The Physics of Built Space

Nothing a room does with light reaches alertness, mood or the body clock except through the spectral irradiance arriving at the cornea along the occupant's gaze, weighted by the five receptor sensitivities of CIE S 026 and expressed for the non-visual channel as melanopic equivalent daylight illuminance (mEDI) on a vertical plane at about 1.2 m in the direction of view. Horizontal desk lux and colour temperature are not that quantity: a uniformly luminous ceiling delivers a vertical plane half of what it delivers a horizontal plane (configuration factor 0.5, a radiometric model result); two sources of equal photopic lux differ by about two-fold in mEDI across the 2700 K to 6500 K range; a brow-shaped occluder fitted to a spectroradiometer at 1.2 m cut the measured melanopic quantity by amounts from negligible up to 41 to 71 percent under steep overhead fixtures in a controlled physical measurement of room layouts (2023). Field consequence: offices meeting the 500 lux horizontal task standard under 3000 to 4000 K fluorescent light averaged below 150 lux mEDI at the eye in the field studies Brown et al. (2022) summarize, below the 250 lux daytime consensus target. Boundary: radiometry and metrology; it states what the dose is, not what the dose does. Population: office workers in the summarized field studies; evidence class MODEL (standard and geometry) plus a controlled physical measurement.

Photoreceptors absorb photons and receive no label saying daylight or electric. The retina integrates radiance over the solid angle the eye faces, so direction, gaze, surface reflectance, glazing and the brow set the dose; a large low-luminance surface in the sightline (a wall, a window) delivers more to the eye than a bright small fixture overhead at equal room lux.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 (the writer claimed FORGED) · Survived the adversary's attack · Sources: 4 cited, 3 opened and confirmed

• PROBATION: 1 misquote flag elsewhere in this vector (below the kill line of 2); this axiom's own sources were clean -> capped at PROBABLE
• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 51% (>=90%), resolved 95% (>=80%) -> BELOW FLOOR
• dead links: 18 of 37 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED PROBABLE: one of {grounded cross-lab adversary, verified sources} but not both
• CLAMPED: claimed FORGED -> granted PROBABLE (gate lowers, never raises)

S1 (the alpha-opic standard and the two-fold spectral range), S2 (vertical 1.2 m specification; control offices below 150 lux mEDI), S23 (occluder measurement, negligible to 41 to 71 percent loss), S37 (direction-of-view measurement, EML to mEDI by 0.9058). All four legs carried S1 and S2; S23 from the Gemini leg; the configuration-factor ratio is the Anthropic leg's radiometric derivation. Observation: the occluder measurements and the control-office summary. Inference: the 0.5 ratio and the two-fold range, arithmetic from the standard.

• S1CIE S 026:2018, the alpha-opic metrology standardopened, quote confirmedfive receptor-weighted quantities and melanopic EDI defined; equal photopic lux differs about two-fold in mEDI by spectrum (MODEL, standard)
• S2Brown et al. 2022, PLOS Biology, recommendations for daytime, evening and nighttime indoor light exposureopened, quote confirmed250 lux mEDI by day, at most 10 lux in the three hours before bed, at most 1 lux during sleep, vertical at 1.2 m; control offices average below 150 lux mEDI (ARGUMENT, expert consensus, 2022)
• S23Field-of-view occlusion in spectroradiometric measurement of melanopic exposure at eye levelopened, quote confirmeda brow-shaped occluder on a spectroradiometer at 1.2 m cut measured visual and melanopic quantities by amounts from negligible up to 41 to 71 percent under steep overhead fixtures (INTERVENTION, controlled physical measurement, 2023)
• S37CIE TN 015:2023, second international workshop on circadian and neurophysiological photoreceptiondid not openEML converts to melanopic EDI by 0.9058; the dose is specified in the vertical plane at eye level in the direction of view (ARGUMENT, standards body, 2023)

The page may say light acts on the occupant only if the sentence names light at the eye; any claim stated in room lux or colour temperature is rewritten in mEDI at the eye or dropped.

This is the light channel's estimand row, and it pairs with the sound channel's intelligibility index (2.1), the sightline channel's image statistics (3.1) and the air channel's outdoor air per person (4.2): in every channel the quantity that acts is measured at the occupant's position, never read off the fixture, the decibel meter, the object count or the thermostat. For the architecture page it converts 'natural light' and 'warm light' into one number at the eye, and it is the row that lets the sleep article's room-lux sentences be rewritten rather than dropped.

In dim-adapted healthy young adults exposed in the laboratory during the biological night, about 100 photopic lux at the eye produced half the maximal phase delay and half the maximal melatonin suppression seen near 9000 lux (23 adults, 6.5 h exposure, between-subject dose arms, 2000), and about 100 lux produced half the maximal alerting response measured as sleepiness ratings, slow eye movements and EEG theta-alpha power after hours below 3 lux (2000), with nearly the whole transition from no response to maximal response between about 30 and 300 lux. The curve is logistic in log illuminance, so the actionable unit is the decade, not the percent. Between people the curve slides: in 55 young adults given 5 h of evening light beginning four hours before habitual bedtime, the group illuminance for 50 percent melatonin suppression was about 25 lux and individual values ran from about 6 to about 350 lux, a fifty-fold spread within one protocol (2019). Boundary: melatonin suppression, phase delay and night-time alerting in dim-adapted laboratory subjects; the half-maximum moves with prior light history and timing, so 100 lux is a protocol result, not a universal threshold. Evidence class INTERVENTION, laboratory.

Melanopsin-driven ganglion cells signal irradiance after adaptation, which compresses input logarithmically; saturation of the pathway and of melatonin suppression caps the top of the curve and dim adaptation lowers the bottom, so evening room light lands on the steep part while a daytime office increment of a few hundred lux lands near the flat top.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 (the writer claimed FORGED) · Survived the adversary's attack · Sources: 3 cited, 1 opened and confirmed

• PROBATION: 1 misquote flag elsewhere in this vector (below the kill line of 2); this axiom's own sources were clean -> capped at PROBABLE
• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 51% (>=90%), resolved 95% (>=80%) -> BELOW FLOOR
• dead links: 18 of 37 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED PROBABLE: one of {grounded cross-lab adversary, verified sources} but not both
• CLAMPED: claimed FORGED -> granted PROBABLE (gate lowers, never raises)

S3 (half-maximum near 100 lux for phase delay and suppression, 23 adults), S4 (half-maximum alerting near 100 lux, logistic), S5 (55 adults, group ED50 about 25 lux, individual 6 to 350 lux). OpenAI and Anthropic legs carried all three; Gemini carried S3 as a 54 percent amplitude reduction at 100 to 150 lux against 52 percent at 10,000 lux. Observation: the fitted half-maxima and the 6 to 350 lux range. Inference: the decade-as-unit corollary.

• S3Zeitzer et al. 2000, J Physiol, sensitivity of the human circadian pacemaker to nocturnal lightdid not openabout 100 photopic lux gave half the maximal phase delay and melatonin suppression near 9000 lux, 23 young adults, 6.5 h, transition between 30 and 300 lux (INTERVENTION, lab, 2000)
• S4Cajochen et al. 2000, Behavioural Brain Research, dose-response relationship for light intensity and alertnessopened, quote confirmedhalf the maximal alerting response (sleepiness, slow eye movements, EEG theta-alpha) at about 100 lux after hours below 3 lux; logistic curve (INTERVENTION, lab, 2000)
• S5Phillips et al. 2019, PNAS, high sensitivity and interindividual variability in the response of the human circadian system to evening lightdid not open55 young adults, 5 h evening exposure, group ED50 for melatonin suppression about 25 lux, individual ED50 about 6 to 350 lux (INTERVENTION, lab, 2019)

A page sentence about light's strength must say when (biological night versus day) and for whom (the sensitive tail); a room specification cannot deliver one physiological dose to a household.

The logistic-in-log-dose shape explains why evening room light does so much and a daytime office increment does so little (1.5): the same photons land on different parts of one curve. The fifty-fold spread between people is the light channel's instance of a law that recurs in the sightline channel as a suppression trait (3.4): at ordinary room levels the variance between occupants exceeds the variance a room can deliver, so no specification yields one dose to a household.

The Physics of Sleep (February 2026) holds that light acts on the body clock; newly opened sources verify and bound that claim by name. In healthy adults under constant-routine laboratory conditions, a 6.7 h stimulus near 10,000 lux delayed the clock when centred before the core body temperature minimum and advanced it after, with a 5.02 h peak-to-trough amplitude and no dead zone in the subjective day (2003), so the same photons delay, advance or do nothing by internal time and a saturating stimulus buys at most about plus or minus 2.5 h. Duration pays off nonlinearly: in 39 young adults a 0.2 h pulse near 10,000 lux delayed the clock 1.07 plus or minus 0.36 h and a 4.0 h pulse 2.65 plus or minus 0.24 h, the first minutes over five times more effective per minute (2012). Ordinary evening room light acts: in 116 healthy adults aged 18 to 30 in a research centre, room light under 200 lux in the 8 h before bedtime, against under 3 lux, delayed melatonin onset in 99.0 percent of individuals and shortened melatonin duration by about 90 min (2011). The phase and acute channels dissociate: millisecond flashes shifted phase with no dose-dependent melatonin suppression. Consensus targets from this body of work: at most 10 lux mEDI at the eye in the three hours before bed and at most 1 lux during sleep (2022). Boundary: healthy young adults in inpatient protocols; the magnitudes are laboratory ceilings, not what a room delivers. Evidence class INTERVENTION, laboratory.

The suprachiasmatic clock's light sensitivity is itself rhythmic (a phase-response curve), so the sign of a shift is set by internal phase, not clock time; the response to a pulse saturates quickly, so a long exposure adds less per minute than a short one; acute suppression and phase resetting run on different integration rules downstream of the same cells.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 (the writer claimed FORGED) · Survived the adversary's attack · Sources: 5 cited, 3 opened and confirmed

• PROBATION: 1 misquote flag elsewhere in this vector (below the kill line of 2); this axiom's own sources were clean -> capped at PROBABLE
• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 51% (>=90%), resolved 95% (>=80%) -> BELOW FLOOR
• dead links: 18 of 37 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED PROBABLE: one of {grounded cross-lab adversary, verified sources} but not both
• CLAMPED: claimed FORGED -> granted PROBABLE (gate lowers, never raises)

S6 (PRC, 5.02 h amplitude, no dead zone), S7 (0.2 h versus 4.0 h pulses in 39 adults), S8 (116 adults, 99.0 percent delayed onset, 90 min shorter), S32 (flashes shift phase without dose-dependent suppression), S2 (10 and 1 lux mEDI targets). Khalsa and Brown in all four legs; Chang and Gooley from the Exa leg; the flash study from the Anthropic leg. Observation: the amplitude, the two pulse shifts, the 99.0 percent of 116. Inference: the plus or minus 2.5 h ceiling and the channel dissociation.

• S6Khalsa et al. 2003, J Physiol, a phase response curve to single bright light pulses in human subjectsdid not open6.7 h stimulus near 10,000 lux; delays before the core body temperature minimum, advances after; 5.02 h peak-to-trough amplitude; no dead zone (INTERVENTION, lab, 2003)
• S7Chang et al. 2012, J Physiol, human responses to bright light of different durationsopened, quote confirmed39 young adults; 0.2 h pulse near 10,000 lux delayed 1.07 plus or minus 0.36 h, 4.0 h pulse 2.65 plus or minus 0.24 h; first minutes over five times more effective per minute (INTERVENTION, lab, 2012)
• S8Gooley et al. 2011, JCEM, exposure to room light before bedtime suppresses melatonin onset and shortens melatonin durationdid not open116 healthy adults 18 to 30; room light under 200 lux in the 8 h before bed versus under 3 lux delayed melatonin onset in 99.0 percent and shortened duration about 90 min (INTERVENTION, research center, 2011)
• S32Millisecond light flashes and circadian phase shifting in humansopened, quote confirmed2 ms flashes at 3 to 9500 lux and 10 microsecond to 10 s flashes at 2000 lux shifted phase, with only slight dependence on duration and no dose-dependent melatonin suppression (INTERVENTION, lab, 2022)
• S2Brown et al. 2022, PLOS Biology, recommendations for daytime, evening and nighttime indoor light exposureopened, quote confirmed250 lux mEDI by day, at most 10 lux in the three hours before bed, at most 1 lux during sleep, vertical at 1.2 m; control offices average below 150 lux mEDI (ARGUMENT, expert consensus, 2022)

The architecture page may import the sleep article's clock claim only as 'evening room light under 200 lux delays the clock in nearly everyone; light after the temperature minimum advances it,' with the 10 lux evening target at the eye; a sentence implying brightness alone sets the shift is rewritten to name timing.

This is the row that carries the adjacent-article verdict: the sleep article's clock claim is verified against newly opened phase-response and evening-light sources and bounded by sign, ceiling and duration, never restated. Across the run it is also the clearest case of a law that recurs in sound (2.3): the sign and size of a room's effect are set by a contextual variable the room does not control alone, here the occupant's internal time.

In a randomized crossover with 29 adults, one evening hour of about 60 photopic lux in two visually matched (metameric) spectra, 123 versus 59 lux mEDI, produced about 14 percent more relative melatonin suppression under the higher melanopic dose and no detectable difference in sleepiness, vigilance, subsequent sleep or sleep quality (2022). In 72 men given four pre-bedtime hours of metameric light, the lower melanopic condition lowered alertness and melatonin suppression and shifted melatonin onset by about half an hour measured during the exposure, a suppression effect and not a persisting clock shift (2023). By day, in 39 young adults from 09:00 to 11:00, blue light (about 1442 lux mEDI) and red light (about 3.8 lux mEDI) both improved some psychomotor-vigilance measures against dim light while blue-enriched white light (about 1156 lux mEDI) did not, so higher mEDI did not predict better daytime vigilance (2025). Millisecond flashes shifted the clock without dose-dependent melatonin suppression. Boundary: healthy young adults in laboratory crossovers; the law is that no single endpoint stands in for another, not that any endpoint is unresponsive. Evidence class INTERVENTION.

Melanopsin cells project to several targets (the clock, arousal nuclei, and in rodents a thalamic mood pathway); each target integrates light on its own time constant with its own cone and rod inputs, so a dose that moves one hormone can leave vigilance unchanged and vice versa.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 (the writer claimed FORGED) · Survived the adversary's attack · Sources: 4 cited, 2 opened and confirmed

• PROBATION: 1 misquote flag elsewhere in this vector (below the kill line of 2); this axiom's own sources were clean -> capped at PROBABLE
• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 51% (>=90%), resolved 95% (>=80%) -> BELOW FLOOR
• dead links: 18 of 37 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED PROBABLE: one of {grounded cross-lab adversary, verified sources} but not both
• CLAMPED: claimed FORGED -> granted PROBABLE (gate lowers, never raises)

S9 (29 adults, 14 percent suppression difference, null alertness), S28 (72 men, metameric evening), S27 (39 adults, three daytime conditions), S32 (flash dissociation). OpenAI leg for S9, S27, S28; Anthropic leg for S32; Gemini leg supplied the divergent-pathway framing. Observation: the suppression difference with null alertness; the three daytime conditions. Inference: the separate-targets mechanism.

• S9Blume et al. 2022, metameric evening light and melatonin, sleepiness, vigilance and sleepopened, quote confirmed29 adults, one hour at about 60 photopic lux, 123 versus 59 lux mEDI: about 14 percent more relative melatonin suppression, no difference in sleepiness, vigilance or sleep (INTERVENTION, randomized crossover, 2022)
• S272025 Scientific Reports daytime light and psychomotor vigilance trialdid not open39 young adults 09:00 to 11:00: blue (about 1442 lux mEDI) and red (about 3.8 lux mEDI) improved some PVT measures against dim light, blue-enriched white (about 1156 lux mEDI) did not (INTERVENTION, 2025)
• S28Schöllhorn et al. 2023, Communications Biology, melanopic irradiance and evening alertness, metameric lightdid not open72 men, four pre-bedtime hours of metameric light: lower melanopic exposure lowered alertness and melatonin suppression; melatonin onset differed about 30 min measured during exposure (INTERVENTION, 2023)
• S32Millisecond light flashes and circadian phase shifting in humansopened, quote confirmed2 ms flashes at 3 to 9500 lux and 10 microsecond to 10 s flashes at 2000 lux shifted phase, with only slight dependence on duration and no dose-dependent melatonin suppression (INTERVENTION, lab, 2022)

A page sentence may not move from 'light suppresses melatonin' to 'light makes you alert' or 'lifts your mood'; each endpoint carries its own row.

The light channel's dissociation row. Read beside the sound channel's triple dissociation (2.4), the clutter channel's arousal-without-mood finding (3.8) and the heat channel's comfort-versus-performance split (4.7), it establishes that 'state of mind' is not one variable a room sets: melatonin, vigilance and mood each carry their own row. It is the row that forbids the page from sliding from 'light suppresses melatonin' to 'light lifts your mood'.

Pooled across 29 crossover and parallel studies with 1,210 healthy participants, light interventions improved subjective alertness by a standardized mean difference of -0.28 (95 percent CI -0.49 to -0.06, sleepiness scales) and objective alertness by -0.34 (95 percent CI -0.68 to -0.01, reaction-time tasks), with wide heterogeneity (2022). Restricted to daytime electric light, a 2025 review found faster responses at higher intensity with summary r = 0.06 (95 percent CI 0.01 to 0.109, I-squared about 45 percent) and no significant summary effect of spectral composition. Where the melanopic contrast is large the effect shows within a day: 16 sleep-restricted young adults in an inpatient 8 h simulated workday improved on an addition task by about 3.15 percent against 0.93 percent from baseline when a task lamp raised the eye-level dose from about 30 to about 250 lux mEDI (50 to 210 photopic lux), with better vigilance and subjective alertness (2023). Under 1000 versus 200 lux by day, self-reported alertness changed at the first rating and performance changed gradually (2012). Where the contrast is small nothing shows: a randomized simulated-classroom arm with 55 students scored 159.57 versus 157.69 on concentration, about 0.06 SD, not significant, while the uncontrolled field arms reported r about 0.30 and 0.64; a two-week 470 to 490 nm classroom intervention cut attentional errors with partial eta-squared 0.09 and changed neither mood nor sleep (2025). Office daylight packages are total effects, not light doses: 30 workers in an unblinded crossover slept 37 min longer by actigraphy and scored 42 percent higher on one decision-making simulation under optimized daylight and views against blinds, with the view changed too (2020), and a case-control pilot of 49 workers (actigraphy in 21) found window workers slept longer, confounded by seat allocation (2014). Boundary: healthy or sleep-restricted adults and adolescents by day; pooled numbers average heterogeneous doses; each effect size names the sleepiness or reaction-time scale it was measured on. Evidence class INTERVENTION (pooled and single trials) with the two office daylight rows as a package effect and a correlational pilot.

By day most occupants sit on the upper, flatter part of the log-dose curve and the acute alerting signal is one input beside circadian phase and sleep pressure, so a few hundred lux more at the eye yields a small shift in reaction time, large only when the baseline room is melanopically dark or the person is sleep-deprived.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 (the writer claimed FORGED) · Survived the adversary's attack · Sources: 8 cited, 4 opened and confirmed

• PROBATION: 1 misquote flag elsewhere in this vector (below the kill line of 2); this axiom's own sources were clean -> capped at PROBABLE
• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 51% (>=90%), resolved 95% (>=80%) -> BELOW FLOOR
• dead links: 18 of 37 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED PROBABLE: one of {grounded cross-lab adversary, verified sources} but not both
• CLAMPED: claimed FORGED -> granted PROBABLE (gate lowers, never raises)

S13 (SMD -0.28 and -0.34, 1,210 participants), S14 (r = 0.06, daytime electric light), S15 (16 sleep-restricted adults, 3.15 versus 0.93 percent), S26 (1000 versus 200 lux, immediate subjective change), S20 (randomized classroom arm null, n = 55), S30 (two-week classroom, partial eta-squared 0.09), S22 (30 workers, 37 min and 42 percent, package effect), S21 (49 workers, case-control). Exa leg for S13, S14, S15; OpenAI for S20, S22; Anthropic for S21; Anthropic for S26 and S14 (marked UNVERIFIED there, opened by Exa); Gemini for S30. Observation: the pooled SMDs, the percentages, the two classroom scores. Inference: the flat-curve explanation.

