Skip to content

Study 28 Wet Bulb Threshold Court

Jump to bottom
rg78803 edited this page Aug 27, 2026 · 2 revisions

Study 28 — Wet-bulb threshold court

Charter published 2026-08-27. Act 1 corpus ingested and the section-zero gate run the same day — results in the Act 1 section below; the law here is frozen as run. Act 2, the 2026 casualty-window census, stands gated on the 2026 archive.

The season's measured record and the instrument's own confession

In July 2026, Japan's Fire and Disaster Management Agency counted 41,174 people transported to hospital by ambulance for heatstroke in one month, 79 of them confirmed dead; the week of July 13–19 alone recorded 10,857 transports, the first weekly total above 10,000 since monitoring began May 1 (REPORTED — Japan Times 2026-08-25 and 2026-07-23, quoting FDMA figures). Across the strait, Yangsan, South Gyeongsang Province, reached 42.5 °C at 13:26 KST on 2026-08-02 — the highest temperature measured in Korea's 122 years of modern observation, begun 1904 — after five consecutive days above 40 °C, with more than 20 localities under the country's new top-tier emergency heatwave alert (REPORTED — Korea Times, Al Jazeera, Japan Times, all 2026-08-02, all quoting the Korea Meteorological Administration). Korea's disease agency counted 1,871 heat-related illnesses and 13 deaths between May 15 and July 31; by early August, about 2,220 and 19 (REPORTED — KDCA via AP wire, early August 2026).

Every operational verdict rendered during that record — every alert issued or withheld, every work-stop triggered or not — turned on a computed humid-heat number, and the instruments that compute it carry their own published confessions:

  • The National Weather Service's operational heat index is a regression. Its own technical attachment, SR 90-23 (Lans P. Rothfusz, 7/1/90), states verbatim: "No true equation for the Heat Index exists. Heat Index values are derived from a collection of equations that comprise a model." The document offers what it itself calls "an ersatz version of the Heat Index equation" and closes with: "Because this equation is obtained by multiple regression analysis, the heat index value (HI) has an error of ±1.3 F." (VERIFIED — https://www.weather.gov/media/ffc/ta_htindx.PDF fetched and read 2026-08-27; all three quotes verbatim.)
  • The most widely used single-expression wet-bulb formula, Stull's (J. Appl. Meteor. Climatol. 50, 2267–2269, 2011), is an empirical fit valid at sea-level pressure only, for RH 5–99% and T −20…+50 °C, mean absolute error ~0.28 °C, error range −1 °C to +0.65 °C (REPORTED — band and domain consistently stated by the NCAR NCL wetbulb_stull documentation, PMC9941479, and multiple independent calculators; the AMS journal page 403-blocked live fetch this session, so the band is carried as reported wording of a real published figure, not a verbatim quote). The formula itself is VERIFIED from a Stull-authored source — Practical Meteorology, Ch. 4, p. 95, eq. (4.19), read directly.
  • The survivability literature puts the stakes in the same unit. Sherwood & Huber (PNAS 107:9552–9555, 2010), verbatim: "Any exceedence of 35 °C for extended periods should induce hyperthermia in humans and other mammals, as dissipation of metabolic heat becomes impossible." (VERIFIED — PMC2906879 fetched 2026-08-27.) The chamber work moved the line lower: Vecellio et al. (J. Appl. Physiol. 132:340–345, 2022), 24 young healthy adults at everyday-activity workloads, verbatim: "Mean Twb,crit values were relatively constant across 36 °C–40 °C humid environments and averaged 30.55 ± 0.98 °C" and "all means were significantly lower than the theoretical 35 °C threshold." (VERIFIED — PMC8799385 fetched 2026-08-27.)

The thesis is one sentence: at a survivability threshold, the published error band of the operational instrument is verdict-sized — a formula licensed to be wrong by up to 1.0 °C, graded at a line where 0.98 °C is the empirical standard deviation of the human limit itself, will render EXCEEDS where exact arithmetic renders BELOW, and the reverse, on real archived station-hours. This court computes the thermodynamic wet-bulb temperature T_W as an exact rational — bounded root-bracketing on the psychrometric identity, from the integer-tenths (T, T_d, P) triples NOAA's global hourly archive already serves in native integer form — and seals every station-hour where the exact key and each named float instrument render opposite verdicts at a frozen threshold, with station id and timestamp. The verdict is about the instrument, never about physiology: a presented station-hour is verified; a body's response is not modelled.