• S13Mu et al. 2022, alerting effects of light in healthy individuals, systematic review and meta-analysisopened, quote confirmed29 studies, 1,210 healthy participants: subjective alertness SMD -0.28 (95 percent CI -0.49 to -0.06), objective alertness SMD -0.34 (-0.68 to -0.01) (INTERVENTION pooled, 2022)
• S14Lasauskaite et al. 2025, LEUKOS, non-image-forming effects of daytime electric light exposure, systematic review and meta-analysesdid not opendaytime electric light: faster response at higher intensity r = 0.06 (95 percent CI 0.01 to 0.109), I-squared about 45 percent; no significant summary effect of spectral composition (INTERVENTION pooled, 2025)
• S15Grant et al. 2023, SLEEP, supplementation of ambient lighting with a task lamp improves daytime alertness and cognitive performance in sleep-restricted individualsdid not open16 sleep-restricted young adults, 8 h simulated workday, task lamp raising about 30 to about 250 lux mEDI (50 to 210 photopic lux): addition task +3.15 versus +0.93 percent, better vigilance and alertness (INTERVENTION, crossover, inpatient, 2023)
• S26Smolders et al. 2012, Physiology and Behavior, a higher illuminance induces alertness even during office hoursdid not open1000 versus 200 lux by day in non-sleep-deprived adults: subjective alertness changed at the first rating, performance and heart-rate variability changed gradually (INTERVENTION, lab-office, 2012)
• S20Sleegers et al., lighting affects students' concentration positively, findings from three Dutch studiesopened, quote confirmedrandomized simulated-classroom arm n = 55: concentration 159.57 versus 157.69, about 0.06 SD, not significant; uncontrolled field arms r about 0.30 and 0.64 (INTERVENTION and field, 2012)
• S30Two-week 470 to 490 nm enriched classroom lighting intervention in adolescentsopened, quote confirmedattentional errors on the Frankfurter Attention Inventory fell, partial eta-squared 0.09; no significant change in self-reported mood or sleep (INTERVENTION, school, 2025)
• S22Boubekri et al. 2020, the impact of optimized daylight and views on the sleep duration and cognitive performance of office workersopened, quote confirmed30 workers, crossover, optimized daylight and views versus blinds: 37 min longer actigraphic sleep, 42 percent higher decision-making simulation score; views changed too (INTERVENTION, crossover, unblinded, 2020)
• S21Boubekri et al. 2014, JCSM, impact of windows and daylight exposure on overall health and sleep quality of office workers, a case-control pilot studydid not opencase-control pilot, 49 workers, actigraphy in 21; window workers had more light exposure and longer sleep; confounded by seat allocation (CORRELATIONAL, 2014)

The page may say daytime light at the eye sharpens alertness by a small, measurable amount; it may not say light makes people productive, and a school or office claim needs the melanopic contrast stated.

The daytime magnitude row, and the light channel's entry in the design-rigour gradient (4.5): pooled interventions with named denominators give a few tenths of a standard deviation on sleepiness and reaction-time scales, single field studies and preprints give more, and a randomized classroom arm with a small melanopic contrast gives nothing. For the page it fixes what 'light sharpens you' may mean by day and forbids 'light makes you productive'.

In nonseasonal major depression, 10,000 lux white light for 30 min each morning over eight weeks improved the MADRS score by 13.4 points against 6.5 with sham, groups of 32 and 30, a standardized effect near 0.8 (2016); pooled over 11 trials and 858 patients, bright light raised remission to 40.7 against 23.5 percent (odds ratio 2.42) and response to 60.4 against 38.6 percent (2024); an earlier meta-analysis gave d = 0.84 (95 percent CI 0.60 to 1.08) in seasonal depression and 0.53 (0.18 to 0.89) as nonseasonal monotherapy, with -0.01 as an adjunct to drugs (2005). The one room-scale blinded trial is in the elderly: 12 care facilities with 189 residents (mean age 85.8, 87 percent with dementia) randomized to about 1000 versus 300 lux at the eye from 10:00 to 18:00 over a mean 15 months improved depressive symptoms by 1.5 points on the Cornell scale (95 percent CI 0.24 to 2.70) and slowed MMSE decline by 0.9 points (0.04 to 1.71), assessed every six months (2008). In healthy adults in ordinary rooms the evidence is small and confounded: 15 students after one hour of 6500 K against 3500 K at 500 lux rated mood d = 0.58 better, unblinded, colour appearance changed (2019); 25 subjects spending afternoons under daylight against electric light preferred daylight for glare and acceptance, and mood tracked the photometric variables through that preference (2015); an ICU trial of 734 patients with dynamic lighting to 1700 lux found delirium in 38 against 33 percent (odds ratio 1.24, 95 percent CI 0.92 to 1.68), no benefit (2016). Boundary: the measured mood effects belong to depressed patients and institutionalized elderly under repeated high doses over weeks to months; the rodent perihabenular pathway is a mechanism with no human magnitude. Evidence class INTERVENTION (RCTs and pooled trials), with the healthy-adult rows confounded.

Mood responses to light accumulate through repeated daily exposure acting on clock timing, sleep and (in animals) a direct thalamic pathway, so they appear on the scale of weeks in people whose baseline is low and are indistinguishable from preference and glare in a healthy person's afternoon.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 (the writer claimed FORGED) · Survived the adversary's attack · Sources: 7 cited, 4 opened and confirmed

• PROBATION: 1 misquote flag elsewhere in this vector (below the kill line of 2); this axiom's own sources were clean -> capped at PROBABLE
• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 51% (>=90%), resolved 95% (>=80%) -> BELOW FLOOR
• dead links: 18 of 37 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED PROBABLE: one of {grounded cross-lab adversary, verified sources} but not both
• CLAMPED: claimed FORGED -> granted PROBABLE (gate lowers, never raises)

S17 (MADRS 13.4 versus 6.5), S33 (858 patients, remission 40.7 versus 23.5 percent), S18 (d = 0.84 seasonal; 0.53 nonseasonal; -0.01 adjunct), S16 (189 residents, Cornell 1.5, MMSE 0.9), S36 (15 students, d = 0.58), S35 (25 subjects, preference confound), S19 (734 ICU patients, OR 1.24). OpenAI for S16, S17, S19; Anthropic for S16, S18, S33; Exa for S35, S36; Gemini for the pathway framing. Observation: every score and interval. Inference: the accumulation mechanism and the preference confound.

• S16Riemersma-van der Lek et al. 2008, JAMA, effect of bright light and melatonin on cognitive and noncognitive function in elderly residents of group care facilitiesopened, quote confirmed12 facilities, 189 residents (mean age 85.8, 87 percent with dementia), about 1000 versus 300 lux at the eye 10:00 to 18:00, mean 15 months: Cornell depression 1.5 points (95 percent CI 0.24 to 2.70), MMSE decline attenuated 0.9 points (0.04 to 1.71) (INTERVENTION, cluster RCT, 2008)
• S17Lam et al. 2016, JAMA Psychiatry, efficacy of bright light treatment, fluoxetine and the combination in nonseasonal major depressionopened, quote confirmed10,000 lux 30 min each morning for 8 weeks: MADRS improvement 13.4 versus 6.5 points with sham, groups of 32 and 30, effect about 0.8 (INTERVENTION, RCT, 2016)
• S18Golden et al. 2005, Am J Psychiatry, the efficacy of light therapy in the treatment of mood disorders, meta-analysisdid not openseasonal depression d = 0.84 (95 percent CI 0.60 to 1.08); nonseasonal monotherapy 0.53 (0.18 to 0.89); adjunct -0.01 (-0.36 to 0.34) (INTERVENTION pooled, 2005)
• S19Simons et al. 2016, Lancet Respiratory Medicine, dynamic light application therapy to reduce delirium in intensive caredid not open734 ICU patients, dynamic lighting to 1700 lux versus 300 lux: delirium 38 versus 33 percent, odds ratio 1.24 (95 percent CI 0.92 to 1.68), no benefit (INTERVENTION, RCT, 2016)
• S332024 systematic review and meta-analysis of bright light therapy for nonseasonal depressive disordersopened, quote confirmed11 trials, 858 patients: remission 40.7 versus 23.5 percent (OR 2.42, 1.50 to 3.91), response 60.4 versus 38.6 percent (OR 2.34) (INTERVENTION pooled, 2024)
• S35Borisuit et al. 2015, Lighting Research and Technology, effects of realistic office daylighting and electric lighting on visual comfort, alertness and mooddid not open25 young subjects, two afternoons under daylight versus electric light without view: daylight preferred for acceptance and glare; mood and alertness changes tracked photometric variables and comfort (INTERVENTION, within-person, 2015)
• S36Choi et al. 2019, Scientific Reports, awakening effects of blue-enriched morning light on university studentsopened, quote confirmed15 students, one hour of 6500 K versus 3500 K at 500 lux: sleepiness d = 0.70, mood d = 0.58, melatonin fell more under 6500 K; unblinded, colour appearance changed (INTERVENTION, within-person, 2019)

The page may not say light lifts mood in a room; it may say repeated bright morning light treats depression and that all-day 1000 lux slowed decline in dementia care, each with its dose and population.

The row that moves mood out of the room and into the clinic: measured mood effects of light are repeated high doses over weeks in depressed or institutionalized populations, and a healthy adult's mood under ordinary room light is confounded with preference and glare. Together with 3.8 (clutter raises arousal without changing reported mood) it is why the page's 'state of mind' cannot mean mood on the evidence this run opened.

The onset ladder, each rung a measurement: in three blind individuals retaining non-image-forming responses, two seconds of intense blue light altered EEG responses in an auditory task without conscious vision (2013); in 28 men exposed in the evening, the half-maximum response time was about 48 s for tonic pupil constriction and about 4, 8 and 12 min for EEG delta, beta and gamma changes (2019); blue ambient light modulated amygdala and emotional-voice responses in fMRI during the exposure (2010); the CIE position statement records pupil and EEG responses in milliseconds to seconds (2024); self-reported alertness under 1000 versus 200 lux changed at the first rating while performance changed gradually (2012); melatonin suppression accrues over an hour (2022); the clock shift appears on the next cycle (2003); mood changes are measured over weeks. Awareness: in two subjects without rod or cone function, 480 nm light suppressed melatonin, reset the clock, constricted the pupil and raised alertness while 555 nm did nothing, and one subject could report the presence of intense blue light, so melanopsin activation faintly reaches awareness at high short-wavelength intensity (2007). The occupant cannot report the melanopic dose (two rooms matched for perceived brightness differ about two-fold in mEDI), but the channel is not fully unconscious. Boundary: the fast rungs are pupil, EEG and fMRI in small laboratory samples (n = 2, 3, 28); no measurement places a change in thought or mood before the first seconds, so 'before a thought forms' is wider than the evidence. Evidence class INTERVENTION, laboratory, small n.

Intrinsic melanopsin phototransduction is slow and sustained compared with cones, so its grip on the pupil and on subcortical arousal circuits builds over tens of seconds and outlasts the stimulus, while hormonal, clock and mood endpoints integrate over hours to weeks; the signal runs beside image vision and only its brightest short-wavelength extreme is noticed.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 · Survived the adversary's attack · Sources: 8 cited, 4 opened and confirmed

• PROBATION: 1 misquote flag elsewhere in this vector (below the kill line of 2); this axiom's own sources were clean -> capped at PROBABLE
• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 51% (>=90%), resolved 95% (>=80%) -> BELOW FLOOR
• dead links: 18 of 37 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED PROBABLE: one of {grounded cross-lab adversary, verified sources} but not both

S11 (2 s blue light, EEG, n = 3), S12 (48 s pupil; 4, 8, 12 min EEG; 28 men), S29 (amygdala fMRI during exposure), S34 (millisecond to second responses, CIE), S26 (immediate subjective alertness), S9 (hour-scale suppression), S6 (next-cycle phase), S10 (two blind subjects; awareness of intense blue). OpenAI for S11, S12; Anthropic for S10, S26; Gemini for S29 and the pupil timeline; Exa for S34. Observation: each latency and the blind subjects' reports. Inference: the sustained-signal mechanism and that mood is never among the fast rungs.

• S10Zaidi et al. 2007, Current Biology, short-wavelength light sensitivity of circadian, pupillary and visual awareness in humans lacking an outer retinadid not opentwo blind subjects: 480 nm suppressed melatonin, reset the clock and raised alertness, 555 nm did not; one subject could report intense blue light (INTERVENTION, n = 2, 2007)
• S11Vandewalle et al. 2013, blue light stimulates cognitive brain activity in visually blind individualsopened, quote confirmedthree blind individuals: two seconds of blue light altered auditory-task EEG responses without conscious vision (INTERVENTION, n = 3, 2013)
• S12Rahman et al. 2019, Frontiers in Neuroscience, dynamics of non-visual responses to lightopened, quote confirmed28 men, evening: half-maximum response times about 48 s for tonic pupil constriction and about 4, 8 and 12 min for EEG delta, beta and gamma (INTERVENTION, lab, 2019)
• S26Smolders et al. 2012, Physiology and Behavior, a higher illuminance induces alertness even during office hoursdid not open1000 versus 200 lux by day in non-sleep-deprived adults: subjective alertness changed at the first rating, performance and heart-rate variability changed gradually (INTERVENTION, lab-office, 2012)
• S29Vandewalle et al. 2010, PNAS, spectral quality of light modulates emotional brain responses in humansdid not openblue ambient light modulated amygdala and voice-area responses to emotional stimuli in fMRI during the exposure (INTERVENTION, fMRI, 2010)
• S34CIE Position Statement PS 001:2024, recommending proper light at the proper time, third editionopened, quote confirmedlight elicits fast responses in milliseconds and seconds in the pupil reflex and brain activity; integrative lighting defined (ARGUMENT, standards body, 2024)
• S9Blume et al. 2022, metameric evening light and melatonin, sleepiness, vigilance and sleepopened, quote confirmed29 adults, one hour at about 60 photopic lux, 123 versus 59 lux mEDI: about 14 percent more relative melatonin suppression, no difference in sleepiness, vigilance or sleep (INTERVENTION, randomized crossover, 2022)
• S6Khalsa et al. 2003, J Physiol, a phase response curve to single bright light pulses in human subjectsdid not open6.7 h stimulus near 10,000 lux; delays before the core body temperature minimum, advances after; 5.02 h peak-to-trough amplitude; no dead zone (INTERVENTION, lab, 2003)

The light component of 'before you have formed a thought' is bounded: the page may say light acts on the pupil and brain rhythms within seconds and outside awareness of the dose; it may not say light changes mood or thinking before thought.

The light rung of the run's onset ladder, read with 2.5 (sound), 2.6 (one's own sound) and 3.2 (the sightline): what is measured in the first seconds to minutes is the pupil, brain rhythms and an imaging response, never a change in thought or mood, and the melanopic dose is unreportable yet not sealed from awareness. The page's 'before you have formed a thought' is upheld here for reflexes and bounded against mood and thinking.

Daylight glare probability is a function containing vertical illuminance at the eye alongside source luminance, solid angle and position, with bands at 0.35 perceptible, 0.40 disturbing and 0.45 intolerable, and the unified glare rating has steps at 19 just acceptable, 22, 25 and 28 uncomfortable (model formulas fitted to office observers' comfort ratings). Because mEDI also rides vertical eye illuminance, raising the non-visual dose by brute illuminance raises glare along a monotone function; the degrees of freedom that raise dose without raising glare are melanopic efficacy (spectrum enters mEDI and not the glare terms), source solid angle and position (both enter glare squared or inverse-squared and corneal irradiance only linearly). In a simulation of electrochromic glazing, 250 lux mEDI at the eye was met without exceeding 1500 lux vertical in only 24.7 to 33.3 percent of hours depending on orientation (2020). The outdoor-indoor gap is large: across 675 measured scenes, outdoor mEDI averaged above 12,000 lux, windowed rooms about 2,300 and windowless rooms under 300 (2025 preprint, field measurement). Boundary: glare bands are fitted to comfort ratings; the glazing figure is one simulation; the scene survey is a preprint. Evidence class MODEL plus one correlational field measurement.

Discomfort glare grows with source luminance squared times solid angle against adaptation luminance, while the non-visual dose grows linearly with irradiance summed over the field, so spreading the same flux over a large dim surface in the sightline raises dose and lowers glare at once.

Granted tier: SIGNAL, set by the code gate on October 4, 2026 (the writer claimed PROBABLE) · Refuted by the adversary on a source said less than the claim · Sources: 3 cited, 0 opened and confirmed

• TRIPWIRE: this axiom's own source(s) S24 (number_not_on_page) were wrong -> SIGNAL (the tripwire hits the axiom whose use of a source was wrong)
• CLAMPED: claimed PROBABLE -> granted SIGNAL (gate lowers, never raises)

S24 (UGR and DGP bands; DGP contains vertical eye illuminance), S25 (24.7 to 33.3 percent of hours), S31 (675 scenes: above 12,000, about 2,300, under 300 lux mEDI). Anthropic leg for S24, S25; Gemini for S31 and the dose-glare correlation. Observation: the band values, the hours fraction, the scene means. Inference: the separability argument read off the formulas.

• S24DesignLights Consortium, glare metrics technical noteopened, the number was not on the pageUGR steps 19 just acceptable, 22, 25, 28 uncomfortable; DGP bands 0.35 perceptible, 0.40 disturbing, 0.45 intolerable; DGP contains vertical illuminance at the eye (MODEL, 2021)
• S25Electrochromic glazing, melanopic EDI and vertical illuminance study, Buildings 2020did not openat least 250 lux mEDI without exceeding 1500 lux vertical illuminance in only 24.7 to 33.3 percent of hours depending on orientation (MODEL, simulation, 2020)
• S312025 bioRxiv analysis of melanopic exposure across 675 natural and indoor scenesdid not openoutdoor scenes mean mEDI above 12,000 lux, indoor spaces with windows about 2,300, windowless rooms under 300 (CORRELATIONAL field measurement, preprint, 2025)

If the page keeps a lighting claim, the design corollary it may state is 'light the walls and the sky view, not the eye'; a window claim must say when and how often it meets the day target.

The light channel's one design corollary: because non-visual dose and glare share vertical eye illuminance, the degrees of freedom are spectrum, source size and position, so large dim surfaces in the sightline raise dose cheaply. Its source record carried a number the gate could not find on the cited page, so the glazing fraction and the scene survey ride as a direction, not a figure, until a second opening; the formula-level separability argument does not depend on them.

In adult laboratory subjects doing verbal short-term memory tasks in simulated offices, the performance cost of irrelevant speech is a function of the Speech Transmission Index (STI, the share of the speech modulation envelope that survives the room, 0 to 1), not of A-weighted level. The revised model built on fourteen laboratory studies and 34 tests (S2, 2020) places the onset of loss near STI 0.21 and its maximum near STI 0.44; the original three-study model (S1, 2005) placed the plateau at about 7 percent mean decrement for STI 0.7 to 1.0 with a 4 to 41 percent spread across tasks; both are model fits to intervention data, so the 7 percent is an asymptote of a fitted curve, not a measured population mean. With overall level held at 55 dBA, meaningful speech raised modelled serial-recall error from 26.3 percent (nonspeech office sound) to 37.2 percent, a 10.9 point rise with a 95 percent credible interval of 7.0 to 14.7 points (S3, 2023, adults, within-person). Near 51 dBA, shifting only the speech-to-masker balance cost about 5 percent on word memory and under 3 percent on arithmetic, most of the word-memory loss already present at STI 0.34 (S4). The FinGer study (S18, 2023, N = 98 across Finland and Germany) found STI 0.3 and 0.1 both outperform STI 0.8, with 0.1 adding only subjective benefit. Within a reasonable range (about 40 to 60 dBA) level itself does not move performance (S1, S2). Boundary: the curve is for irrelevant speech during verbal memory work by adults in laboratories; for a wanted voice (a teacher) higher STI is the goal, so the same metric has opposite sign by room purpose.

The room is a linear transfer system from each source to each ear (S39); reverberation and noise each reduce envelope modulation depth, and the two enter the modulation transfer function multiplicatively, which is what STI integrates across octave and modulation bands. Cognition reads the 2 to 16 Hz syllabic envelope, so a room that preserves modulation delivers segmentable speech whatever the level, and a room that destroys it delivers unsegmentable sound whatever the level. Once the envelope is intelligible enough to be parsed into lexical units the obligatory processing is fully engaged, which is why the curve saturates near STI 0.5 and why the standards put distraction distance rD at STI 0.50 (S38). Lowering background level without lowering transmission raises STI, so a quieter room can be a worse room; the lever is the ratio, reached by absorption, screens and geometry (S40: D2,S 1.3 to 8.5 dB per doubling across treatments) or by a steady spectrum-shaped masking floor.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 (the writer claimed FORGED) · Refuted by the adversary on a number larger than the sources support · Sources: 8 cited, 8 opened and confirmed

• adversary REFUTED the claim
• citations: refetched 95% (>=90%), resolved 100% (>=80%) -> OK
• dead links: 2 of 40 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED PROBABLE: one of {grounded cross-lab adversary, verified sources} but not both
• CLAMPED: claimed FORGED -> granted PROBABLE (gate lowers, never raises)

S1 (2005 model, three studies, 7 percent plateau, 4 to 41 percent spread), S2 (2020 revised model, 14 studies and 34 tests, onset 0.21, maximum 0.44, plateau not rejected), S3 (2023, 55 dBA, 26.3 vs 37.2 percent errors, CrI 7.0 to 14.7), S4 (about 5 percent word memory, under 3 percent arithmetic near 51 dBA, STI 0.34), S18 (2023, N = 98, STI 0.8 vs 0.3 vs 0.1), S38 (rD at STI 0.50), S39 (transfer-system physics), S40 (2024 full-scale measurement range). All four legs carry the STI dose-response; openai-dr and anthropic carry the plateau and the spread; exa carries the 34-test review; gemini-dr carries the same thresholds with an unsourced 'up to 16 percent' figure that is not used.

• S1Hongisto 2005, Indoor Air: a model predicting the effect of speech of varying intelligibility on work performanceopened, quote confirmedThe original STI-to-performance model: decrement rising from STI about 0.2, plateau about 7 percent from STI 0.7; built on three studies; the 4 to 41 percent spread across tasks (MODEL on INTERVENTION data; 2005)
• S2Haapakangas, Hongisto and Liebl 2020, Indoor Air: the revised STI-performance model from laboratory studiesopened, quote confirmedFourteen laboratory studies, 34 tests: performance loss begins near STI 0.21 and reaches its maximum near STI 0.44; verbal short-term memory the most reliably impaired task; the plateau not rejected (MODEL on INTERVENTION data; 2020)
• S3Yadav, Georgi, Leist, Klatte, Schlittmeier and Fels 2023, Applied Acoustics: cognitive performance in open-plan office acoustic simulationsopened, quote confirmedAt 55 dBA, serial-recall error 26.3 percent with nonspeech office sound vs 37.2 percent with meaningful speech (10.9 points, 95 percent credible interval 7.0 to 14.7); meaningful speech worse than foreign speech; raising the irrelevant background's T30 from 0.4 to 0.8 s cut errors 34.3 to 30.2 percent (CrI 0.7 to 7.5), 0.8 vs 1.1 s not robust; spatial separation not robust (INTERVENTION, adults, lab; 2023)
• S4Jahncke, Lulea University of Technology thesis compilation: cognitive performance in open-plan office noise (the STI-balance experiment with Hongisto and Virjonen, and the 39 vs 51 dBA two-hour experiment)opened, quote confirmedWord memory about 5 percent and arithmetic under 3 percent worse as the speech-to-masker balance rose near 51 dBA, most of the word-memory loss already at STI 0.34; 39 vs 51 dBA over two hours changed recall, fatigue and motivation but not cortisol or catecholamines (INTERVENTION, adults, lab; 2011 to 2013)
• S18Hongisto, Keranen, Schiller, Schlittmeier, Tervahartiala and Radun 2023, Forum Acusticum: the FinGer study, room acoustic design and work performanceopened, quote confirmedIdentical experiments in Finland and Germany, N = 98 merged: STI 0.8 (regulation violated) vs 0.3 (fulfilled) vs 0.1 (surpassed); 0.3 and 0.1 gave the best performance, 0.1 further improved subjective experience (INTERVENTION, adults, lab; 2023)
• S38Odeon application note: ISO 3382-3 open-plan office parametersopened, quote confirmedISO 3382-3 measures the spatial decay of speech: D2,S, speech level at 4 m, background level, STI, comfort distance rC (under 45 dBA) and distraction distance rD (STI under 0.50); under 5 m good, over 10 m poor; a simulated ordinary office at D2,S 4.33 dB, rD 15 m (standard and worked example)
• S39arXiv 2111.01900: room acoustics from the wave equation, the diffuse-field energy balance and the Sabine approximationopened, quote confirmedThe room as a linear transfer system from source to each ear; uncorrelated levels add in mean-square (two equal sources +3 dB); T60 about 0.161 V/A as a conditional diffuse-field approximation; averaged level discards event structure (MODEL)
• S40How ISO 3382-3 parameter values are affected by furniture, barriers and sound absorption in a typical open-plan office (2024)opened, quote confirmedFull-scale office: A-weighted speech level at 4 m varied 46 to 60 dB and spatial decay D2,S 1.3 to 8.5 dB per doubling across absorption and screen conditions; bootstrapped 95 percent CIs about 1.2 m for rD and 0.8 dB for D2,S (controlled physical measurement; 2024)

The page may say that what a room does to your attention through sound is set by how much of other people's speech it lets you understand, not by how loud it is, and that the effect stops growing once speech is fully intelligible. It may not say that silence is best or that open offices kill productivity; the measured cost is a bounded 5 to 11 point rise in verbal-memory error in adults, saturating, task-specific, and reversible by room geometry that cuts intelligibility rather than level.

The sound channel's estimand row and the one that refutes 'open offices kill productivity' and 'silence is best' at once: the cost of other people's speech follows how much of it survives the room, not how loud it is, and it saturates once speech is fully intelligible. The gate's adversary refuted part of the claim as first written, so the thresholds and the plateau ride as a model fit over interventions rather than a measured population mean; the direction, the saturation and the sign reversal by room purpose stand, and the row sits in the magnitude group with 1.5, 4.4 and 4.5.

In adult laboratory subjects, an irrelevant auditory stream impairs serial recall of visually presented items when the stream changes state (pitch, segmentation, token identity), and steady-state sound does not disrupt relative to silence (S5, 1993; S6). Meaning is not necessary: non-speech changing streams disrupt (S5). The largest increment of disruption comes from going from one token to two, with further growth as the set grows (S6). Meaning adds a second layer: with level fixed at 55 dBA, meaningful speech produced substantially more disruption than foreign speech, so semantics compounds the acoustic-change effect rather than replacing it (S3, 2023), with the caveat that same dBA did not mean identical spectra and temporal structure, so the whole difference cannot be assigned to meaning. The vulnerable tasks are those maintaining serial order by rehearsal: serial recall, word memory, writing, proofreading, mental arithmetic with carry; visuo-motor and simple vigilance tasks show small or null effects, and the 34-test review found the most reliable loss in verbal short-term memory with weaker evidence for working memory and complex verbal tasks (S2). In full-scale offices the effect neither grew nor faded across a three-hour work period (S17, 2014, N = 98), so it is a steady tax while the sound is present, not a fatigue curve. Boundary: measured in adults, mostly students and office volunteers, in laboratories; children show the same effect with less executive suppression (S7).

Two processes that each maintain the order of a sequence compete: covert articulatory rehearsal of the to-be-remembered items and the automatic seriation of a changing auditory stream. A steady sound has no sequence to seriate and so nothing to collide with. The mechanism predicts the task asymmetry (order-dependent tasks exposed, others not), the token-set-size effect (more distinct events, more seriation), and the semantic add-on (speech is a changing stream that also engages lexical processing). It is specific interference, not a fixed subtraction from a general attention budget, which is why a dB meter or a noise-annoyance survey cannot predict it.

Granted tier: FORGED, set by the code gate on October 4, 2026 · Survived the adversary's attack · Sources: 6 cited, 5 opened and confirmed

• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 95% (>=90%), resolved 100% (>=80%) -> OK
• dead links: 2 of 40 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED FORGED: survived a grounded cross-lab adversary AND primary sources confirmed

S5 (1993, changing-state disruption of visual serial recall, meaning unnecessary), S6 (token set size, steady state does not disrupt), S3 (2023, meaningful vs foreign speech at 55 dBA), S2 (2020, verbal short-term memory most reliably impaired across 34 tests), S17 (2014, N = 98, no amplification or attenuation across three hours), S7 (2010, children more impaired). openai-dr, anthropic and exa carry the changing-state and serial-order account; gemini-dr frames the same evidence as a 'duplex mechanism' of phonological interference plus attentional capture, which is a competing theoretical gloss on the same data, not a contradiction of the measurements.

• S2Haapakangas, Hongisto and Liebl 2020, Indoor Air: the revised STI-performance model from laboratory studiesopened, quote confirmedFourteen laboratory studies, 34 tests: performance loss begins near STI 0.21 and reaches its maximum near STI 0.44; verbal short-term memory the most reliably impaired task; the plateau not rejected (MODEL on INTERVENTION data; 2020)
• S3Yadav, Georgi, Leist, Klatte, Schlittmeier and Fels 2023, Applied Acoustics: cognitive performance in open-plan office acoustic simulationsopened, quote confirmedAt 55 dBA, serial-recall error 26.3 percent with nonspeech office sound vs 37.2 percent with meaningful speech (10.9 points, 95 percent credible interval 7.0 to 14.7); meaningful speech worse than foreign speech; raising the irrelevant background's T30 from 0.4 to 0.8 s cut errors 34.3 to 30.2 percent (CrI 0.7 to 7.5), 0.8 vs 1.1 s not robust; spatial separation not robust (INTERVENTION, adults, lab; 2023)
• S5Jones, Macken and Murray 1993, Memory and Cognition: disruption of visual short-term memory by changing-state auditory stimuliopened, quote confirmedChanging pitch and segmentation of an irrelevant stream impair serial recall of visually presented items; speech meaning is not necessary; the changing-state mechanism (INTERVENTION, adults, lab; 1993)
• S6Campbell, Beaman and Berry, Role of habituation in the irrelevant sound effect: token set size and rate of transitiondid not openDisruption grows with the number of distinct tokens; the largest step is from one token to two; steady-state sound does not disrupt relative to silence (INTERVENTION, adults, lab)
• S7Klatte, Lachmann and Meis 2010, Noise and Health: noise and reverberation on speech perception and listening comprehension of children and adults in a classroom-like settingopened, quote confirmedTwo virtual classrooms, RT 0.47 vs 1.1 s: RT had no effect on speech perception in silence but severely increased the impairment from background sound in all age groups; children more impaired than adults; babble hurt comprehension more than noise did (INTERVENTION, children and adults, lab; 2010)
• S17Haapakangas, Hongisto, Hyona, Kokko and Keranen 2014, Applied Acoustics: unattended speech on performance and subjective distraction, the role of acoustic designopened, quote confirmed98 volunteers in an 84 m2 open-plan office laboratory, four physically built conditions: background speech impaired short-term and working memory and raised disturbance; effects neither grew nor faded over three hours; cutting STI by room-acoustic means reduced disturbance, with smaller performance gains than expected (INTERVENTION, adults, full-scale lab; 2014)

The page may say that the sound that costs you is sound that keeps changing, other people's speech above all, and that it costs you most when you are holding a sequence in mind: writing, reading, arithmetic. It may not say that noise in general drains attention, and it may not treat a steady hum, a fan, or constant traffic as the enemy; those do not disrupt serial memory relative to silence and can be the fix when they mask speech.

The mechanism under the sound channel: a changing stream is seriated automatically and collides with rehearsal of a remembered sequence, which is why steady sound does not disrupt and why a fan or a masking floor can be the fix. Read with 3.3 and 3.7, it is the second channel in which the cost is a specific interference with the task in the head rather than a general drain, and read with 3.6 it marks the contrast in persistence: this tax neither grew nor faded across three hours while the sightline tax halved in two weeks.

Reverberation time (RT, the seconds for sound to decay 60 dB) has no measured effect on speech perception in silence; it multiplies the harm of background sound. In two virtual classrooms (RT 0.47 vs 1.1 s), the longer RT left perception in silence unchanged and severely increased the impairment from background noise in children and adults alike, with children more impaired (S7, 2010). In 68 scored ten-year-olds, recall of spoken words rose from 0.195 at +3 dB signal-to-noise to 0.316 at +12 dB, and RT (0.3 vs 1.2 s) interacted with SNR: at +12 dB the short RT helped (0.335 vs 0.296) and at +3 dB it hurt (0.185 vs 0.205), raw marginal means from a single study (S9, 2016). When the reverberation acts on the irrelevant stream instead of the wanted one, the sign flips: lengthening the irrelevant background's T30 from 0.4 to 0.8 s at matched level cut adult serial-recall error from 34.3 to 30.2 percent (4.1 points, CrI 0.7 to 7.5), with 0.8 vs 1.1 s not robust (S3, 2023); and in a Chinese open-plan simulation rising STI lowered recall under short RT but not under long RT (S10, 2023). A real classroom treatment that cut T20 from 1.6 to 0.6 s improved modelled speech reception thresholds by up to about 6 dB (S11, 2023, model fitted to the measured room). In 487 children across 21 classrooms with RT 0.49 to 1.1 s, more reverberant rooms went with poorer speech perception and spoken short-term memory, an unrandomized field association (S8, 2010). Boundary: the crossover at low SNR is one study awaiting replication; the classroom findings are children listening to a wanted voice, the office findings adults ignoring an unwanted one.

Reverberation convolves the signal with a decaying exponential that fills the dips between syllables; its modulation-depth loss and the noise's modulation-depth loss enter the modulation transfer function as a product, so RT and SNR cannot be traded one-for-one and an intervention on one has conditional returns. Early reflections add useful energy to a wanted voice while late energy masks the next syllable, which is why short RT helps at high SNR and can hurt at low SNR when the reflections were the listener's only extra signal. Applied to an unwanted voice, the same smearing lowers its intelligibility and therefore its interference (Axiom 1), so the designer's target is a product, not a time.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 (the writer claimed FORGED) · Refuted by the adversary on a number larger than the sources support · Sources: 6 cited, 5 opened and confirmed

• adversary REFUTED the claim
• citations: refetched 95% (>=90%), resolved 100% (>=80%) -> OK
• dead links: 2 of 40 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED PROBABLE: one of {grounded cross-lab adversary, verified sources} but not both
• CLAMPED: claimed FORGED -> granted PROBABLE (gate lowers, never raises)

S7 (2010, RT 0.47 vs 1.1 s, no effect in silence, amplifies noise, children worse), S9 (2016, 68 children, SNR 0.195 vs 0.316, RT by SNR crossover), S3 (2023, irrelevant T30 0.4 to 0.8 s cuts errors 4.1 points), S10 (2023, STI effect only under short RT), S11 (2023, T20 1.6 to 0.6 s, SRT up to about 6 dB), S8 (2010, 487 children, 21 classrooms, field association). openai-dr, anthropic and exa carry the interaction and the Yadav sign flip; gemini-dr states a monotonic rule that RT over 0.6 s cuts word recognition 35 to 45 percent with no opened denominator, which is not used.

• S3Yadav, Georgi, Leist, Klatte, Schlittmeier and Fels 2023, Applied Acoustics: cognitive performance in open-plan office acoustic simulationsopened, quote confirmedAt 55 dBA, serial-recall error 26.3 percent with nonspeech office sound vs 37.2 percent with meaningful speech (10.9 points, 95 percent credible interval 7.0 to 14.7); meaningful speech worse than foreign speech; raising the irrelevant background's T30 from 0.4 to 0.8 s cut errors 34.3 to 30.2 percent (CrI 0.7 to 7.5), 0.8 vs 1.1 s not robust; spatial separation not robust (INTERVENTION, adults, lab; 2023)
• S7Klatte, Lachmann and Meis 2010, Noise and Health: noise and reverberation on speech perception and listening comprehension of children and adults in a classroom-like settingopened, quote confirmedTwo virtual classrooms, RT 0.47 vs 1.1 s: RT had no effect on speech perception in silence but severely increased the impairment from background sound in all age groups; children more impaired than adults; babble hurt comprehension more than noise did (INTERVENTION, children and adults, lab; 2010)
• S8Klatte, Hellbruck, Seidel and Leistner 2010, Environment and Behavior: effects of classroom acoustics on performance and well-being in elementary school childrenopened, quote confirmed487 children in 21 classrooms, mean RT 0.49 to 1.1 s; more reverberant rooms showed poorer speech perception and spoken short-term memory; classrooms not randomly assigned (CORRELATIONAL field; 2010)
• S9Hurtig, Keus van de Poll, Pekkola, Hygge, Ljung and Sorqvist 2016, Frontiers in Psychology: children's recall of words spoken in their first and second languageopened, quote confirmed72 ten-year-olds (68 scored): recall 0.195 at +3 dB SNR vs 0.316 at +12 dB; RT 0.3 vs 1.2 s interacted with SNR, short RT better at +12 dB (0.335 vs 0.296) and worse at +3 dB (0.185 vs 0.205) (INTERVENTION, children, lab; published January 2016)
• S10Kang, Mak, Ou and Zhou 2023, Applied Acoustics: speech intelligibility and reverberation time on serial recall in Chinese open-plan officesdid not openRising STI raised disturbance and lowered serial recall in the short-RT condition but not the long-RT condition; at STI 0.42 speech was more disruptive under short RT (INTERVENTION, adults, lab; 2023)
• S11Minelli, Puglisi, Astolfi, Hauth and Warzybok 2023, IJERPH: binaural benefit from acoustical treatment in a real primary school classroomopened, quote confirmedTreatment cut T20 from 1.6 to 0.6 s; predicted speech reception thresholds improved by up to about 6 dB, longer RT worsened them by about 11 dB on average (MODEL fitted to a measured room; 2023)

The page may say that a reverberant room does not simply sound worse: it makes whatever noise is present cost more, and it can make a neighbour's speech cost less. It may not say that less reverberation is always better; in a quiet room added absorption buys little, in a noisy one it can nearly double the benefit, and at low signal-to-noise the one crossover study points the other way.

The multiplicative law of reverberation, and the sound channel's instance of the context rule shared with 1.3: reverberation has no measured effect in silence, multiplies the harm of noise on a wanted voice and lowers the cost of an unwanted one. The adversary refuted the universal form, so the low-signal crossover rides on one study awaiting replication; the interaction itself is carried by two experiments and a field association.

In randomized or within-person laboratory exposures of adults to office-level sound, the three outcomes dissociate. Forty female clerical workers randomized to three hours of low-intensity simulated open-office noise showed raised urinary epinephrine but unchanged norepinephrine and cortisol, made fewer attempts at unsolvable puzzles (40 percent fewer per the institutional release) and half as many postural adjustments, yet did not differ from the quiet group on perceived stress (S12, 2000; S13). Forty adults proofreading for eight minutes at 59.1 vs 36.3 dBA, within-person, showed heart rate 75.49 to 76.83 beats per minute (p = .013, 38 usable), phasic skin conductance 2.47 to 3.06 (p = .044) and negative affect 12.85 to 16.05 (p = .001) with proofreading errors detected unchanged (10.33 vs 11.00, p = .283) and RMSSD unchanged (S14, S15, 2021). Fifty-nine adults working about 48 minutes in 65 dB speech, 65 dB spectrally matched noise or 35 dB silence: the noise was rated more annoying, the speech more loading; both raised cortisol against silence and speech produced the stronger physiological pattern including a rising HRV LF/HF ratio (S16, 2020, between-person). Over two hours at 39 vs 51 dBA, recall, fatigue and motivation moved while cortisol and catecholamines did not (S4). In 98 volunteers in an 84 m2 built office, cutting STI reduced subjective disturbance with performance gains smaller than expected (S17, 2014). In the field, 23 preschool teachers under LAeq 74.7 dBA for four hours showed heart rate correlating with level at r = 0.40 (p = .04 one-tailed) and HRV falling across level classes while annoyance and noise sensitivity showed no effect, a correlational design that cannot separate speech, reverberation or causality (S19, 2023). Boundary: the Evans and Johnson pattern rests on one N = 40 single-sex single-occupation study; the eight-minute heart-rate change is 1.34 beats per minute, not a percentage of stress.

Annoyance is an appraisal of fit between the environment and the person's goals; performance loss is specific interference with a task's representational channel (Axiom 2); autonomic and endocrine change is the body's arousal response to a demand, which compensatory effort can mask in the performance channel. They have different inputs, different time constants and different adaptation rates, so a survey, an error count and a hormone assay sample three different systems. Compensatory effort is the bridge: it holds performance while raising physiological cost, which is the pattern the eight-minute and 48-minute experiments show.

Granted tier: FORGED, set by the code gate on October 4, 2026 · Survived the adversary's attack · Sources: 8 cited, 8 opened and confirmed

• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 95% (>=90%), resolved 100% (>=80%) -> OK
• dead links: 2 of 40 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED FORGED: survived a grounded cross-lab adversary AND primary sources confirmed

S12 (2000, N = 40 randomized, epinephrine up, cortisol null, perceived stress null), S13 (2001 release, 40 percent fewer attempts, half as many adjustments), S14 and S15 (2021, N = 40 within-person, 8 minutes, heart rate 75.49 to 76.83, affect 12.85 to 16.05, errors null), S16 (2020, N = 59, 48 minutes, speech vs matched noise vs silence, cortisol), S4 (39 vs 51 dBA two hours, cortisol null), S17 (2014, N = 98, disturbance down more than performance up), S19 (2023, 23 teachers, r = 0.40, correlational). All four legs carry the dissociation; the measurement sentences are observation, the compensatory-effort sentence is inference.