Section zero — the discrimination gate

Before any station-hour is graded, the instrument must be shown able to fail. A flip census that fires on every hour is a turn counter, not a court, and the program has retired an instrument for exactly that defect before. The gate is the study's three loss conditions, each graded as a loss and published as such:

  • Loss (i) — the flip set is empty at mortality-relevant thresholds. The float band is real but harmless: no archived station-hour sits close enough to a survivability line for a licensed error to change the verdict. If the sealed flip set at the frozen thresholds is empty over the frozen corpus, the study publishes LOSS with the zero on the page.
  • Loss (ii) — no flip-hour falls inside a REPORTED 2026 humid-heat casualty window. Flips that never coincide with the season people were actually carried to hospitals have no practical stake, and the page says so. The named windows are frozen below. This criterion is UNGRADEABLE today and the page says that too: the NOAA archive's 2026 directory does not yet exist (measured three independent ways, next section). The criterion is frozen now and graded only when access/2026/ appears; substituting 2025 hours into a 2026-window criterion is forbidden by this paragraph.
  • Loss (iii) — the flip census fires everywhere. A census that flips on hours far from every threshold distinguishes nothing — it measures its own noise, not the float band. The guard-band and agreement rules below make this decidable in integers.

The controls, frozen here with real archive rows:

  • Always-green control (scored first). A winter station-hour, read live 2026-08-27 (VERIFIED, row quoted from the served CSV): Ulsan, KS, station 47152099999, 2025-01-01T00:00:00, TMP +0012,1, DEW -0087,1, SLP 10233,1 (T = +1.2 °C, T_d = −8.7 °C, P = 1023.3 hPa). It sits more than 25 °C below the nearest frozen threshold; the exact key and every float adversary MUST render the identical verdict (BELOW) at every threshold. A flip here is not evidence; it is a broken scorer — the study halts on an instrument defect and no physics is scored. A hot, in-domain control is frozen with it: Abu Dhabi Intl, station 41217099999, 2025-07-15T12:00:00, TMP +0397,1, DEW +0199,1, SLP 09941,1 (VERIFIED, row quoted), whose exact T_W lies several degrees below 30.55 °C; all arms must agree BELOW there as well. The second admitting winter control, pinned 2026-08-27 under S6: Pohang, KS, station 47138099999, 2025-01-01T00:00:00, TMP +0015,1, DEW -0078,1, SLP 10238,1, ELEVATION 3.58 m (VERIFIED, row quoted from the served CSV) — coastal, synoptic, SLP-carrying, inside the elevation bound.
  • The absence control, frozen from a measured feed. Gimhae Intl, station 47153099999, 2025-01-01T00:00:00, TMP +0000,1, DEW -0120,1, SLP 99999,9 — the missing sentinel (VERIFIED, row quoted from the served CSV 2026-08-27). This is not one missing hour: the served Gimhae feed is METAR-class (FM-15) and was measured to carry the SLP sentinel on 372 consecutive January 2025 hours, every one, with the altimeter setting present in MA1 instead. Measured at year scale when the corpus ran: 6,397 of Gimhae's 6,412 rows print ELEMENT_MISSING — and 15 rows do carry SLP and admit, so the earlier drafting claim that no Gimhae hour can admit was too strong and is corrected here by the census itself. The control row stands: the exact arm must print ELEMENT_MISSING on it — never a verdict, never a substituted standard atmosphere, never a silent drop. A station that answers with absence is a different answer from a station that answers BELOW, and the gate proves the court keeps them distinct on a real feed.
  • No clean always-red control exists, and none is minted. A guaranteed flip would require knowing, before the corpus, a station-hour whose exact T_W sits inside a specific adversary's error at a specific threshold — the corpus's job to find, not the charter's to fabricate. In its place stands a domain-refusal arm: the Stull adversary evaluated outside its published validity domain (RH < 5%, or T outside −20…+50 °C) must print DOMAIN_REFUSED, never a number. Refusal and absence are different answers, and both are different from a verdict.
  • The guard band, frozen: G = 15 tenths (1.5 °C). A sealed flip may only occur on an hour whose exact T_W lies within G of the threshold being graded — G covers the widest published adversary band (−1.0/+0.65 °C) with margin. A flip sealed at distance > G is an S1-class instrument defect: halt, publish, fix, restart the seal.
  • The agreement floor, frozen: on every admitted hour whose exact T_W lies farther than G from every frozen threshold, exact and float verdicts must agree — 0 disagreements permitted. This is loss (iii)'s decidable form: a census that cannot hold 100% agreement far from the lines is firing everywhere and is retired on the page.
  • The flip ceiling, frozen: among hours within G of a given threshold, the flip fraction per adversary must be ≤ 1/2 as an exact ratio of integers. A near-threshold census that flips more often than a coin is not reading the float band; it is broken, and the study is void at that threshold.

Every integer in this gate — G = 15 tenths, 0 disagreements, the 1/2 ceiling, the four control rows with their station ids and timestamps — is frozen in this charter before any corpus byte is scored. None of it moves after scoring. A FAIL is published as a FAIL.

The two readings and where they disagree

Reading What it holds constant The observation that separates it Sourcing
Float regression — humid heat is a fitted surface; the verdict is the surface's value in binary64 The magnitude: tracks T_W to roughly ±1 °C everywhere, which is why it served for decades Stull 2011 takes (T, RH%) at fixed sea-level pressure 1013.25 hPa — it never sees the station's own pressure, and RH must itself be derived from (T, T_d), so the float chain is two formulas deep before any verdict. Rothfusz is a regression over a table from a model with frozen inputs (5-kt wind, 84% clothing cover, 1.6 kPa ambient vapor pressure, 98.6 °F core — VERIFIED, enumerated in SR 90-23), self-described "ersatz," licensed ±1.3 °F. ASHRAE-class solvers iterate the true identity, but in floats, to a float tolerance, on whichever libm the host ships. VERIFIED (Stull eq.; Rothfusz quotes); REPORTED (Stull band; ASHRAE basis in Hyland & Wexler 1983)
Exact rational key — humid heat is the root of a thermodynamic identity; the verdict is which side of a frozen rational the root lies on The shape: the crossing verdict at the boundary, decided identically on every machine The identity with frozen rational coefficients (below) has one root T_W in [T_d, T]; bounded rational root-bracketing decides EXCEEDS/BELOW at every frozen threshold with no float constructed anywhere, or prints ON_BOUNDARY as its own answer. No libm, no rounding mode, no platform enters. Charter construction; coefficients VERIFIED/REPORTED per row below
The null — the flip census is an artifact of the exact pipeline's own choices (Taylor-bound depth, guard band, quantisation), not of the float band Nothing; it predicts flips scattered without regard to the adversaries' published bands This is what Section zero exists to test, and it is not a straw man: every crossing below is a choice that could, if botched, manufacture flips. The null is retired only by the gate — agreement floor, guard band, ceiling, and the S2 replication arm requiring the measured Stull deviation to land inside Stull's own published band. Never by citation. Design statement of this charter

The adversaries are separated on shape, never magnitude. Stull is not the adversary because its numbers are the wrong size — over most of the chart they are nearly the right size. It is the adversary because it has the wrong shape: a pressure-blind fitted surface standing in for a pressure-dependent thermodynamic root. Rothfusz's shape is a regression over a model table with frozen physiology. The ASHRAE class is the sharpest separation of all: it iterates the same identity this court roots on, so the only thing distinguishing it from the exact key is the arithmetic substrate — precisely the object under trial.