• S4Jahncke, Lulea University of Technology thesis compilation: cognitive performance in open-plan office noise (the STI-balance experiment with Hongisto and Virjonen, and the 39 vs 51 dBA two-hour experiment)opened, quote confirmedWord memory about 5 percent and arithmetic under 3 percent worse as the speech-to-masker balance rose near 51 dBA, most of the word-memory loss already at STI 0.34; 39 vs 51 dBA over two hours changed recall, fatigue and motivation but not cortisol or catecholamines (INTERVENTION, adults, lab; 2011 to 2013)
• S12Evans and Johnson 2000, Journal of Applied Psychology 85(5) 779-783: stress and open-office noiseopened, quote confirmed40 female clerical workers randomized to quiet or 3 hours of low-intensity simulated open-office noise: urinary epinephrine up, norepinephrine and cortisol not; fewer attempts at unsolvable puzzles; fewer postural adjustments; perceived stress did not differ (INTERVENTION, office workers, lab; 2000)
• S13Cornell Chronicle 2001: even low-level office noise can increase health risks (the Evans and Johnson release)opened, quote confirmedThe release's figures: 40 percent fewer attempts at an unsolvable puzzle and half as many ergonomic adjustments in the noise group (institutional summary of S12; 2001)
• S14Sander, Marques, Birt, Stead and Baumann, Open-plan office noise is stressful: multimodal stress detection in a simulated work environment (full text)opened, quote confirmed40 participants, within-person counterbalanced, 59.1 vs 36.3 dBA, 8 minutes of proofreading each: heart rate 75.49 to 76.83 bpm (p = .013, 38 usable), phasic skin conductance 2.47 to 3.06 (p = .044), negative affect 12.85 to 16.05 (p = .001), proofreading errors detected 10.33 vs 11.00 (p = .283), RMSSD unchanged (INTERVENTION, adults, lab; 2021)
• S15Sander et al. 2021, Journal of Management and Organization: the published record of the multimodal stress studyopened, quote confirmedThe journal record of S14 (same experiment; 2021)
• S16Radun, Maula, Rajala, Scheinin and Hongisto 2020, Indoor Air: speech is special, the stress effects of speech, noise and silence during tasks requiring concentrationopened, quote confirmed59 participants, one of three conditions for about 48 minutes: speech 65 dB LAeq, spectrally matched noise 65 dB, silence 35 dB; noise more annoying, speech more loading; both raised cortisol vs silence; speech raised the HRV LF/HF ratio with time and produced the stronger physiological pattern (INTERVENTION, adults, lab, between-person; 2020)
• S17Haapakangas, Hongisto, Hyona, Kokko and Keranen 2014, Applied Acoustics: unattended speech on performance and subjective distraction, the role of acoustic designopened, quote confirmed98 volunteers in an 84 m2 open-plan office laboratory, four physically built conditions: background speech impaired short-term and working memory and raised disturbance; effects neither grew nor faded over three hours; cutting STI by room-acoustic means reduced disturbance, with smaller performance gains than expected (INTERVENTION, adults, full-scale lab; 2014)
• S19Trimmel and Brachtl 2023, Noise and Health 25(118): noise in preschools and its cardiovascular effect on preschool teachersopened, quote confirmed23 teachers in nine Vienna preschools with 24-hour ECG: LAeq over the first four hours 74.7 dBA (SD 1.74); heart rate correlated with level r = 0.40 (p = .04, one-tailed); heart rate rose and HRV fell across level classes; annoyance and sensitivity showed no effect (CORRELATIONAL field; 2023)

The page may say that a room can be raising your heart rate and your stress hormones while your work output looks unchanged and you report no stress, and the converse: that a room can annoy you without costing you performance. It may not infer stress from annoyance, cost from complaints, or harmlessness from silence in a satisfaction survey, and it may not call the eight-minute heart-rate change a stress percentage.

The strongest dissociation row in the run: in adults under office-level sound, annoyance, task output and the body's stress markers move independently, and compensatory effort holds output while physiology rises. It anchors the same-law group with 1.4, 3.8 and 4.7, and it is the reason the page may say a room can raise your heart rate while your work and your own report sit still, but may never infer stress from complaints or harmlessness from a satisfaction survey.

In adult laboratory subjects, the response to an unexpected sound unfolds on nested timescales: pre-attentive deviance detection (mismatch negativity) around 100 ms, involuntary attention capture (P3a) at about 250 to 350 ms, voluntary reorienting (RON) at 400 to 600 ms (S20, 2010; S23, 2025), a behavioural perturbation in a tapping task at 750 ms that has died out by 1750 ms (S22, 2023), autonomic and affect differences measurable within an eight-minute work block (S14), cortisol differences after about 48 minutes (S16), and catecholamine and motivational change after three hours (S12). The early capture index is necessary but not sufficient for a cost: novel and deviant sounds elicited the P3a regardless of informational content, yet impaired performance only when the sound carried target-related information (S21, 2022). Boundary: the latencies come from electrophysiology in laboratory adults and the chain from capture to cost depends on context and task load (S20); these experiments do not locate the first millisecond of impairment, and an exposure duration, a neural latency and a delay imposed on feedback are three different quantities.

Acoustic arrival takes milliseconds (about 9 ms over 3 m at 343 m/s); the auditory system then runs a cascade of detection, capture and reorienting whose each stage has its own latency and recovery, so a single event costs about a second of degraded control and recovers fully. What a room costs over a day is the rate of such events times the unrecovered residue plus the slow autonomic drift, which is why 'I got used to it' (adaptation on the annoyance channel) is compatible with a continuing measured cost on the autonomic channel. The capture response is automatic evaluation of potential significance; whether it costs anything depends on whether the task has a channel for the sound to collide with (Axiom 2).

Granted tier: PROBABLE, set by the code gate on October 4, 2026 · Refuted by the adversary on a source said less than the claim · Sources: 7 cited, 7 opened and confirmed

• adversary REFUTED the claim
• citations: refetched 95% (>=90%), resolved 100% (>=80%) -> OK
• dead links: 2 of 40 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED PROBABLE: one of {grounded cross-lab adversary, verified sources} but not both

S20 (2010, subsecond orienting, magnitude context-dependent), S23 (2025, P3a 250 to 350 ms in real soundscapes), S22 (2023, tapping perturbation 750 to 1750 ms), S21 (2022, P3a without performance cost absent target-related information), S14 (2021, differences within 8 minutes), S16 (2020, cortisol at about 48 minutes), S12 (2000, catecholamines at 3 hours). openai-dr, anthropic and exa keep the clocks separate; anthropic alone carries the 'necessary but not sufficient' experiment (S21), which is why the tier claimed is PROBABLE.

• S12Evans and Johnson 2000, Journal of Applied Psychology 85(5) 779-783: stress and open-office noiseopened, quote confirmed40 female clerical workers randomized to quiet or 3 hours of low-intensity simulated open-office noise: urinary epinephrine up, norepinephrine and cortisol not; fewer attempts at unsolvable puzzles; fewer postural adjustments; perceived stress did not differ (INTERVENTION, office workers, lab; 2000)
• S14Sander, Marques, Birt, Stead and Baumann, Open-plan office noise is stressful: multimodal stress detection in a simulated work environment (full text)opened, quote confirmed40 participants, within-person counterbalanced, 59.1 vs 36.3 dBA, 8 minutes of proofreading each: heart rate 75.49 to 76.83 bpm (p = .013, 38 usable), phasic skin conductance 2.47 to 3.06 (p = .044), negative affect 12.85 to 16.05 (p = .001), proofreading errors detected 10.33 vs 11.00 (p = .283), RMSSD unchanged (INTERVENTION, adults, lab; 2021)
• S16Radun, Maula, Rajala, Scheinin and Hongisto 2020, Indoor Air: speech is special, the stress effects of speech, noise and silence during tasks requiring concentrationopened, quote confirmed59 participants, one of three conditions for about 48 minutes: speech 65 dB LAeq, spectrally matched noise 65 dB, silence 35 dB; noise more annoying, speech more loading; both raised cortisol vs silence; speech raised the HRV LF/HF ratio with time and produced the stronger physiological pattern (INTERVENTION, adults, lab, between-person; 2020)
• S20BMC Neuroscience 2010: novelty and deviance detection, electrophysiological and behavioural orienting to unexpected soundsopened, quote confirmedUnexpected sounds elicit detection and orienting responses on a subsecond scale whose magnitude depends on context and task load (INTERVENTION, adults, lab; 2010)
• S21Memory and Cognition 2022: distractor sounds, the P3a and task performanceopened, quote confirmedNovel and deviant sounds elicited the P3a regardless of informational content, yet impaired performance only when the sound carried target-related information (INTERVENTION, adults, lab; 2022)
• S22Psychonomic Bulletin and Review 2023: the behavioural time course of auditory distraction in a tapping taskopened, quote confirmedTapping speed transiently changes 750 ms after distractor onset and the effect dies out by 1750 ms, visible in first-trial data (INTERVENTION, adults, lab; 2023)
• S23eNeuro 2025: event-related potentials to spectral novelty in real-world soundscapesopened, quote confirmedP3a at 250 to 350 ms reliably elicited by spectral novelty in real-world soundscapes (INTERVENTION, adults, lab; 2025)

The page may say that a room's sound reaches your brain and bends your behaviour within a second, before deliberate thought, and that recovery from any one event is fast; it should add that what accumulates is the rate of events and the slow bodily drift. It may not read a cost off a startle, nor claim that a fast neural response proves a cost to attention.

The sound rung of the onset ladder (with 1.7 and 3.2): detection, capture, reorienting, a behavioural perturbation and autonomic drift run on nested latencies, and an early capture response fires without a performance cost when the sound carries nothing task-relevant. The adversary refuted the claim as first written, so the latencies ride as laboratory electrophysiology and the rule that a fast neural response is not a cost is the part that stands.

In adult EEG and psychophysics laboratories, a sound the listener's own action produces evokes a smaller auditory N1 (about 100 ms) and often a smaller P2 (about 200 ms) than the same sound presented externally. The attenuation holds for tones triggered by the toe as well as the hand in 12 seated participants, and action-to-sound delays from 0 to 1000 ms did not modulate it (S24, 2014); it is the same whether attention is on the sound, the movement or a visual task (S25, 2013); and it falls on the unspecific N1 component, so it mainly reflects a damped orienting response rather than a sensory-specific cancellation (S26, 2013). Perceptually, self-generated sounds are judged softer than external ones when intense and louder when faint (S27, 2015), and self-generation shifts loudness bias without changing detection thresholds or just-noticeable differences (S28, 2021). Boundary: the stimuli are button- or toe-triggered tones, not natural footsteps; N1 suppression weakens at short onset asynchronies near walking cadence (about 0.5 s) where P2 carries more of the effect; at least one study reports enhancement rather than suppression; and no attention, memory or stress outcome is measured in this literature. The microvolt magnitudes gemini-dr gives (minus 1.1 vs minus 4.3 microvolts) have no opened source and are not carried.

When the motor system issues a command it also issues an efference copy, a forward model's prediction of the sensory consequence; the auditory cortex compares arrival against prediction and reduces its response to what was predicted. Prediction does not remove the pressure waveform or its masking power, and it does not change detectability; it changes the gain on the orienting response and the felt loudness. Source identity cannot act as a hidden acoustic property: if two sounds give the same cochlear input, any self-versus-other difference must enter through motor prediction, context, belief or another sense.

Granted tier: FORGED, set by the code gate on October 4, 2026 · Survived the adversary's attack · Sources: 5 cited, 5 opened and confirmed

• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 95% (>=90%), resolved 100% (>=80%) -> OK
• dead links: 2 of 40 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED FORGED: survived a grounded cross-lab adversary AND primary sources confirmed

S24 (2014, 12 participants, hand and toe tones, 0 to 1000 ms delays), S25 (2013, attenuation independent of attention), S26 (2013, unspecific N1 component), S27 (2015, intensity-dependent loudness bias), S28 (2021, bias not sensitivity). All four legs carry the attenuation; exa carries the sensitivity dissent (S27, S28) and openai-dr the delay-insensitivity bound (S24); anthropic carries the attention-independence and the component bound (S25, S26); gemini-dr's microvolt figures are unsourced and dropped.

• S24van Elk, Salomon, Kannape and Blanke 2014, Biological Psychology: suppression of the auditory N1 by self-generated sounds from the upper and lower limbsopened, quote confirmed12 seated participants: tones triggered by hand or toe showed reduced N1 compared with external tones; action-sound delays of 0 to 1000 ms did not modulate the suppression (INTERVENTION, adults, EEG lab; 2014)
• S25Saupe, Widmann, Trujillo-Barreto and Schroger 2013, BMC Neuroscience: sensorial suppression of brain responses to self-generated sounds is independent of attentionopened, quote confirmedN1 attenuation for self-initiated vs externally initiated sounds was the same whether attention went to the sounds, the motor act or visual stimuli (INTERVENTION, adults, EEG lab; 2013)
• S26SanMiguel, Todd and Schroger 2013, Psychophysiology: suppression to self-initiated sounds reflects attenuation of the unspecific N1 componentopened, quote confirmedOnly the unspecific, not the sensory-specific, N1 component was suppressed, so N1 suppression mainly reflects an attenuated orienting response (INTERVENTION, adults, EEG lab; 2013)
• S27Reznik, Henkin, Levy and Mukamel 2015, PLoS ONE: perceived loudness of self-generated sounds is differentially modified by expected sound intensityopened, quote confirmedSelf-generated sounds judged softer than external ones at high intensity but louder at low intensity; attenuation is intensity-dependent, not fixed (INTERVENTION, adults, lab; 2015)
• S28Paraskevoudi and SanMiguel 2021, Scientific Reports: self-generation and sound intensity modulate perceptual bias but not sensitivityopened, quote confirmedSelf-generation did not change detection thresholds or just-noticeable differences; it shifted loudness bias in interaction with intensity (INTERVENTION, adults, lab; 2021)

The page may say that your brain turns down the sound of your own movements before you notice them, a measured discount at about a tenth of a second that holds for the feet as well as the hands. It may not say that this spares your attention or lowers your stress; the discount is on felt loudness and the orienting response, not on detection, and no study has measured it against attention or memory.

The efference-copy row: the brain damps its orienting response to the sound of its own action, feet included, at about the first tenth of a second, lowering felt loudness without changing detection. With 3.4 it forms the run's gating law, that prediction and suppression act before any cost is incurred, and it is the measured half of the page's footstep sentence.

In adults walking freely or on treadmills, the heard consequence of a footstep sits inside a closed control loop with millisecond tolerance. With shoe-tip microphones and delayed headphone playback in 11 walkers, 'these are my footsteps' judgments were about 90 percent at 16 and 100 ms delay, fell to about 34 and 28 percent near 250 and 450 ms, rose again at later step-cycle alignments, and stride timing changed systematically while participants did not recognise the change (S29, 2010). Shifting the frequency spectrum of the walker's own footstep sound in real time changed perceived body weight, gait (a more dynamic swing, a shorter heel strike with high frequencies augmented) and reported motivation (S30, 2015). Simulated gravel and snow footstep sounds changed the sense of effort and of sinking on a treadmill while kinematics and metabolic cost did not change (S31, 2017). A 2026 preprint in 30 adults found amplification without delay cut gait variability about 10 percent relative to baseline (S33, not peer-reviewed), while a 2026 symposium study with seven delays from 0 to 1500 ms found no consistent monotonic change in speed, cadence or step length (S32), so the gait effect is contested in direction and size. In 15 seated listeners, synchronising a synthesized footstep masker to the step rhythm lowered the speech-following threshold by about 1.1 dB (95 percent CI 0.7 to 1.8) with spectra matched, so timing alone changes a footstep's masking power (S34, 2015). Boundary: every outcome here is agency, gait, body perception, effort or masking; no opened study measures attention, memory, annoyance or a stress marker for natural self-generated footsteps against external footsteps, replay or generic noise under matched movement, and that comparison does not exist in anything the four legs could retrieve.

A footfall supplies an extended, surface-diagnostic waveform through the floor and air together with tactile, proprioceptive and vestibular input, and the forward model (Axiom 6) predicts and discounts it; a timing or spectral mismatch registers as prediction error, which the body resolves by adjusting gait and the mind reads as reduced agency or a changed body. The room's floor finish and early reflections set the spectrum and the first-reflection delay of the sound returning to the walker (carpet takes out high frequencies; a hard lobby adds them plus a 10 to 40 ms reflection), so those parameters lie on the measured axis for body perception and gait. Whether any of this reaches attention or stress is an inference from adjacent literatures, not a measurement; replay is not automatically external, because synchronized replay can remain predictable and feel self-produced, which is why the missing experiment must separate identity, contingency and predictability.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 · Refuted by the adversary on a number larger than the sources support · Sources: 6 cited, 6 opened and confirmed

• adversary REFUTED the claim
• citations: refetched 95% (>=90%), resolved 100% (>=80%) -> OK
• dead links: 2 of 40 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED PROBABLE: one of {grounded cross-lab adversary, verified sources} but not both

S29 (2010, 11 walkers, agency curve, unrecognised gait change), S30 (2015, spectral footstep illusion), S31 (2017, effort without kinematic change), S33 (2026 preprint, gait variability about 10 percent), S32 (2026, no consistent gait change across 0 to 1500 ms), S34 (2015, 15 listeners, synchrony shifts masking 1.1 dB). openai-dr, anthropic and exa each state that the matched-movement attention or stress comparison does not exist; gemini-dr supplies a comparison table for self, external, replay and generic noise that no opened study measured, and it is not carried.

• S29Menzer, Brooks, Halje, Faller, Vetterli and Blanke 2010, Cognitive Neuroscience 1(3) 184-192: feeling in control of your footstepsopened, quote confirmed11 walkers with shoe-tip microphones and delayed playback: 'my footsteps' judgments about 90 percent at 16 and 100 ms delay, about 34 and 28 percent near 250 and 450 ms, rising again at later step-cycle alignments; stride timing changed systematically and participants did not recognise the change (INTERVENTION, adults, free walking; 2010)
• S30Tajadura-Jimenez, Basia, Deroy, Fairhurst, Marquardt and Bianchi-Berthouze 2015, CHI: as light as your footstepsopened, quote confirmedA shoe prototype shifted the frequency spectrum of the walker's own footstep sound in real time; high-frequency augmentation produced a thinner felt body, a more dynamic swing, a shorter heel strike and higher reported motivation (INTERVENTION, adults, lab; 2015)
• S31Turchet, Camponogara, Nardello, Zamparo and Cesari 2017: interactive footstep sounds modulate the sense of effort without affecting kinematics or metabolic parameters during treadmill walkingopened, quote confirmedSimulated gravel and snow footstep sounds changed perceived effort and the feeling of sinking while kinematics and metabolic cost did not change (INTERVENTION, adults, treadmill lab; 2017)
• S32Kawasaki and Tanaka 2026, ISASE: effects of temporal delays in self-generated footstep sounds on gait parametersopened, quote confirmedSeven delays from 0 to 1500 ms of the walker's own footfall sound with inertial motion capture: no consistent monotonic change in walking speed, cadence or step length (INTERVENTION, adults, lab; symposium abstract March 2026, on J-STAGE May 2026)
• S33Gu and Dotov 2026 preprint: amplified and delayed self-footstep feedback versus pink-noise masking in 30 healthy adultsopened, quote confirmedAmplification without delay cut gait variability about 10 percent relative to baseline; larger delay raised cadence; masking changed gait little; outcomes are gait, not attention or stress (INTERVENTION, adults; preprint dated 26 January 2026, not peer-reviewed)
• S34Larsson et al. 2015: effects of sounds of locomotion on speech perceptionopened, quote confirmed15 seated listeners, synthesized gravel-footstep maskers at 40 to 70 dBA: synchronising the masker to the listener's step rhythm lowered the just-follow-conversation threshold by about 1.1 dB (95 percent CI 0.7 to 1.8) with spectra matched; body conduction and self-generation flagged as uninvestigated (INTERVENTION, adults, lab; 2015)

The page may say that the sound of your own footsteps changes how you walk, how heavy you feel and how hard the walking feels, before you notice any of it; that is measured. It may not say that the sound of your own footsteps acts on your attention, your mood or your stress; that is unmeasured, and the one experiment that would settle it has not been run.

The footstep adjudication row: the sound of one's own footsteps is measured on agency, gait, felt weight and effort, with no measured effect on attention, memory or stress and no matched-movement comparison in existence. The adversary refuted the claim as first written, so the gait direction rides as contested; the named absence of the attention and stress measurement is the finding the page must carry, alongside 3.8's named absence of a cognitive outcome in the home-clutter data.

In controlled measurements of lightweight wood floors, the heel impact carries the main excitation energy below 100 Hz (S37, 2021); occupants hear low-frequency thumping when someone walks, especially barefoot, as a near pure tone audible above about 15 to 16 Hz, with the floors' lowest natural frequencies at 14 to 25 Hz and poor insulation below 200 Hz (S35, 2023). In surveys of wooden residential buildings, 30 to 67 percent of participating residents in half the studied buildings were moderately to highly annoyed by footstep noise, and listening-test annoyance tracked loudness, thumpiness and reverberation and also age and prior exposure (S36, 2020, correlational). Barefoot walking has more audible low-frequency content and is rated more annoying than shod, and annoyance rose with level except for adult footsteps at 45 dB (S37). Boundary: these are residents of wooden multi-storey housing and listening-test subjects, with annoyance as the outcome; no attention, memory or stress outcome is measured; a cited higher alpha-EEG response to footsteps through low-impact-level floors reached one leg only as a citing-article summary and is not carried. Only the anthropic leg opened this literature, so the tier claimed is SIGNAL.

Porous absorption, screens and broadband masking act above roughly 200 Hz; the dominant energy of an external footfall sits below 100 Hz and couples through the floor structure as vibration and radiated low-frequency sound near the floor's own modes. Room-acoustic treatment therefore cannot address it; the levers are structural mass, decoupling and resilient flooring. The occupant's own footstep is the mirror case: its airborne, high-frequency, surface-diagnostic component is what the walker hears and what floor finish can change (Axiom 7), so one floor serves two acoustic optimisations that can conflict.

Granted tier: SIGNAL, set by the code gate on October 4, 2026 · Refuted by the adversary on a source said less than the claim · Sources: 3 cited, 3 opened and confirmed

• adversary REFUTED the claim
• citations: refetched 95% (>=90%), resolved 100% (>=80%) -> OK
• dead links: 2 of 40 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED PROBABLE: one of {grounded cross-lab adversary, verified sources} but not both

S35 (2023, thumping, 14 to 25 Hz modes, poor insulation below 200 Hz), S36 (2020, 30 to 67 percent annoyed in half the buildings, correlates), S37 (2021, heel impact below 100 Hz, barefoot vs shod, 45 dB exception). anthropic alone; the other three legs did not open the structure-borne literature.

• S35Unlocking the myth of low-frequency footstep thumping noise in lightweight wood floor construction (2023)opened, quote confirmedOccupants hear low-frequency thumping when a person walks, especially barefoot; audible above about 15 to 16 Hz; floor lowest natural frequencies 14 to 25 Hz; lightweight wood floors insulate poorly below 200 Hz (controlled physical measurement; 2023)
• S36Acta Acustica 2020: footstep noise annoyance in wooden residential buildingsopened, quote confirmedIn half of the studied wooden buildings 30 to 67 percent of participating residents were moderately to highly annoyed by footstep noise; listening-test annoyance tracked loudness, thumpiness and reverberation and also age and prior exposure (CORRELATIONAL survey plus listening test; 2020)
• S37Acta Acustica 2021: spectrum and perception of walking impact sound on lightweight floorsopened, quote confirmedHeel impact carries the main excitation energy below 100 Hz; barefoot walking has more audible low-frequency content and is rated more annoying than shod; annoyance rose with level except for adult footsteps at 45 dB (controlled physical measurement and listening test; 2021)

The page may say that the footsteps you hear from others arrive mostly through the structure, below the range any room finish controls, and that what to do about them is a construction question, not a decoration one. It may not attach a cognitive or stress cost to them; the measured outcome is annoyance in housing surveys.

A single-leg row on structure-borne footfall: another person's footsteps arrive below the range any room finish controls and are measured only as annoyance in housing surveys. It is kept as a construction note for the page, not a cognitive claim; it was considered and declined as a pair with 1.8 because both are single-direction rows about different physics.

In visual search by laboratory subjects over maps and photographs of real scenes, the cost of finding a target scales with image-based measures of the scene (Rosenholtz, Li and Nakano 2007: feature congestion, the local variability of color, orientation and luminance; subband entropy, the bits needed to encode the image; edge density, the share of pixels that are edges), not with a count of objects. Rosenholtz 2007 (INTERVENTION, laboratory, within-person, map backgrounds): the contrast threshold needed to find an arrow target correlated r = .93 with feature congestion, r = .68 with subband entropy and r = .83 with edge density across backgrounds. Henderson, Chanceaux and Smith 2009 (INTERVENTION, laboratory, real-scene photographs, eye tracking): all three measures related to search response time with R-squared 0.27 (feature congestion), 0.18 (subband entropy) and 0.28 (edge density), p < .001, so in real scenes clutter explains about a fifth to a third of the variance in search time and the rest is target-distractor similarity, scene meaning and prior knowledge. The measures only loosely agree with one another (a 2025 clutter-metric paper reporting feature congestion at 0.77 with edge density but 0.20 with subband entropy rides as an unfetched flag). Occlusion is a separate mechanism that subtracts target information rather than adding competitors: intracranial recordings in 18 epilepsy patients (2014 preprint) showed selective responses to objects shown 9 to 25 percent visible arriving about 100 ms later than to whole objects; what occlusion costs depends on whether it hides the target or a competitor. Boundary: laboratory subjects, search tasks, static images; outcome is search time, error and contrast threshold. No opened source measured a Rosenholtz-quantified room against a validated working-memory outcome in a dwelling.

The eye samples one high-resolution patch of about 2 degrees, repositioned 3 to 4 times a second; what a room costs is the number and difficulty of samples needed to resolve the target against its local surround. Salience is local feature contrast, so removing competitors raises the relative salience of the survivors (Rosenholtz's desk analogy: the note is conspicuous on the empty desk and invisible on the full one). Decluttering does not lower the amount of attention-capturing content; it concentrates it in fewer items, and whether that helps depends on whether the survivors are task-relevant.

Granted tier: SIGNAL, set by the code gate on October 4, 2026 (the writer claimed FORGED) · Refuted by the adversary on a claim no opened source could verify · Sources: 3 cited, 1 opened and confirmed

• ATTACK VOID: the refutation rests on a source that would not open (reason_class unverifiable), a miss, never a refutation; neither a kill nor a pass, the grade follows the sources test alone
• citations: refetched 71% (>=90%), resolved 100% (>=80%) -> BELOW FLOOR
• dead links: 7 of 24 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED SIGNAL: neither a grounded cross-lab adversary survived nor sources confirmed
• CLAMPED: claimed FORGED -> granted SIGNAL (gate lowers, never raises)

S1 (Rosenholtz 2007: r = .93/.68/.83 with contrast threshold), S2 (Henderson 2009: R-squared 0.27/0.18/0.28 of search RT in real scenes), S22 (occlusion recordings: about 100 ms later at 9 to 25 percent visibility). Legs: openai-dr, gemini-dr, exa and anthropic all carry S1; exa and anthropic carry S2; openai-dr carries S22.

• S1Rosenholtz, Li, Nakano 2007, Measuring visual clutter, Journal of Visiondid not openImage-based clutter measures (feature congestion, subband entropy, edge density) predict search cost; r = .93/.68/.83 with contrast threshold; INTERVENTION, lab.
• S2Henderson, Chanceaux, Smith 2009, The influence of clutter on real-world scene search, Journal of Visiondid not openAll three clutter measures relate to search RT in real scenes, R-squared 0.27/0.18/0.28; clutter as proxy for set size; INTERVENTION, eye tracking.
• S22Object completion under partial visibility, 2014 preprint (intracranial recordings)opened, quote confirmed18 epilepsy patients: selective responses to objects 9 to 25 percent visible arrive about 100 ms later than to whole objects; occlusion subtracts target information; INTERVENTION.

The page may use 'a cluttered sightline' only as a measured image property at the occupant's eye position (edge density, feature congestion), never as a tidiness judgment; it may say clutter raises the price of each visual search (about a fifth to a third of search-time variance in real scenes) and must not say it fills or clutters the mind.

The sightline channel's estimand row: clutter is image statistics around the target at the occupant's eye position, not an object count, and those statistics explain a fifth to a third of search time in real scenes. The gate graded it on its sources alone because the adversary's refutation rested on a page that would not open, so the exchange rate rides as a direction; it belongs with 1.1 and 4.2 as the law that every channel's acting quantity is measured at the occupant.

In laboratory search displays with electroencephalography (INTERVENTION, within-person, adults), a distractor's priority signal and its suppression or selection resolve within about 100 to 350 ms of display onset: a distractor-elicited positivity (Pd) near 100 ms, the target-selection negativity (N2pc) at 180 to 200 ms, a later suppression positivity at 260 to 340 ms (Gaspar et al. 2016, PNAS; the label of the pre-N2pc positivity is contested in a 2025 paper the anthropic leg surfaced unfetched, the timing is not). Ansorge, Kiss and Eimer 2009 (INTERVENTION, n = 8): target-colored cues shown 17 ms and masked produced selection and cueing effects while cue localization did not differ from chance. Adams and Gaspelin 2020 (INTERVENTION): trial-by-trial reports did distinguish capture episodes (probe accuracy at the distractor 41.7 percent on reported-capture trials versus 33.8 percent on no-capture trials), but participants' global ratings of how distracted they were correlated r = .01 with their measured response-time cost. At the scale of a screen, eye tracking on 40 website screenshots classified by a clutter algorithm (Ergonomics 2021, INTERVENTION, within-person) showed a wider spread of attention on high-clutter pages within the first 3 seconds of search. Boundary: millisecond onsets are for a salient item entering a controlled display and are the onset of selection, not of a learning or stress effect; room-scale attention effects are measured over minutes, and the home-cortisol literature has no acute onset.

Selection is resolved by competition in visual cortex and its biasing by frontoparietal signals in the first few hundred milliseconds; verbal report requires later, slower processes. Preference reports about a room are a different, later channel and not a readout of attentional cost; r = .01 is the measured gap between the two channels.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 (the writer claimed FORGED) · Survived the adversary's attack · Sources: 4 cited, 3 opened and confirmed

• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 71% (>=90%), resolved 100% (>=80%) -> BELOW FLOOR
• dead links: 7 of 24 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED PROBABLE: one of {grounded cross-lab adversary, verified sources} but not both
• CLAMPED: claimed FORGED -> granted PROBABLE (gate lowers, never raises)

S8 (Gaspar 2016: Pd near 100 ms, N2pc 180 to 200 ms, suppression 260 to 340 ms), S9 (Ansorge 2009: 17 ms masked cues select, localization at chance, n = 8), S10 (Adams and Gaspelin 2020: 41.7 vs 33.8 percent; global ratings r = .01 with RT cost), S20 (website eye tracking: spread differs within 3 s). Legs: anthropic (S8), openai-dr (S9, S10), exa (S20), gemini-dr (pre-conscious suppression via secondary pages).

• S8Gaspar et al. 2016, Inability to suppress salient distractors predicts low visual working memory capacity, PNASopened, quote confirmedPd timing and size predict capacity; target N2pc at 180 to 200 ms does not; suppression positivity 260 to 340 ms; INTERVENTION, EEG.
• S9Ansorge, Kiss, Eimer 2009, Goal-driven attentional capture by invisible colorsopened, quote confirmedTarget-colored cues shown 17 ms and masked produce selection and cueing effects while localization is at chance (n = 8); INTERVENTION, EEG.
• S10Adams, Gaspelin 2020, Assessing introspective awareness of attention captureopened, quote confirmedTrial-by-trial reports distinguish capture (41.7 vs 33.8 percent probe accuracy); global distraction ratings correlate r = .01 with RT cost; INTERVENTION, lab.
• S20How do we react to cluttered displays? Evidence from the first seconds of visual search in websites, Ergonomics 2021did not open40 website screenshots classified by clutter algorithm: attention spread differs on high-clutter pages within the first 3 seconds; INTERVENTION, within-person.

'Act on you before you have formed a thought' is upheld for the sightline component at the level of selection, with the number 100 to 350 ms; the page must not extend it to mood or stress, and must not use anyone's impression of a room as evidence of its attentional cost. This is the onset row the synthesis vector needs.

The sightline rung of the onset ladder, and the run's cleanest evidence that what people later say about a room does not track its measured cost: selection or suppression resolves within a few hundred milliseconds, before report, and retrospective distraction ratings are uncorrelated with the response-time cost. With 1.7 and 2.5 it upholds the page's 'before you have formed a thought' for selection and bounds it against mood and stress.

Lavie, Hirst, de Fockert and Viding 2004 (INTERVENTION, within-person, laboratory adults, flanker-type distractors, about 10 participants per experiment): raising the perceptual load of the task cut distractor interference from a mean 92 ms to 14 ms; adding a working-memory load raised distractor interference by 39 ms under low perceptual load and 43 ms under high perceptual load; in another experiment incompatible-versus-compatible interference was 140 ms under low memory load and 193 ms under high (n = 10). 'More cognitive load' is not one thing: load on perception protects against distraction and load on control exposes to it. Boundary: response-time interference from a single distractor in a controlled display, adults, tens of milliseconds; the manipulation loaded the task, not the room; load theory's mechanism is disputed (critics argue the perceptual versus working-memory split may reduce to one attentional load), so the direction of the two effects is carried, not the model.

Finite perceptual capacity: a demanding foveal task consumes the capacity that would otherwise spill to peripheral competitors, while the frontal control system that keeps the task goal active is what a memory load taxes, and a weakened goal lets competitors through. For a room this predicts that a busy field behind the screen costs most during perceptually easy, memory-heavy work (holding a plan, an argument, a number) and least during dense visual work (reading fine text, close inspection). The room's cost is not a property of the room.

Granted tier: SIGNAL, set by the code gate on October 4, 2026 (the writer claimed PROBABLE) · Refuted by the adversary on a claim no opened source could verify · Sources: 1 cited, 0 opened and confirmed

• ATTACK VOID: the refutation rests on a source that would not open (reason_class unverifiable), a miss, never a refutation; neither a kill nor a pass, the grade follows the sources test alone
• citations: refetched 71% (>=90%), resolved 100% (>=80%) -> BELOW FLOOR
• dead links: 7 of 24 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED SIGNAL: neither a grounded cross-lab adversary survived nor sources confirmed
• CLAMPED: claimed PROBABLE -> granted SIGNAL (gate lowers, never raises)

S6 (Lavie 2004: 92 to 14 ms under perceptual load; +39 to +43 ms under working-memory load; 140 vs 193 ms, n = 10). Legs: openai-dr and anthropic both opened S6 and report the same tables; gemini-dr argued the working-memory side from templates without the paper; exa silent.

• S6Lavie, Hirst, de Fockert, Viding 2004, Load theory of selective attention and cognitive control, JEP: Generaldid not openPerceptual load cuts distractor interference 92 to 14 ms; working-memory load adds 39 to 43 ms; 140 vs 193 ms (n = 10); INTERVENTION, lab.

'Sightlines act on you' may stand only with the conditional that the cost depends on what you are doing and that a demanding visual task gates it out; any page sentence that gives a room a fixed cognitive cost is overstated.

The task-in-the-head row: a perceptually demanding task gates distractors out and a working-memory load lets them in, so the same wall costs different amounts by what the occupant is doing. The gate graded it on sources alone after an unopenable refutation, so the millisecond magnitudes ride as a direction; with 2.2 it is the second channel where the cost is specific interference, and with 3.7 it predicts the on-desk versus off-desk geometry.

Gaspelin, Leonard and Luck 2015, Experiment 4 (INTERVENTION, within-person, n = 24): a salient irrelevant color singleton made target responses 17 ms faster, not slower (630 versus 647 ms, p < .001), and probe recall at the singleton's location was 24.2 percent against 32.6 percent at ordinary distractor locations, the signature of active suppression. Vogel, McCollough and Machizawa 2005 (INTERVENTION with individual differences, EEG): higher-capacity individuals stored only the relevant objects, lower-capacity individuals also stored distractors; filtering efficiency correlated r = .69 (color task) and r = .62 (location task) with working-memory capacity. Gaspar et al. 2016 (PNAS, EEG): the target N2pc did not vary with capacity, but the timing and size of the distractor Pd strongly predicted capacity, so low-capacity individuals suppress a salient distractor too late to stop it diverting attention. Fine and Minnery 2009 (INTERVENTION, 3 to 5 icons on a map, eye tracking): recall of an icon's location improved with its computed salience, more so as difficulty rose, with no difference in fixation time, so more visual energy on a task-relevant item helps memory. Boundary: adults in controlled displays; outcomes are response time, probe recall and EEG indices; the r values are between-person correlations, not an estimate of how much a cluttered room reduces anyone's capacity. McMains and Kastner 2011 (fMRI) and Kastner 1998 (Science) show competing stimuli suppress each other's cortical response (gemini-dr read the 1998 paper as about a fifth less signal in that vector with four stimuli than one); that measures neural signal under passive viewing, not task outcome.

Biased competition: items in nearby receptive fields mutually suppress; attention biases the competition toward the goal; a proactive suppression signal can veto a salient item before it captures. The strength and timing of that veto is a stable person variable. Because the same variable plausibly affects planning and tidying, a person who suppresses poorly is more likely both to keep a cluttered space and to be distracted in it: the room-mind correlation has a common cause inside the person, a structural feature of any cross-sectional clutter study.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 · Survived the adversary's attack · Sources: 6 cited, 6 opened and confirmed

• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 71% (>=90%), resolved 100% (>=80%) -> BELOW FLOOR
• dead links: 7 of 24 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED PROBABLE: one of {grounded cross-lab adversary, verified sources} but not both

S7 (Gaspelin 2015: 630 vs 647 ms; probe recall 24.2 vs 32.6 percent), S5 (Vogel 2005: r = .69 and .62), S8 (Gaspar 2016: Pd predicts capacity), S19 (Fine and Minnery 2009: salience helps recall), S4 (McMains and Kastner 2011), S3 (Kastner 1998). Legs: openai-dr (S7, S5, S4), anthropic (S8), exa (S19), gemini-dr (S3).

• S7Gaspelin, Leonard, Luck 2015, Direct evidence for active suppression of salient-but-irrelevant sensory inputs, Psychological Scienceopened, quote confirmedSalient singleton present: 630 vs 647 ms (17 ms faster), probe recall 24.2 vs 32.6 percent, n = 24; INTERVENTION, lab.
• S5Vogel, McCollough, Machizawa 2005, Neural measures reveal individual differences in controlling access to working memory, Natureopened, quote confirmedFiltering efficiency correlates r = .69 (color) and r = .62 (location) with working-memory capacity; low-capacity people store distractors; INTERVENTION, EEG.
• S8Gaspar et al. 2016, Inability to suppress salient distractors predicts low visual working memory capacity, PNASopened, quote confirmedPd timing and size predict capacity; target N2pc at 180 to 200 ms does not; suppression positivity 260 to 340 ms; INTERVENTION, EEG.
• S19Fine, Minnery 2009, Visual salience affects performance in a working memory task, J Neurosciopened, quote confirmedRecall of an icon's location improves with its computed salience, more so as difficulty rises, with no difference in fixation time; INTERVENTION, eye tracking.
• S4McMains, Kastner 2011, Interactions of top-down and bottom-up mechanisms in human visual cortex, J Neurosciopened, quote confirmedPerceptual grouping reduces competition without changing element count; INTERVENTION, fMRI.
• S3Kastner et al. 1998, Mechanisms of directed attention in the human extrastriate cortex, Scienceopened, quote confirmedSimultaneously presented stimuli evoke smaller cortical signals than sequential ones (about a fifth less in V4); biased competition; INTERVENTION, fMRI.

The page must not say visible objects take up slots in the mind; it may say the brain spends effort vetoing them and that people differ in how well they do it. Any argument from 'my space, my mind' is confounded by the person.

The suppression-trait row and the hidden common cause behind every 'cluttered room, cluttered person' observation: salient irrelevant items are often vetoed at no cost, and how well a person vetoes predicts both distractibility and working-memory capacity. With 2.6 it forms the gating law, and with 1.2's fifty-fold sensitivity spread it is why the person's variance confounds the room's in any cross-sectional study.