The identity, and the coefficients frozen as exact rationals. Saturation vapor pressure over water is pinned to the Alduchov–Eskridge relative-error-minimizing Magnus form (named AEKR/AERK in the primary report):

e_s(t) = (30547/5000) · exp( (141/8) · t / ( (6076/25) + t ) ) hPa, t in °C

— that is 6.1094, 17.625, 243.04, each a finite decimal and hence an exact rational (VERIFIED — the NOAA/DOE primary report, Alduchov & Eskridge, DOE/ER/61011-T6, OSTI 548871, fetched 2026-08-27: Table II gives the form, Table III its maximum relative error 0.384% over −40…+50 °C; journal citation J. Appl. Meteor. 35, 601–609, 1996; the 17.625/243.04 pair independently REPORTED as the minimax choice across secondary dew-point references — the cross-check source read this session is a tertiary reference page, so the tag is REPORTED, while the coefficients themselves stand VERIFIED from the primary report). The moist-air enhancement factor is pinned from the same report, eq. (20): f_w(p) = (100071/100000) · exp( (9/2000000) · p ), p in hPa, maximum relative error 0.0773% (VERIFIED, same fetch). The psychrometer equation closes the identity: actual vapor pressure e = f_w(P) · e_s(T_d), and T_W is the root of

F(T_W) = f_w(P) · e_s(T_W) − e − (653/1000000) · (1 + (118/125000) · T_W) · P · (T − T_W)

with 6.53×10⁻⁴ and 0.000944 the Assmann-psychrometer coefficients of the WMO Guide to Instruments and Methods of Observation — VERIFIED and frozen 2026-08-27: WMO-No. 8, Volume I, 2024 edition, Annex 4.B, formula (4.B.8), read verbatim from the fetched PDF: e′ = e′w(p,tw) − 6.53 · 10⁻⁴ · (1 + 0.000944 tw) · p · (t − tw) (printed page 197; document sha256 4f020a5970bb3fca44f61df3ae63f6c02a74b6b60cfd6951b49aa520b1c94187; library.wmo.int sits behind a human-verification interstitial for automated clients, so the byte-identical official PDF was retrieved from the Internet Archive capture of the library's own download URL — the gate evidence is kept with the corpus). The Guide's own Section 4.3.2.5 records that screen psychrometers use coefficients from 6.50 to 6.78×10⁻⁴ K⁻¹ — the plurality this court referees, in the instrument's own reference. S6's re-read requirement is satisfied; the pair above is the frozen law. The over-water saturation form is the only form this court evaluates: every frozen threshold sits far above freezing, the admitted-domain exponent arguments are bounded, and the sign-correct interval rule of Crossing B covers dew points below 0 °C (where winter control hours live) without any over-ice coefficient entering a verdict.

The plurality this court referees is itself measured: a live-fetched comparison paper exhibits five distinct Magnus-family coefficient sets in simultaneous common use for the same saturation curve — August-Roche-Magnus, Bolton, Tetens, Buck 1981, Wexler — and states "dozens of formulas have appeared in the literature" (VERIFIED — tc.copernicus.org preprint tc-2023-8, pages 3–5 read directly, 2026-08-27). The full duplication census is in the protection section below.

The response archive

Everything the court reads is public, anonymous, and integer-native on the wire. All entries verified live on 2026-08-27 unless marked REPORTED.

Source URL Format / keys Version / cadence Auth Sourcing
NOAA ISD global hourly https://www.ncei.noaa.gov/data/global-hourly/access/<year>/<station>.csv — station key is the 11-digit USAF+WBAN id CSV; header "STATION","DATE","SOURCE","LATITUDE","LONGITUDE","ELEVATION","NAME","REPORT_TYPE","CALL_SIGN","QUALITY_CONTROL","WND","CIG","VIS","TMP","DEW","SLP",…; TMP/DEW signed scaled integer tenths of °C, one-character QC code after the comma; SLP scaled integer tenths of hPa Yearly directories; newest is 2025/ (dir mtime 2025-10-02); sampled 2025 files end 2025-08-24T21:00:00 None — anonymous GET VERIFIED — listing plus Ulsan/Gimhae/Abu Dhabi files fetched, rows quoted 2026-08-27
ISD format document https://www.ncei.noaa.gov/data/global-hourly/doc/isd-format-document.pdf Air temperature: "SCALING FACTOR: 10, +9999 = Missing" (MIN −0932, MAX +0618 °C); dew point: same scaling (MIN −0982, MAX +0368); sea-level pressure: "UNITS: Hectopascals, SCALING FACTOR: 10, 99999 = Missing"; QC codes 0–9 plus letters A/C/I/M/P/R/U for TMP/DEW, digits only for SLP Static document None VERIFIED — PDF read page-by-page; wording quoted from pp. 10–12
AWS open-data mirror noaa-global-hourly-pds.s3.amazonaws.com Same files under <year>/<station>.csv; key 2025/47152099999.csv is 1,906,289 bytes, LastModified 2025-10-04T13:28:53Z Mirrors NCEI None VERIFIED — S3 ListObjectsV2 exercised live