Fisher, Godwin and Seltman 2014 (INTERVENTION, within-person, 23 to 24 kindergarten children, mostly high-SES, six 5-to-7-minute science lessons alternating a heavily decorated and a sparse laboratory classroom): off-task time 38.58 percent of lesson time decorated against 28.42 percent sparse, a 10.16-point difference, d = 0.85 (paired 95 percent interval reconstructed by the openai-dr leg from the reported t statistics, about 5.9 to 14.5 points); post-lesson accuracy 42 percent against 55 percent, d = 0.65 (about 3.9 to 22.1 points); learning gain above pretest 18 against 33 points, d = 0.73. Godwin et al. 2022 (INTERVENTION, within-person, n = 42, laboratory classroom, two weeks): off-task time 46.56 percent decorated against 32.86 percent streamlined, paired t(41) = 10.01, Hedges g = 1.51; learning not measured. Hanley et al. 2017 (INTERVENTION, within-person, 34 primary children with and without autism, eye tracking of a teacher composited over decorated or blank backgrounds): gaze shifted markedly to the decorated background, more so in the autistic group, yet mini-lesson scores were 10.1 against 9.6 out of 18 (about 2.8 points of the scale), p = .06 with age adjustment and p = .418 with nonverbal-ability adjustment; time looking at displays was the strongest within-child predictor of learning, a correlation, not an experimental effect. Boundary: children aged about 5 to 9, first exposure to an unfamiliar decorated room, lessons of minutes, an extreme contrast (full decoration against bare walls); the manipulation also changed novelty and visual boundary, so the package is measured, with gaze plausibly mediating part of it; not adults, not offices, not steady state; children learned in both conditions.

Decoration adds high-salience task-irrelevant content inside the child's field during a low-perceptual-load listening task, the condition under which the load gate predicts the largest distractor cost; overt gaze moves to it, and the share of the lesson encoded drops by less than the share of gaze lost because listening continues while looking away.

Granted tier: SIGNAL, set by the code gate on October 4, 2026 (the writer claimed FORGED) · Refuted by the adversary on a claim no opened source could verify · Sources: 3 cited, 1 opened and confirmed

• ATTACK VOID: the refutation rests on a source that would not open (reason_class unverifiable), a miss, never a refutation; neither a kill nor a pass, the grade follows the sources test alone
• citations: refetched 71% (>=90%), resolved 100% (>=80%) -> BELOW FLOOR
• dead links: 7 of 24 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED SIGNAL: neither a grounded cross-lab adversary survived nor sources confirmed
• CLAMPED: claimed FORGED -> granted SIGNAL (gate lowers, never raises)

S11 (Fisher 2014: 38.58 vs 28.42 percent, d = 0.85; 42 vs 55 percent, d = 0.65; 18 vs 33 points, d = 0.73), S13 (Godwin 2022: 46.56 vs 32.86 percent, g = 1.51, n = 42), S12 (Hanley 2017: 10.1 vs 9.6 of 18, p = .06 / .418, n = 34). Legs: all four carried Fisher; openai-dr and anthropic opened Godwin 2022; openai-dr opened the Hanley manuscript, exa the abstract.

• S11Fisher, Godwin, Seltman 2014, Visual environment, attention allocation, and learning in young children, Psychological Sciencedid not openDecorated vs sparse lab classroom, 23 to 24 kindergarteners, 5 to 7 minute lessons: off task 38.58 vs 28.42 percent (d = 0.85), accuracy 42 vs 55 percent (d = 0.65), gains 18 vs 33 points (d = 0.73); INTERVENTION, within-person.
• S13Godwin et al. 2022, Cognitive Science, classroom decorations and habituationdid not openOff task 46.56 vs 32.86 percent, g = 1.51 (n = 42); environmental off-task difference falls 16.41 to 6.22 points in two weeks (SE 1.29); 4.6 percent of time on the environment in real classrooms over 15 weeks; INTERVENTION plus CORRELATIONAL.
• S12Hanley et al. 2017, Classroom displays, attraction or distraction, Developmental Psychology (accepted manuscript)opened, quote confirmed34 children, eye tracking of a teacher over decorated or blank backgrounds: large gaze shift; learning 10.1 vs 9.6 of 18, p = .06 (age) and .418 (nonverbal ability); INTERVENTION, within-person.

The one place the page's claim is measured as a cognitive outcome is a kindergarten laboratory classroom; the page may cite about 10 points more off-task time and 13 points lower test accuracy in one small experiment with five-year-olds and must not generalize it to adults or to a familiar room.

The one place the page's clutter claim is measured as a cognitive outcome: decorated versus bare laboratory classrooms, five-year-olds, first exposure, lessons of minutes, with the gaze shift about twice the learning loss. The gate graded it on sources alone after an unopenable refutation, so the points and effect sizes ride as the owning vector's numbers rather than gate-confirmed figures; read with 3.6 and 3.7 it is the top of the clutter channel's own rigour gradient.

Godwin et al. 2022 (INTERVENTION, within-person, laboratory classroom, two weeks): the decorated-minus-streamlined difference in environmental off-task behavior fell from an estimated 16.41 points at the start to 6.22 points after two weeks (final estimate SE 1.29, p < .001); the children were still more distracted in the decorated room in week 2, but the share of time attending the decorations fell significantly between weeks. In real elementary classrooms observed over 15 weeks (CORRELATIONAL, same paper), children spent about 4.6 percent of time attending to the environment, close to the 4.7 percent in the sparse laboratory room's third lesson; environmental distraction did not decline across the 15 weeks, but the displays themselves kept changing. Imuta and Scarf 2014 (ARGUMENT) argued the original 2014 effect carried a novelty component, and the 2014 analysis collapsed across lessons in a way that could mask habituation. Boundary: children, classrooms, overt gaze and off-task coding; learning was not measured in the habituation study, so persistent looking cannot be converted to a persistent learning deficit; no adult habituation measurement was opened.

Priority signals for static items decay with repeated exposure (habituation of orienting), while the suppression signal becomes cheaper with practice; a wall that no longer changes stops generating new priority. The sightline channel therefore has a fast onset and a fast decay, which separates it from the acoustic and air channels, where adaptation is weak; a synthesis that gives light, sound, air and clutter one persistence coefficient is wrong on this term (an inference from the measured decay, not a measured comparison).

Granted tier: SIGNAL, set by the code gate on October 4, 2026 (the writer claimed FORGED) · Refuted by the adversary on a claim no opened source could verify · Sources: 2 cited, 1 opened and confirmed

• ATTACK VOID: the refutation rests on a source that would not open (reason_class unverifiable), a miss, never a refutation; neither a kill nor a pass, the grade follows the sources test alone
• citations: refetched 71% (>=90%), resolved 100% (>=80%) -> BELOW FLOOR
• dead links: 7 of 24 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED SIGNAL: neither a grounded cross-lab adversary survived nor sources confirmed
• CLAMPED: claimed FORGED -> granted SIGNAL (gate lowers, never raises)

S13 (Godwin 2022: 16.41 to 6.22 points, SE 1.29; 4.6 percent in real classrooms; 4.7 percent in the sparse lab room's third lesson), S21 (Imuta and Scarf 2014: the novelty account). Legs: openai-dr (16.41 to 6.22), anthropic (significant decrease between weeks; 4.6 percent), gemini-dr (habituation acknowledged via secondary pages).

• S13Godwin et al. 2022, Cognitive Science, classroom decorations and habituationdid not openOff task 46.56 vs 32.86 percent, g = 1.51 (n = 42); environmental off-task difference falls 16.41 to 6.22 points in two weeks (SE 1.29); 4.6 percent of time on the environment in real classrooms over 15 weeks; INTERVENTION plus CORRELATIONAL.
• S21Imuta, Scarf 2014, commentary on Fisher et al., Frontiers in Psychologyopened, quote confirmedThe novelty account of the 2014 decorated-classroom effect; ARGUMENT.

'A cluttered sightline creates a cluttered mind' read as a standing condition of a room one lives in is refuted on this axiom: measured steady-state attention to decoration is about 5 percent of time, not the 39 percent of first exposure; the page may say the first look costs most.

The decay row, and the run's one measured persistence term: the decorated-room distraction fell by more than half within two weeks and runs near a twentieth of lesson time in familiar classrooms. The gate graded it on sources alone after an unopenable refutation. Its inference that the clutter channel habituates where the speech cost (2.2, no change over three hours) and the body's response to sound (2.5) do not is the seed of the persistence meta-axiom, and it is the row that refutes 'a cluttered sightline creates a cluttered mind' as a standing condition.

Norberg et al. 2023 (INTERVENTION, within-person, counterbalanced, 123 adults: 63 with elevated hoarding symptoms and 60 controls, tested in a cluttered and a tidy room; Journal of Behavioral Addictions 12(3)): the room effect on a sustained-attention detectability measure (Conners CPT-3) was d = 0.02, 95 percent CI −0.15 to 0.20, p = .74, a null (read from the paper's results on 2026-10-05); the room-by-group interaction on the Victoria Stroop was d = 0.42 (0.17 to 0.67) for the hoarding group only; because participants changed rooms, room identity rode with the manipulation. Vohs, Redden and Rahinel 2013 (INTERVENTION, between-person): participants in a disorderly room produced the same number of uses for a ping-pong ball and ideas judges rated more creative; orderly-room participants chose the 'classic' option. The 2018 replication in Frontiers in Psychology (INTERVENTION, about twice the original sample, multiple measures) found no difference in creativity or executive control between order and disorder and wrote that the relation 'does not warrant the claims it generated in the popular press'. The same paper reports Chae and Zhu's workstation study, where clutter displayed vertically behind and above the desk worsened Stroop performance and shortened persistence on an unsolvable puzzle, and notes that this more salient off-surface placement may be what produced a cost that on-desk clutter did not. Boundary: adults, single sessions, laboratory or office rooms; outcomes are sustained attention, inhibition, creativity ratings and Stroop; no study measured a Rosenholtz-quantified room against validated working memory; the Chae and Zhu numbers were not opened by any leg and ride only as a direction.

The load gate predicts this geometry: clutter on the work surface sits inside the task's own perceptual-load envelope and is gated out; clutter in the surrounding field is governed by the loadable, failable control system. The operative variable is on-task-surface versus off-task-surface and salience, not quantity.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 · Survived the adversary's attack · Sources: 3 cited, 3 opened and confirmed

• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 71% (>=90%), resolved 100% (>=80%) -> BELOW FLOOR
• dead links: 7 of 24 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED PROBABLE: one of {grounded cross-lab adversary, verified sources} but not both

S14 (Norberg 2023: room effect on CPT-3 detectability d = 0.02, −0.15 to 0.20, p = .74, n = 123; VST room-by-group d = 0.42), S17 (Vohs 2013: disorderly room, more creative ideas), S18 (2018 replication: null on creativity and executive control; carries the Chae and Zhu account). Legs: openai-dr (S14), anthropic (S17, S18); gemini-dr and exa carried no adult room experiment.

• S14Norberg et al. 2023, Effect of environmental clutter on attention performance in hoarding, Journal of Behavioral Addictions 12(3) 827-839opened, quote confirmed123 adults (63 hoarding, 60 control), cluttered vs tidy room counterbalanced: room effect on CPT-3 detectability d = 0.02 (−0.15 to 0.20), p = .74; VST room-by-group d = 0.42 for the hoarding group; INTERVENTION, within-person.
• S17Vohs, Redden, Rahinel 2013, Physical order produces healthy choices, generosity, and conventionality, whereas disorder produces creativity, Psychological Scienceopened, quote confirmedDisorderly room: same number of ideas, rated more creative; orderly room: the classic option; INTERVENTION, between-person.
• S18Workspace disorder does not influence creativity and executive functions, Frontiers in Psychology 2018opened, quote confirmedReplication at about twice the original sample: no difference in creativity or executive control between order and disorder; reports Chae and Zhu's off-surface clutter cost on Stroop and persistence; INTERVENTION.

'Space is a cognitive instrument' is bounded here: for adults in a room for a session, measured instrument effects on attention and executive tasks are d about 0 to small and geometry-dependent; the page must not say clutter impairs adult thinking, and may say a busy field behind and above the work surface is the placement with a measured cost.

The adult null and the geometry rule: a cluttered room moved adult sustained attention by nothing, workspace disorder moved neither creativity nor executive control in a doubled-sample replication, and the one measured adult cost came from clutter behind and above the desk. It is the row that refutes the page's clutter sentence for adults and the clutter channel's entry in the design-rigour gradient (4.5); with 2.2 and 3.3 it belongs to the interference law, placement relative to the task standing where the task's load stands there.

Saxbe and Repetti 2010 (CORRELATIONAL, 60 adults in 30 dual-income middle-class couples, three weekdays, four saliva samples a day): the exposure was language in a self-narrated home tour (a 'stressful home' score counting words of clutter, mess and unfinished projects), not any measurement of the sightline; wives with higher scores had flatter diurnal cortisol slopes (stressful-home by time coefficient 0.08, SE 0.02, on the model's scale, with a quadratic time term) and mood that declined across the day; husbands showed no such pattern; results held with marital satisfaction and neuroticism controlled; attention and working memory were not measured (on 2026-10-05 the primary PDF opened as a scan with no text layer; the sample, days, word-count exposure and flatter-slope sign were confirmed from a summary page; the coefficient rides on the openai-dr leg's reading of the primary). Bodrij et al. 2021 (INTERVENTION, within-person, 96 female students aged 18 to 25, an infant simulator in a neutral and a chaotic living room): salivary alpha-amylase was higher in the chaotic room; self-reported negative emotion did not differ by condition; neither mediated caregiving sensitivity. Roster, Ferrari and Jurkat 2016 (CORRELATIONAL, cross-sectional SEM survey of U.S. and Canadian adults) linked self-rated clutter to lower psychological home and well-being with no direction identified. Fingerman et al. 2021 (CORRELATIONAL, 286 adults 65 and over, living rooms rated from photographs): more functional limitations went with more clutter, and among them clutter went with fewer depressive symptoms, so the sign of the clutter-mood association is not fixed across populations. Boundary: women in couples with children (2010), female students in a caregiving simulation (2021), survey respondents (2016), older adults (2021); at least four causal models fit the 2010 data (visible demands raise stress; stress changes how the same home is described; stress makes maintenance harder; a third cause such as time scarcity, money or caregiving load drives both); sampling middle-class couples narrows but does not remove socioeconomic confounding; the 2021 experiment improves causal inference for acute arousal only and says nothing about cortisol, cognition or chronic exposure.

What the 2010 study plausibly captured is appraisal: an object matters because it signifies an unfinished obligation, a route distinct from visual competition (a bill has low edge density; patterned wallpaper carries no obligation). The appraisal route runs through the stress axis over hours and days; the competition route runs through selection in milliseconds. No opened source connects them.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 (the writer claimed FORGED) · Survived the adversary's attack · Sources: 4 cited, 3 opened and confirmed

• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 71% (>=90%), resolved 100% (>=80%) -> BELOW FLOOR
• dead links: 7 of 24 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED PROBABLE: one of {grounded cross-lab adversary, verified sources} but not both
• CLAMPED: claimed FORGED -> granted PROBABLE (gate lowers, never raises)

S15 (Saxbe and Repetti 2010: 30 couples, 3 weekdays, 0.08 SE 0.02, wives only, no cognitive outcome), S16 (Bodrij 2021: sAA up, mood unchanged, n = 96), S23 (Roster 2016: cross-sectional SEM), S24 (Fingerman 2021: clutter with fewer depressive symptoms among the functionally limited, n = 286). Legs: openai-dr (opened S15), exa (S16), anthropic (S23; marked Saxbe UNVERIFIED and the secondary accounts contradictory), gemini-dr (S15 via secondary pages; S24 in its citation list).

• S15Saxbe, Repetti 2010, No place like home: home tours correlate with daily patterns of mood and cortisol, Personality and Social Psychology Bulletinopened, quote confirmed60 adults in 30 dual-income couples, 3 weekdays, 4 saliva samples a day; exposure is word counts in self-narrated home tours; wives with higher stressful-home scores had flatter diurnal cortisol slopes (stressful-home by time 0.08, SE 0.02) and declining mood; husbands did not; no attention or memory measured; CORRELATIONAL.
• S16Bodrij et al. 2021, The causal effect of household chaos on stress and caregiving: an experimental study, Comprehensive Psychoneuroendocrinologyopened, quote confirmed96 female students, neutral vs chaotic living room with an infant simulator: salivary alpha-amylase higher in chaos; negative emotion unchanged; no mediation of caregiving sensitivity; INTERVENTION, within-person.
• S23Roster, Ferrari, Jurkat 2016, The dark side of home: assessing possession clutter on subjective well-being, Journal of Environmental Psychologydid not openCross-sectional SEM survey of U.S. and Canadian adults: self-rated clutter with lower psychological home and well-being; no direction identified; CORRELATIONAL.
• S24Fingerman et al. 2021, Late life in the living room: room decor, functional limitations, and personality, The Gerontologistopened, quote confirmed286 adults 65 and over, living rooms rated from photographs: more functional limitations, more clutter; among them clutter with fewer depressive symptoms; CORRELATIONAL.

The page may not cite the cortisol literature for 'a cluttered mind'; the cognitive half of that sentence is absent from the dataset usually cited for it. It may say, with the 2010 numbers, that women who describe their homes as cluttered show flatter daily cortisol curves, as an association in 30 couples.

The home-clutter-and-stress row: the dataset usually cited for 'a cluttered mind' measures self-narrated home descriptions and a cortisol slope in wives of thirty couples, never attention or memory, and the one experiment found arousal without a change in reported mood. With 1.4, 2.4 and 4.7 it belongs to the dissociation law, and with 2.7 it is the second named absence the page must carry.

A room acts on an occupant through two transfers only, heat and inhaled air, and each obeys a conservation law. Heat: metabolic production minus work equals convection plus radiation plus conduction plus evaporation plus storage, so air temperature enters only through the convective and vapor-pressure terms while mean radiant temperature, air speed, humidity, clothing and metabolic rate carry comparable leverage (NIOSH 2016 criteria document). Air: for a well-mixed room V dC/dt = N g + Q (C_out minus C), so the steady-state excess of any occupant-sourced gas is N g / Q and the approach time constant is V / Q (NIST 2017 CO2 generation report; NIST CONTAM framework). Illustrative model consequences, not measurements: with g = 0.005 L/s per person and 450 ppm outdoors, 10 L/s per person gives about 950 ppm, 5 gives about 1,450, 2 gives about 2,950; a 50 m3 room with two people and 10 L/s has a time constant of about 1.4 hours and reads about 960 ppm after one hour, so a low early reading proves nothing about ventilation adequacy (three time constants to reach 95 percent of the rise). Neither law supplies a dose-response for thinking: a unique cognitive optimum at 22°C, a universal impairment line at 1,000 ppm, or a fixed percentage gain per doubling of ventilation would each need an empirical dose-response, and the opened pooled studies do not agree on one (Porras-Salazar 2021; ASHRAE 2025 position document). Evidence class: MODEL and physics reference; population-independent because it is physics, cognition-silent for the same reason.

Energy and mass conservation. Heat must balance or be stored; occupant-sourced gases accumulate in proportion to source rate over outdoor-air flow with a lag set by room volume over flow. Because per-person CO2 generation is a near-constant of metabolism, the CO2 reading is an algebraic transform of outdoor air per occupant, a flow gauge, and the same denominator Q governs every other human-sourced emission.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 (the writer claimed FORGED) · Survived the adversary's attack · Sources: 5 cited, 5 opened and confirmed

• PROBATION: 1 misquote flag elsewhere in this vector (below the kill line of 2); this axiom's own sources were clean -> capped at PROBABLE
• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 97% (>=90%), resolved 96% (>=80%) -> OK
• dead links: 1 of 29 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED FORGED: survived a grounded cross-lab adversary AND primary sources confirmed
• PROBATION CAP: earned FORGED -> PROBABLE (one misquote elsewhere in the vector)
• CLAMPED: claimed FORGED -> granted PROBABLE (gate lowers, never raises)

S1 (heat-balance determinants: temperature, humidity, air movement, radiation, clothing, metabolic rate), S2 (CO2 generation rates and the steady-state and time-constant relations), S3 (contaminant balance with sources, penetration, deposition, filtration), S4 (ASHRAE: CO2 is a ventilation indicator, not an overall air-quality indicator), S20 (no universal temperature-performance curve across 18 to 34°C). Observation: the balance equations and their determinants. Inference: CO2 as flow gauge; no cognition coefficient in either law.