The 2026 directory does not exist, measured three independent ways with a healthy-population control (all VERIFIED 2026-08-27): access/2026/ returns HTTP 404; a direct probe of access/2026/47152099999.csv returns 404; the S3 mirror returns KeyCount 0 for prefix 2026/ while the control arm — prefix 2025/47152 — returns exactly the Ulsan file named above. The archive's live edge is 2025-08-24T21:00Z. Consequence, stated as a standing gate rather than worked around: no 2026 station-hour is ingestable today, loss condition (ii) cannot be graded in either direction, and the corpus plan is two acts — Act 1 (calibration and gate) runs on the 2025 summer, complete through August 24 at the same endpoints; Act 2 (the casualty-window census) begins the day access/2026/ appears, with the law already frozen and unmoved.

The QC admission rule, frozen: an hour is admitted only if the TMP, DEW, and SLP quality characters are each in {0, 1, 4, 5, 9} — the "passed" family of the ISD grammar; codes 2, 3, 6, 7 and every letter code are excluded, and the missing sentinels +9999/99999 exclude the hour outright. A QC exclusion is recorded QC_EXCLUDED with its characters; a missing element is recorded ELEMENT_MISSING. Neither is a zero, neither is a verdict, and both denominators are published in the seal. The court never substitutes a standard atmosphere for a missing SLP — pressure-blindness is the named defect of one of its own adversaries. The station-elevation admission bound, frozen: |elevation| ≤ 50 m. The exact arm's P is ISD's SLP — sea-level-reduced pressure — which equals the station's own air pressure only at sea level; at an elevated site it is a convention, not the air the psychrometric identity describes. Admitting only stations whose published ISD ELEVATION field lies within 50 m of sea level makes SLP-for-station-pressure a bounded frozen approximation, stated as such in Honest limits, rather than an unbounded one. All three frozen control stations satisfy the bound: Abu Dhabi Intl 27 m, Ulsan 34 m, Gimhae Intl 2 m.

The named station anchors (ids VERIFIED as served files): 41217099999 (Abu Dhabi Intl — Gulf block), 47152099999 (Ulsan) and 47153099999 (Gimhae) in the Korean 4715x block — Gimhae with its measured role stated: its METAR-class feed carries the altimeter in MA1 and the SLP sentinel throughout (372 consecutive January 2025 hours measured), so it anchors the absence census, not the verdict census. The full corpus station list — the Gulf shoreline stations of the humid-heat literature (Ras al-Khaimah, Jeddah, Dhahran, Jacobabad — REPORTED, Raymond et al. 2020) plus the Korean and Japanese stations covering the 2026 windows — is pinned by exact 11-digit id at corpus stage, before the gate runs, under the S6 discipline. A station whose file is absent for a named year is recorded FILE_ABSENT, never silently dropped.

The frozen law (to be sealed at corpus stage)

Drafted here; frozen verbatim, with every integer bound to a named file and station id, before the first station-hour is scored.

The declared crossings — two, each at its own site, each performed once.

  • Crossing A — wire integers → exact rationals. TMP, DEW, SLP arrive as signed scaled integers (tenths) and are read once, at load, as exact rationals with denominator 10. Integer-to-rational: no float parse exists anywhere in the exact arm, because the archive is already integer-native (VERIFIED above).
  • Crossing B — the transcendental, bounded, with the interval ends assigned by sign. exp is the only non-rational operation in the identity. It is evaluated as a rational interval whose end-assignment is frozen sign-correct, because a truncated Taylor sum bounds exp from different sides on the two half-lines. For an argument x ≥ 0: the degree-32 truncated Taylor sum in exact rationals is the LOWER bound, and that sum plus the frozen geometric tail bound is the upper. For x < 0 — where admitted winter hours live: the Ulsan control's e_s(T_d) argument is −0.654 — the degree-32 even-ending truncated sum is an UPPER bound of exp(x), because the first omitted term x³³/33! is negative, and the degree-33 truncated sum is the LOWER bound. The interval assignment swaps ends by the sign of the argument and by nothing else; both cases are valid because every exponent argument on the admitted domain is bounded by 4 in absolute value. Everything downstream of exp is a rational interval; a comparison is decided only when the whole interval sits on one side.