• S1NIOSH Criteria for a Recommended Standard: Occupational Exposure to Heat and Hot Environments (2016)opened, quote confirmedHeat balance determinants: air temperature, humidity, air movement, radiation, clothing and metabolic rate interact; air temperature alone does not set the body's heat load (A1, A6, A7).
• S2Persily and de Jonge, NIST, Carbon dioxide generation rates for building occupants (2017)opened, quote confirmedPer-person CO2 generation and the well-mixed mass balance: steady-state excess = N g / Q, time constant V / Q; occupant CO2 to O2 relation (A1, A3).
• S3NIST TN 1887r1, CONTAM contaminant transport frameworkopened, quote confirmedEvery indoor contaminant obeys a balance of sources, penetration, deposition, filtration and reaction with the same outdoor-air denominator; CO2 cannot certify the absence of other pollutants (A1, A2).
• S4ASHRAE Position Document on Indoor Carbon Dioxide (approved 2025-02-12)opened, quote confirmedIndoor CO2 evaluates ventilation through indoor-outdoor differences, is not an overall indoor-air-quality indicator, occupants are exposed to mixtures not pure CO2, and evidence for direct CO2 effects at common indoor levels is inconsistent (A1, A2, A3).
• S20Porras-Salazar et al. 2021, Building and Environment: meta-analysis of 35 studies examining the effect of indoor temperature on office work performanceopened, quote confirmedNo usable temperature-performance relation over 20 to 30°C or 18 to 34°C; the Seppänen 2006 curve refit on its own data gives R² = 0.05, no better than a flat model (A1, A6).

The architecture page may describe heat and air as physical inputs that reach the body before any thought, but only as transfers; no sentence may attach a cognition coefficient to a thermostat setting or a CO2 number.

The conservation row: heat balance and mass balance fix what a room can deliver to a body, and neither contains a cognition coefficient. With 1.1's radiometry it states the run's dose law, that physics fixes the dose at the occupant and says nothing about what the dose does; the gate capped it only because a sibling row in the vector misquoted a source.

Bioeffluents, human-borne volatile compounds, skin-oil oxidation products and shed particles each follow N g_j / Q with their own source term, so they are structurally collinear with CO2 excess in any real room; a field regression of performance on CO2 therefore measures the occupancy-to-ventilation ratio and everything riding on it. Only a chamber that breaks the ratio identifies the molecule. Zhang, Wargocki and Lian 2017 (INTERVENTION; 25 adults; 255-minute exposures; 500, 1,000, 3,000 ppm) did so: at the same 3,000 ppm, pure CO2 added at unchanged ventilation produced no statistically significant change in cognitive performance, acute symptoms or perceived air quality, while restricted ventilation with bioeffluents at 3,000 ppm reduced perceived air quality, raised headache, fatigue, sleepiness and difficulty thinking, and slowed addition. The dissociation shows the mixture, not the gas, carried the effect; it does not identify which companion did. ASHRAE's 2025 position document states the structure as policy: CO2 evaluates ventilation through indoor-outdoor differences, is not an overall indoor-air-quality indicator, and occupants breathe mixtures. The 302-worker office cohort (CORRELATIONAL, 2021) reads its own coefficient correctly as 'lower ventilation rates, as assessed by CO2.' Boundary: in 36 children aged 10 to 12 (INTERVENTION, 2023), neither 2,000 to 3,000 ppm pure CO2 nor bioeffluents during sleep changed next-morning CANTAB scores, with 45 to 70 minutes of ventilated washout weakening the test. The chess study that instrumented CO2, fine particles and temperature together reached the legs through news pages only and is UNVERIFIED here.

Shared-denominator collinearity from the mass balance: every occupant-sourced contaminant is diluted by the same outdoor-air flow Q, so in field data CO2 excess and companion emissions rise and fall together and cannot be separated statistically; only an experimental break of the ratio (pure dosing at fixed flow, or scrubbing) identifies the gas.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 (the writer claimed FORGED) · Survived the adversary's attack · Sources: 5 cited, 5 opened and confirmed

• PROBATION: 1 misquote flag elsewhere in this vector (below the kill line of 2); this axiom's own sources were clean -> capped at PROBABLE
• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 97% (>=90%), resolved 96% (>=80%) -> OK
• dead links: 1 of 29 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED FORGED: survived a grounded cross-lab adversary AND primary sources confirmed
• PROBATION CAP: earned FORGED -> PROBABLE (one misquote elsewhere in the vector)
• CLAMPED: claimed FORGED -> granted PROBABLE (gate lowers, never raises)

S5 (the 2017 dissociation: same 3,000 ppm, pure CO2 null, bioeffluents impaired), S4 (ASHRAE: CO2 is a ventilation indicator; occupants breathe mixtures), S3 (every contaminant shares the balance structure), S17 (the cohort's own reading of CO2 as a ventilation proxy), S13 (children: no next-morning effect of either source during sleep, with washout caveat). Observation: S5's two-source contrast. Inference: companions of CO2, not CO2, act in under-ventilated rooms.

• S5Zhang, Wargocki, Lian 2017, Indoor Air: effects of exposure to carbon dioxide and bioeffluentsopened, quote confirmed25 adults, 255-minute exposures at 500, 1,000 and 3,000 ppm: pure CO2 at unchanged ventilation produced no significant cognitive, symptom or perceived-air-quality effect; bioeffluents at 3,000 ppm did (A2, A3, A5).
• S4ASHRAE Position Document on Indoor Carbon Dioxide (approved 2025-02-12)opened, quote confirmedIndoor CO2 evaluates ventilation through indoor-outdoor differences, is not an overall indoor-air-quality indicator, occupants are exposed to mixtures not pure CO2, and evidence for direct CO2 effects at common indoor levels is inconsistent (A1, A2, A3).
• S3NIST TN 1887r1, CONTAM contaminant transport frameworkopened, quote confirmedEvery indoor contaminant obeys a balance of sources, penetration, deposition, filtration and reaction with the same outdoor-air denominator; CO2 cannot certify the absence of other pollutants (A1, A2).
• S17Cedeño Laurent et al. 2021, Environmental Research Letters: associations between acute exposures to PM2.5 and CO2 and cognitive function in office workersopened, quote confirmed302 office workers, 42 buildings, six countries, 12 months: a 315 ppm CO2 increase (as a ventilation proxy) associated with about 0.85 percent slower Stroop responses (95% CI 0.32 to 1.39); CO2 95th percentile about 1,363 ppm (A2, A4, A5).
• S13The effect of air quality on sleep and cognitive performance in school children aged 10 to 12 years: double-blinded placebo-controlled crossover (2023)opened, quote confirmed36 children sleeping at 700 ppm, 2,000 to 3,000 ppm pure CO2, or 2,000 to 3,000 ppm with bioeffluents: no next-morning CANTAB effect; 45 to 70 minutes of ventilated washout before testing (A2).

Any page sentence about air must name ventilation or the air the room supplies, never carbon dioxide, as the thing acting on thinking; a CO2 sensor is defensible as a flow gauge and tracer only.

The flow-gauge row: in any occupied room CO2 and every human-sourced emission share one denominator, so a field coefficient on CO2 estimates the occupancy-to-airflow bundle, never the molecule, and the one chamber that broke the ratio found the mixture acted and the gas did not. It belongs with 1.1 and 3.1 as the estimand law and is why the page may say 'ventilation' and never 'carbon dioxide' as the thing that acts.

Early positives, both on the Strategic Management Simulation (SMS): Satish 2012 (INTERVENTION; 22 adults; 2.5 h per condition; 600, 1,000, 2,500 ppm added pure CO2) reported raw SMS scores 11 to 23 percent lower at 1,000 ppm and 44 to 94 percent lower at 2,500 ppm than at 600 on seven of nine scales; Allen 2016's pure-CO2 arm (24 office workers; full workdays; about 550, 945, 1,400 ppm) reported SMS scores about 15 and 50 percent lower at the two elevated levels. Later pure-gas tests: Zhang 2017 (25 adults, 3,000 ppm) null; Rodeheffer 2018 (36 submariners; 600, 2,500, 15,000 ppm) no difference on nine SMS measures; Scully 2019 (22 astronaut-like adults; double-blind crossover; 600, 1,200, 2,500, 5,000 ppm) no dose-response, some measures lower at 1,200 but similar to or above baseline at 2,500 and 5,000; Snow 2019 (31 adults; about 830 versus 2,700 ppm; under 40 min) no broad deterioration, two domains missing the practice gain; Flagner 2025 (20 adults; randomized single-blind crossover; 900 versus 3,000 ppm for 8 h) no robust impairment after multiple-testing correction, powered for about a one-standard-deviation effect; Herbig 2026 (398 adults; randomized; pure CO2 to 4,200 ppm with TVOC to 2,100 µg/m3) no systematic effect across eight cognitive domains. Mechanistic bound: raising room CO2 from 450 to 2,500 ppm adds about 1.6 mmHg of inspired partial pressure against an alveolar setpoint regulated near 40 mmHg by metabolic production and the chemoreflex, so a large effect at 1,000 to 2,500 ppm would need a non-respiratory mechanism no study has produced. Boundaries: null is not zero (Flagner's power statement; chambers mostly 2.5 h against 8-h days and nights; Rodeheffer's selected population); Fan 2023's pooled complex-task SMD of minus 2.04 (CI minus 2.62 to minus 1.47) at 1,000 to 1,500 ppm versus minus 0.86 (CI minus 1.38 to minus 0.34) at 1,500 to 3,000 ppm is non-monotonic and driven by the few complex-decision studies, while its children-under-ventilation estimate of minus 0.12 (CI minus 0.23 to minus 0.01) is the bundle, not the gas.

Respiratory homeostasis: alveolar CO2 is set by metabolic production over alveolar ventilation and defended by a chemoreflex, so a millimetre-of-mercury change in inspired CO2 is absorbed inside normal diurnal variation; with no respiratory lever, a direct cognitive effect at room levels lacks a mechanism, and the replication record matches that prediction.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 (the writer claimed FORGED) · Survived the adversary's attack · Sources: 12 cited, 11 opened and confirmed

• PROBATION: 1 misquote flag elsewhere in this vector (below the kill line of 2); this axiom's own sources were clean -> capped at PROBABLE
• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 97% (>=90%), resolved 96% (>=80%) -> OK
• dead links: 1 of 29 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED FORGED: survived a grounded cross-lab adversary AND primary sources confirmed
• PROBATION CAP: earned FORGED -> PROBABLE (one misquote elsewhere in the vector)
• CLAMPED: claimed FORGED -> granted PROBABLE (gate lowers, never raises)

S6 and S7 (the two SMS positives, 2012 and 2016), S5, S8, S9, S10, S11, S12 (seven pure-gas tests 2017 to 2026, null or inconsistent), S18 (pooled SMDs by band, non-monotonic), S29 and S2 (the partial-pressure and metabolic bound), S4 (ASHRAE: direct-effect evidence inconsistent). Observation: nine interventions, two positive sharing one instrument, seven null. Inference: no established direct, large, general effect; small or long-exposure effects unresolved.

• S6Satish et al. 2012, Environmental Health Perspectives: Is CO2 an indoor pollutant?opened, quote confirmed22 adults, 2.5 h per condition at 600, 1,000, 2,500 ppm added pure CO2: SMS scores 11 to 23 percent lower at 1,000 and 44 to 94 percent lower at 2,500 ppm on seven of nine scales (A3, A5).
• S7Allen et al. 2016, Environmental Health Perspectives: cognitive function scores in green and conventional office environmentsopened, quote confirmed24 office workers, six workdays: SMS scores 61 percent higher in Green and 101 percent higher in Green+ than Conventional, with VOCs, ventilation and CO2 changed together; pure-CO2 arm about 15 and 50 percent lower at about 945 and 1,400 ppm (A3, A5).
• S5Zhang, Wargocki, Lian 2017, Indoor Air: effects of exposure to carbon dioxide and bioeffluentsopened, quote confirmed25 adults, 255-minute exposures at 500, 1,000 and 3,000 ppm: pure CO2 at unchanged ventilation produced no significant cognitive, symptom or perceived-air-quality effect; bioeffluents at 3,000 ppm did (A2, A3, A5).
• S8Rodeheffer et al. 2018, Aerospace Medicine and Human Performance: acute CO2 exposure and decision making in submarinersopened, quote confirmed36 submarine-qualified sailors, parallel groups at 600, 2,500 and 15,000 ppm: no significant difference on nine SMS measures (A3, A5).
• S9Scully et al. 2019, npj Microgravity: acute CO2 exposures and decision making in astronaut-like subjectsopened, quote confirmed22 adults, double-blind crossover at 600, 1,200, 2,500, 5,000 ppm: no clear dose-response; some measures lower at 1,200 ppm but similar to or above baseline at higher levels (A3, A5).
• S10Snow et al. 2019, Building and Environment: exploring the physiological, neurophysiological and cognitive performance effects of elevated CO2opened, quote confirmed31 adults at about 830 versus 2,700 ppm added CO2 for under 40 minutes: no broad deterioration in reaction time, attention or working memory; two domains lacked the practice gain (A3).
• S11Flagner et al. 2025, Indoor Environments: eight-hour exposure to 3,000 ppm CO2opened, quote confirmed20 healthy adults, randomized single-blind crossover, 900 versus 3,000 ppm for 8 h: no robust cognitive, economic-preference or physiological impairment after multiple-testing correction; powered for about a one-standard-deviation effect (A3, A5).
• S12Herbig et al. 2026, International Journal of Hygiene and Environmental Health: do carbon dioxide, volatile organic compounds and atmospheric pressure affect cognitive abilitiesopened, quote confirmed398 healthy adults, randomized, 11 combinations of pure CO2 up to 4,200 ppm, TVOC up to 2,100 µg/m³ and pressure, eight cognitive domains: no systematic negative effect (A3, A5).
• S18Fan et al. 2023, Building and Environment: short-term exposure to indoor carbon dioxide and cognitive task performance, systematic review and meta-analysisopened, quote confirmedPooled complex-task SMD minus 2.04 (CI minus 2.62 to minus 1.47) at 1,000 to 1,500 ppm and minus 0.86 (CI minus 1.38 to minus 0.34) at 1,500 to 3,000 ppm, non-monotonic; children under ventilation manipulation SMD minus 0.12 (CI minus 0.23 to minus 0.01); occupational population less impaired (A3, A5).
• S29Respiratory physiology reference on alveolar CO2 regulation (PubMed 35217911)did not openAlveolar CO2 is regulated near 40 mmHg with metabolic production as the dominant term, so a room rise from 450 to 2,500 ppm (about 1.6 mmHg inspired) does not translate to an equal arterial rise (A3).
• S2Persily and de Jonge, NIST, Carbon dioxide generation rates for building occupants (2017)opened, quote confirmedPer-person CO2 generation and the well-mixed mass balance: steady-state excess = N g / Q, time constant V / Q; occupant CO2 to O2 relation (A1, A3).
• S4ASHRAE Position Document on Indoor Carbon Dioxide (approved 2025-02-12)opened, quote confirmedIndoor CO2 evaluates ventilation through indoor-outdoor differences, is not an overall indoor-air-quality indicator, occupants are exposed to mixtures not pure CO2, and evidence for direct CO2 effects at common indoor levels is inconsistent (A1, A2, A3).

The page may not say that CO2 fogs thinking; if it names CO2 at all, it names it as a gauge of how much outdoor air the room supplies per person.

The replication row: pure CO2 at room levels produced no reproducible cognitive decrement in blinded crossovers through the largest randomized test, and the two early effects shared one instrument and one laboratory lineage. Null is not zero, as the owning vector says; but together with 4.5 it refutes the famous air numbers and is the run's clearest refutation of a folk law ('stale air fogs thinking') by the dated record.

Wargocki and Wyon 2007 (INTERVENTION; pupils in classrooms) raised outdoor-air supply from 3.0 to 8.5 L/s per person, moving CO2 from about 1,300 to 900 ppm; four numerical and language tasks became faster with no significant change in errors, and the group's summary puts a doubling of ventilation at about an 8 percent gain in speed of schoolwork with no significant error change. Twardella 2012 (INTERVENTION; cluster-randomized crossover; 20 classrooms; 417 pupils; 2,366 tests; median CO2 1,045 versus 2,115 ppm) found no significant change in the primary concentration-performance score or in items processed, and 1.65 additional errors in the worse-air condition (95% CI 0.42 to 2.87). Haverinen-Shaughnessy 2011 (CORRELATIONAL; 87 classrooms; 1.9 to 15 cfm per person) found each 1 L/s per person of added outdoor air associated with 2.9 percent more students passing state math and 2.7 percent more passing reading, with the authors' caveat that the line may level off at higher rates. A 2023 pooling of five ventilation studies with 3,679 participants (MODEL; reached through one leg) reports arithmetic speed 13.7 percent better (95% CI 6.2 to 20.5), arithmetic error rate 16.1 percent lower (95% CI minus 30.8 to 0.0, touching zero) and general cognition 3.5 percent better (95% CI 0.9 to 6.1) under better ventilation. In offices (CORRELATIONAL; 302 workers; 42 buildings; 12 months; most readings below 1,363 ppm) a 315 ppm CO2 rise as a ventilation proxy went with 0.85 percent slower Stroop responses (95% CI 0.32 to 1.39). Population: schoolchildren for the interventions, office workers for the cohort; outcome: speed on school tasks and pass rates, not learning capacity; dose: the gain sits on the low-ventilation side, 3 to 8.5 L/s per person.

Dilution of the companion emissions of CO2 (A2) acts on arousal and effort rather than capacity: speed rises in every intervention while errors move in one direction in one study, not at all in another, and with an interval touching zero in the pooling, the signature of an alertness channel, not a capacity channel.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 (the writer claimed FORGED) · Survived the adversary's attack · Sources: 5 cited, 5 opened and confirmed

• PROBATION: 1 misquote flag elsewhere in this vector (below the kill line of 2); this axiom's own sources were clean -> capped at PROBABLE
• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 97% (>=90%), resolved 96% (>=80%) -> OK
• dead links: 1 of 29 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED FORGED: survived a grounded cross-lab adversary AND primary sources confirmed
• PROBATION CAP: earned FORGED -> PROBABLE (one misquote elsewhere in the vector)
• CLAMPED: claimed FORGED -> granted PROBABLE (gate lowers, never raises)

S14 (3.0 to 8.5 L/s per person; faster, errors unchanged; about 8 percent per doubling), S15 (cluster-randomized crossover; primary outcome unchanged; 1.65 more errors, CI 0.42 to 2.87), S16 (87 classrooms; 2.9 and 2.7 percent more passing per 1 L/s per person), S19 (pooled 13.7 percent speed, CI 6.2 to 20.5; error CI touching zero), S17 (offices: 0.85 percent per 315 ppm, CI 0.32 to 1.39). Observation: speed effects in every row. Inference: alertness channel.

• S14Wargocki and Wyon 2007, the effects of outdoor air supply rate and supply air filter condition in classrooms on the performance of schoolworkopened, quote confirmedRaising classroom outdoor air from 3.0 to 8.5 L/s per person (CO2 about 1,300 to 900 ppm) made four numerical and language tasks faster with no significant error change; a doubling of ventilation about 8 percent faster schoolwork (A4).
• S15Twardella et al. 2012, Indoor Air: effect of classroom air quality on students' concentration, a cluster-randomized crossoveropened, quote confirmed417 pupils, 20 classrooms, 2,366 tests, median CO2 1,045 versus 2,115 ppm: no significant change in the primary concentration-performance score or items processed; 1.65 additional errors (95% CI 0.42 to 2.87) in worse air (A4).
• S16LBNL IAQ Science: ventilation rates and school performance (summary of Haverinen-Shaughnessy et al. 2011)opened, quote confirmedAcross 87 classrooms with 1.9 to 15 cfm per person, each 1 L/s per person of added outdoor air associated with 2.9 percent more students passing math and 2.7 percent more passing reading; may level off at higher rates (A4).
• S192023 meta-analysis of ventilation interventions on cognitive performance (PMC10139101; reached by the Gemini leg only)opened, quote confirmedPooling five studies with 3,679 participants: arithmetic speed 13.7 percent better (95% CI 6.2 to 20.5), arithmetic error rate 16.1 percent lower (95% CI minus 30.8 to 0.0), general cognition 3.5 percent better (95% CI 0.9 to 6.1) under better ventilation (A4).
• S17Cedeño Laurent et al. 2021, Environmental Research Letters: associations between acute exposures to PM2.5 and CO2 and cognitive function in office workersopened, quote confirmed302 office workers, 42 buildings, six countries, 12 months: a 315 ppm CO2 increase (as a ventilation proxy) associated with about 0.85 percent slower Stroop responses (95% CI 0.32 to 1.39); CO2 95th percentile about 1,363 ppm (A2, A4, A5).

A page sentence may say the air a room supplies moves how fast people work by a few percent, in classrooms and offices, and may not say it makes them smarter or more accurate.

The honest air magnitude: added outdoor air in classrooms buys a small, repeatable gain in speed of work and little or inconsistent change in accuracy, and the office cohort's CO2 coefficient is a ventilation proxy. With 1.5 and 2.1 it fixes the few-percent scale at which room-level channels move healthy people, and it is the air sentence the page may keep.

Ordered by design, the opened studies read: a single-laboratory bundle with 24 participants and one decision-simulation instrument reports SMS scores 61 percent higher in Green and 101 percent higher in Green+ than Conventional, with VOCs, ventilation and CO2 changed together (Allen 2016, INTERVENTION bundle); chamber dosing on the same instrument reports tens of percent (Satish 2012); a 12-month field cohort of 302 workers in 42 buildings reports about 0.85 percent slower Stroop responses per 315 ppm of CO2 as a ventilation proxy (Cedeño Laurent 2021, CORRELATIONAL); blinded pure-gas crossovers report about zero (Zhang 2017, Rodeheffer 2018, Scully 2019, Flagner 2025); the largest randomized test (Herbig 2026, 398 adults) reports no systematic effect to 4,200 ppm. The pooled literature carries the same signature: Fan 2023's complex-task SMD at 1,000 to 1,500 ppm (minus 2.04) is more than twice the effect at 1,500 to 3,000 ppm (minus 0.86), a non-monotonic pattern from a few high-leverage decision studies, and the occupational population is reported less impaired than the general one. A percentage change on an SMS-derived score is a change on that scale, not in intelligence or in correct decisions. This axiom is a MODEL-class reading laid across INTERVENTION and CORRELATIONAL rows: it is not a within-study manipulation, it says nothing about which companion emission acts, and it does not reduce the field estimate to zero (the 0.85 percent per 315 ppm interval excludes zero).

Measurement physics: when effect size is ordered by design quality rather than by dose, the large values are measuring the instrument and the bundled manipulation, and the small values are measuring the exposure; expectancy, learning effects in unblinded crossovers and a single sensitive instrument inflate the top of the gradient.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 (the writer claimed FORGED) · Survived the adversary's attack · Sources: 9 cited, 9 opened and confirmed

• PROBATION: 1 misquote flag elsewhere in this vector (below the kill line of 2); this axiom's own sources were clean -> capped at PROBABLE
• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 97% (>=90%), resolved 96% (>=80%) -> OK
• dead links: 1 of 29 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED FORGED: survived a grounded cross-lab adversary AND primary sources confirmed
• PROBATION CAP: earned FORGED -> PROBABLE (one misquote elsewhere in the vector)
• CLAMPED: claimed FORGED -> granted PROBABLE (gate lowers, never raises)

S7 (61 and 101 percent, bundle, n = 24), S6 (tens of percent, same instrument), S17 (0.85 percent per 315 ppm, CI 0.32 to 1.39, 302 workers), S5, S8, S9, S11 (pure-gas nulls), S12 (398 adults, null to 4,200 ppm), S18 (non-monotonic pooled SMDs; occupational population less impaired). Observation: the ordering. Inference: instrument and bundle, not air, at the top.

• S7Allen et al. 2016, Environmental Health Perspectives: cognitive function scores in green and conventional office environmentsopened, quote confirmed24 office workers, six workdays: SMS scores 61 percent higher in Green and 101 percent higher in Green+ than Conventional, with VOCs, ventilation and CO2 changed together; pure-CO2 arm about 15 and 50 percent lower at about 945 and 1,400 ppm (A3, A5).
• S6Satish et al. 2012, Environmental Health Perspectives: Is CO2 an indoor pollutant?opened, quote confirmed22 adults, 2.5 h per condition at 600, 1,000, 2,500 ppm added pure CO2: SMS scores 11 to 23 percent lower at 1,000 and 44 to 94 percent lower at 2,500 ppm on seven of nine scales (A3, A5).
• S17Cedeño Laurent et al. 2021, Environmental Research Letters: associations between acute exposures to PM2.5 and CO2 and cognitive function in office workersopened, quote confirmed302 office workers, 42 buildings, six countries, 12 months: a 315 ppm CO2 increase (as a ventilation proxy) associated with about 0.85 percent slower Stroop responses (95% CI 0.32 to 1.39); CO2 95th percentile about 1,363 ppm (A2, A4, A5).
• S5Zhang, Wargocki, Lian 2017, Indoor Air: effects of exposure to carbon dioxide and bioeffluentsopened, quote confirmed25 adults, 255-minute exposures at 500, 1,000 and 3,000 ppm: pure CO2 at unchanged ventilation produced no significant cognitive, symptom or perceived-air-quality effect; bioeffluents at 3,000 ppm did (A2, A3, A5).
• S8Rodeheffer et al. 2018, Aerospace Medicine and Human Performance: acute CO2 exposure and decision making in submarinersopened, quote confirmed36 submarine-qualified sailors, parallel groups at 600, 2,500 and 15,000 ppm: no significant difference on nine SMS measures (A3, A5).
• S9Scully et al. 2019, npj Microgravity: acute CO2 exposures and decision making in astronaut-like subjectsopened, quote confirmed22 adults, double-blind crossover at 600, 1,200, 2,500, 5,000 ppm: no clear dose-response; some measures lower at 1,200 ppm but similar to or above baseline at higher levels (A3, A5).
• S11Flagner et al. 2025, Indoor Environments: eight-hour exposure to 3,000 ppm CO2opened, quote confirmed20 healthy adults, randomized single-blind crossover, 900 versus 3,000 ppm for 8 h: no robust cognitive, economic-preference or physiological impairment after multiple-testing correction; powered for about a one-standard-deviation effect (A3, A5).
• S12Herbig et al. 2026, International Journal of Hygiene and Environmental Health: do carbon dioxide, volatile organic compounds and atmospheric pressure affect cognitive abilitiesopened, quote confirmed398 healthy adults, randomized, 11 combinations of pure CO2 up to 4,200 ppm, TVOC up to 2,100 µg/m³ and pressure, eight cognitive domains: no systematic negative effect (A3, A5).
• S18Fan et al. 2023, Building and Environment: short-term exposure to indoor carbon dioxide and cognitive task performance, systematic review and meta-analysisopened, quote confirmedPooled complex-task SMD minus 2.04 (CI minus 2.62 to minus 1.47) at 1,000 to 1,500 ppm and minus 0.86 (CI minus 1.38 to minus 0.34) at 1,500 to 3,000 ppm, non-monotonic; children under ventilation manipulation SMD minus 0.12 (CI minus 0.23 to minus 0.01); occupational population less impaired (A3, A5).

The page may cite air at the one-percent to few-percent scale with a population attached; it may not cite the doubling.

The design-rigour gradient, stated for air and generalized by this run across channels: the reported size of an effect falls by orders of magnitude as blinding, sample and instrument independence rise, in light (1.5, pooled versus field and preprint), in clutter (3.5's first-exposure children against 3.7's adult null) and in sound (2.1's saturating few points). It is the law behind the number rule in the claim ledger.

Porras-Salazar 2021 (MODEL; pooled 35 studies; office-type tasks) found no usable common relation between temperature and office work performance over 20 to 30°C or over 18 to 34°C; refitting the 2006 Seppänen curve (the '2 percent per degree above 25°C' rule) on its own data gave R² = 0.05, a prediction accuracy no better than a model in which temperature does nothing. Lin 2025 (MODEL; pooled 30 studies; 17 to 30°C) found that against 21 to 25°C, temperatures above 25°C slowed responses (p = 0.002) and lowered the composite (p = 0.04) with accuracy inconclusive (p = 0.08), while temperatures below 21°C showed no decrement; the decline was larger on skilled tasks; pooled magnitudes are not in the opened record. Abbasi 2019 (INTERVENTION; 35 male students; four chamber sessions at 18, 22, 26, 30°C) found worse N-back accuracy and slower responses at 18 and 30 than at 22, with heart-rate changes. The model that fits is the extended-U of Hancock and Warm (ARGUMENT, theory): performance is flat across a wide band while cheap homeostatic controls (vasomotor tone, clothing, air movement) hold the heat balance and falls when they no longer do. Boundaries: office workers and students in chambers and offices; office-type tasks and N-back, not learning; 17 to 34°C, nothing above; the two meta-analyses overlap in literature and disagree, so the honest state is an unresolved warm-side effect of small size, not a settled curve; S22 is 35 men in short sessions; humidity and air speed have clear roles in the heat balance but no opened source gives an independent cognitive dose-response for either (UNVERIFIED).

Thermoregulatory reserve: cognition is unloaded while vasomotor control, clothing and air movement hold core temperature; the warm side costs more because heat must be shed by sweating and blood-flow redistribution, which the body pays for before discomfort registers, giving a plateau with a warm-side decline rather than a peak.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 (the writer claimed FORGED) · Survived the adversary's attack · Sources: 6 cited, 6 opened and confirmed

• PROBATION: 1 misquote flag elsewhere in this vector (below the kill line of 2); this axiom's own sources were clean -> capped at PROBABLE
• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 97% (>=90%), resolved 96% (>=80%) -> OK
• dead links: 1 of 29 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED FORGED: survived a grounded cross-lab adversary AND primary sources confirmed
• PROBATION CAP: earned FORGED -> PROBABLE (one misquote elsewhere in the vector)
• CLAMPED: claimed FORGED -> granted PROBABLE (gate lowers, never raises)

S20 (35 studies; no relation 18 to 34°C; Seppänen refit R² = 0.05), S21 (30 studies; above 25°C slower, p = 0.002; below 21°C no decrement), S22 (35 students; 18 and 30 worse than 22°C on N-back), S23 (extended-U model), S1 (heat-balance determinants including humidity and air speed), S25 (fan at 29°C partly restores speed). Observation: the pooled nulls and the one-sided finding. Inference: reserve model.

• S20Porras-Salazar et al. 2021, Building and Environment: meta-analysis of 35 studies examining the effect of indoor temperature on office work performanceopened, quote confirmedNo usable temperature-performance relation over 20 to 30°C or 18 to 34°C; the Seppänen 2006 curve refit on its own data gives R² = 0.05, no better than a flat model (A1, A6).
• S21Lin et al. 2025, the effects of temperature on work performance in the typical office temperature range: a meta-analysis of 30 studiesopened, quote confirmedWithin 17 to 30°C, temperatures above 25°C slowed responses (p = 0.002) and lowered composite performance (p = 0.04) versus 21 to 25°C, accuracy inconclusive (p = 0.08); below 21°C no decrement; larger on skilled tasks (A6).
• S22Abbasi et al. 2019, Health Promotion Perspectives: the impact of indoor air temperature on the executive functions of human brainopened, quote confirmed35 male students at 18, 22, 26 and 30°C in a chamber: 18 and 30°C produced worse N-back accuracy and slower responses than 22°C, with heart-rate changes (A6).
• S23Hancock and Ganey 2003, From the inverted-U to the extended-U: the evolution of a law of psychologyopened, quote confirmedThe maximal-adaptability model: performance is stable across a broad band of stress while adaptive reserve holds and declines only at extremes (A6).
• S1NIOSH Criteria for a Recommended Standard: Occupational Exposure to Heat and Hot Environments (2016)opened, quote confirmedHeat balance determinants: air temperature, humidity, air movement, radiation, clothing and metabolic rate interact; air temperature alone does not set the body's heat load (A1, A6, A7).
• S25Schiavon et al. 2017, Indoor Air: thermal comfort, perceived air quality and cognitive performance when personally controlled air movement is used by tropically acclimatized personsopened, quote confirmed56 acclimatized adults, 90 min at 23, 26, 29°C: task speed best at 26°C; a fan partly mitigated the 29°C loss; comfort at 26 and 29°C with a fan equaled or beat 23°C while performance did not (A6, A7).

The page may say warmth above the mid-twenties Celsius slows work a little and cold to 17°C does not; it may not state a degree-by-degree law or an optimum temperature.

The temperature plateau: air temperature alone predicts office performance poorly, the measured decrement sits on the warm side above the mid-twenties Celsius and is small, and the cool side shows none, which refutes the per-degree rule and the folk 'cooler is sharper'. Like 2.3 it is a channel with no monotone law; the two overlapping meta-analyses disagree and the owning vector keeps the disagreement rather than resolving it.

Lan et al. 2022 (INTERVENTION; 36 adults; 4.5 h each at 24, 26 and 28°C) held thermal comfort constant by adjusting clothing and air speed; subjects rated 24°C comfortably cool, 26°C comfortably neutral and 28°C comfortably warm; self-estimated performance did not differ, yet 12 of 14 objective cognitive metrics fell significantly at the higher temperatures, with lower parasympathetic activity (PNN50) and lower arterial oxygen saturation. The two percentages in the record are 10 percent and 6 percent for 26 and 28°C; two retrieved sources carry them in opposite order, so the ordering is UNRESOLVED and the magnitude is read as order 5 to 10 percent, non-monotonic. Schiavon 2017 (INTERVENTION; 56 tropically acclimatized adults; 90 min at 23, 26, 29°C, 60 percent RH, with and without personal fans): task speed best at 26°C; the fan partly mitigated the loss at 29°C; comfort at 26 and 29°C with a fan equaled or exceeded comfort at 23°C, so comfort equivalence did not deliver performance equivalence. On the air side the inversion runs the other way: in the Zhang and Wargocki program's summary, students rated air quality higher at 1,500 ppm pure CO2 than at 600 ppm while performance did not fall. Boundaries: healthy adults in chambers; objective batteries and self-ratings; 23 to 29°C over 1.5 to 4.5 hours. The 'slightly cool is optimal' extension (gemini-dr's PMV = minus 0.689 from a 2025 pooled model fit, one leg, redirect link) is a MODEL claim with no second opening and is not in the statement.

Autonomic cost precedes sensation: the work of shedding heat (vasodilation, parasympathetic withdrawal, lower SpO2) is spent before conscious discomfort registers, so a comfort vote is a poor instrument for the cognitive cost of a room, and a preference rating is a poor instrument for air.

Granted tier: SIGNAL, set by the code gate on October 4, 2026 (the writer claimed FORGED) · Refuted by the adversary on a source said less than the claim · Sources: 4 cited, 3 opened and confirmed

• TRIPWIRE: this axiom's own source(s) S28 (number_not_on_page) were wrong -> SIGNAL (the tripwire hits the axiom whose use of a source was wrong)
• CLAMPED: claimed FORGED -> granted SIGNAL (gate lowers, never raises)

S24 (36 adults; comfort held; 12 of 14 objective metrics down; self-rating flat; PNN50 and SpO2 lower; 10 and 6 percent, ordering unresolved), S25 (56 acclimatized adults; comfort equal with fan, speed not), S28 (air quality rated higher at 1,500 than 600 ppm pure CO2), S1 (clothing and air speed as comfort levers in the balance). Observation: ratings and scores move independently. Inference: autonomic cost precedes sensation.

• S24Lan et al. 2022, Indoor Air: cognitive performance was reduced by higher air temperature even when thermal comfort was maintained over the 24 to 28°C rangeopened, quote confirmed36 adults, 4.5 h at 24, 26, 28°C with comfort held by clothing and air speed: self-estimated performance unchanged, 12 of 14 objective metrics fell (10 and 6 percent; ordering unresolved), lower PNN50 and SpO2 (A7).
• S25Schiavon et al. 2017, Indoor Air: thermal comfort, perceived air quality and cognitive performance when personally controlled air movement is used by tropically acclimatized personsopened, quote confirmed56 acclimatized adults, 90 min at 23, 26, 29°C: task speed best at 26°C; a fan partly mitigated the 29°C loss; comfort at 26 and 29°C with a fan equaled or beat 23°C while performance did not (A6, A7).
• S28REHVA Journal: direct impact of short-term exposure to pure carbon dioxide levels on cognitive performanceopened, the number was not on the pageResearcher summary: bioeffluent exposures reduce simulated office work above 3,000 ppm, effects observed 1,600 to 3,000 ppm, none below 1,600 except demanding tasks from 950 ppm; pure CO2 results inconsistent, no office-task effects until 20,000 ppm and higher; air quality rated higher at 1,500 than 600 ppm pure CO2 (A7).
• S1NIOSH Criteria for a Recommended Standard: Occupational Exposure to Heat and Hot Environments (2016)opened, quote confirmedHeat balance determinants: air temperature, humidity, air movement, radiation, clothing and metabolic rate interact; air temperature alone does not set the body's heat load (A1, A6, A7).

The page may say a room can feel fine and still cost thinking, with the heat channel as the measured case; it may not say the body 'knows' a good room, because the comfort vote demonstrably does not.

The comfort-versus-performance row: with thermal comfort held constant, objective performance still fell while self-rated performance did not, and air-quality ratings moved against performance in the CO2 chamber. Its source record carried a number the gate could not find on the cited page, so the percentages ride as a direction; the dissociation itself is the heat channel's entry in the same-law group with 1.4, 2.4 and 3.8 and the reason 'if it feels fine it is fine' is refuted.

Park, Goodman, Hurwitz and Smith 2020 (CORRELATIONAL, weather-shock quasi-experiment with student fixed effects; about 10 million PSAT retakers in the United States): hotter school days in the year before the test lowered scores, weekend and summer heat did not (a falsification test), and without air conditioning a 1°F hotter school year reduced that year's learning by about 1 percent, with the loss about three times larger for Black and Hispanic students than for White students, attributed by the authors to who has air conditioning. Cedeño Laurent 2018 (CORRELATIONAL, natural experiment; university students in Boston dormitories during a heat wave): non-air-conditioned residences averaged 26.3°C against 21.4°C in conditioned ones, and Stroop reaction times were about 13.4 percent slower and throughput about 9.9 percent lower in the hot dormitories; overnight exposure, sleep, ventilation and noise differed, so the figure is marked confounded. Boundaries: secondary and university students, not office workers; a standardized test score and a reaction-time task; exposure measured outdoors in S27 with the mechanism inferred (lost instructional time and lost sleep as much as daytime heat on the brain); no randomization in S26; neither transfers to an air-conditioned office at 24°C, and the roughly ten percent of S26 is not a per-degree law (A6).

Cumulative exposure: heat cost scales with concentration times time (dose integrates), and in real buildings a hot season acts through instructional time, sleep and the dwelling's cooling stock rather than through a daytime temperature coefficient, which is why the large numbers appear only in season-long or week-long exposures and fall on the unconditioned.

Granted tier: PROBABLE, set by the code gate on October 4, 2026 · Survived the adversary's attack · Sources: 2 cited, 2 opened and confirmed

• PROBATION: 1 misquote flag elsewhere in this vector (below the kill line of 2); this axiom's own sources were clean -> capped at PROBABLE
• survived a grounded cross-lab adversary (google vs authors ["anthropic"])
• citations: refetched 97% (>=90%), resolved 96% (>=80%) -> OK
• dead links: 1 of 29 source(s) would not open, a miss counted in the ratio above, never a misquote
• EARNED FORGED: survived a grounded cross-lab adversary AND primary sources confirmed
• PROBATION CAP: earned FORGED -> PROBABLE (one misquote elsewhere in the vector)

S27 (10 million students; 1 percent of a year's learning per 1°F without air conditioning; weekend and summer null; threefold racial gap), S26 (26.3 versus 21.4°C; 13.4 percent slower Stroop, 9.9 percent lower throughput; confounded). Observation: both field estimates. Inference: cumulative, mediated by time and cooling stock.

• S27Park, Goodman, Hurwitz and Smith 2020, American Economic Journal: Economic Policy, Heat and Learningopened, quote confirmedAbout 10 million PSAT retakers with student fixed effects: hotter school days in the prior year lower scores, weekend and summer heat do not; without air conditioning a 1°F hotter school year cuts that year's learning by about 1 percent, about three times larger for Black and Hispanic students (A8).
• S26Cedeño Laurent et al. 2018, PLOS Medicine: reduced cognitive function during a heat wave among residents of non-air-conditioned buildingsopened, quote confirmedStudents in non-air-conditioned dormitories (26.3°C) versus air-conditioned (21.4°C) during a heat wave: Stroop reaction time about 13.4 percent slower and throughput about 9.9 percent lower; overnight exposure, sleep, ventilation and noise confounded (A8).

The page may say heat over a season costs learning in buildings without cooling, with the population named; it may not carry the 13 percent as a daytime temperature effect.

The cumulative-exposure row: the large thermal numbers live in a hot school year without cooling and a dormitory heat wave, confounded or population-bound, not in an air-conditioned office. With 1.6 (mood effects in clinical doses over weeks) it marks where the big effects of a built environment actually live, in long exposure and selected populations, and it feeds the persistence meta-axiom as the channel whose cost accrues over a season.