The exact-key algorithm, frozen. For an admitted hour (T, T_d, P), three cases by the sign of T − T_d, each with its own frozen answer. T > T_d: compute e = f_w(P)·e_s(T_d) as a rational interval; bracket the root of F on [T_d, T] (F is strictly increasing in T_W on the admitted domain, and F(T_d) < 0 < F(T) holds strictly here, and only here); bisect with exact rational midpoints. T = T_d — saturation: fog, RH 100%, exactly the humid-heat hours this court cares about — gives F(T_d) = F(T) = 0 and a degenerate bracket; the frozen answer is T_W = T exactly, a rational, with no bisection performed. T < T_d — a sensor error the raw archive really contains — inverts the bracket; the hour prints its own exclusion token INVERTED_DEWPOINT, which is neither absence, nor refusal, nor a verdict, and is published in its own census. Termination, in integers: stop when the bracket width is ≤ 1/2000 °C AND every frozen threshold lies strictly outside the bracket — the verdict at every threshold is then a rational comparison. If a threshold remains inside the bracket, refine to width 1/1000000 °C; if it still straddles, print ON_BOUNDARY for that (hour, threshold) — its own token, its own census, never folded into EXCEEDS or BELOW. If Crossing B's interval is too wide to decide a bisection sign, the Taylor degree doubles once, to the frozen ceiling 64; an hour undecidable there prints PRECISION_EXHAUSTED and is published as such. Same input, same bracket, every machine.

The float adversaries, frozen as named arms. Each is evaluated in IEEE-754 binary64, round-to-nearest-even, with the platform libm's atan/exp, and the libm identity string is recorded in every sealed row — a court whose adversary is float arithmetic records which float arithmetic:

  • Arm STULL: T_w = T·atan(0.151977·(RH% + 8.313659)^{1/2}) − 4.686035 + atan(T + RH%) − atan(RH% − 1.676331) + 0.00391838·(RH%)^{3/2}·atan(0.023101·RH%), radians (VERIFIED — Practical Meteorology p. 95, eq. 4.19; coefficients as exact decimal strings: 151977/1000000, 8313659/1000000, 4686035/1000000, 1676331/1000000, 391838/100000000, 23101/1000000). RH% is derived in float from the same Magnus form, RH = 100·e_s(T_d)/e_s(T) — the two-formula chain is the adversary as deployed. Outside RH 5–99% or T −20…+50 °C: DOMAIN_REFUSED.
  • Arm ITER (ASHRAE-class): the identical identity F(T_W) = 0, iterated in binary64 (bisection to float tolerance 1×10⁻⁶), standing for the iterative psychrometric solver class whose basis is Hyland & Wexler 1983 (REPORTED — ASHRAE Handbook—Fundamentals Ch. 1). Graded on the court's own pinned coefficients, so the only difference from the exact key is the arithmetic substrate.
  • Arm ROTHFUSZ (deferred act, confession standing): the SR 90-23 regression outputs a heat index in °F, not a T_W, so it cannot flip at a T_W threshold and is not in the v1 census. It is chartered for a heat-index act whose thresholds (the NWS operational category cuts) are frozen at corpus stage before any scoring, under the same discipline. Its confession — "No true equation … exists," error ±1.3 °F — stands either way (VERIFIED above).

The thresholds, frozen as exact rationals:

Threshold Exact rational Role Sourcing
T_W = 35.0 °C 350/10 Theoretical survivability ceiling (Sherwood & Huber 2010) VERIFIED (quotes above)
T_W = 30.55 °C 611/20 Empirical young-healthy humid-regime limit — Vecellio et al. 2022's mean, carried with the paper's own caveat that it is humid-regime-specific ("a wet-bulb temperature threshold cannot be applied to human adaptability across all climatic conditions" — VERIFIED) VERIFIED (quotes above)
WBGT = 32.1 °C 321/10 Qatar's statutory stop-work trigger, Ministerial Decision No. 17 of 2021, restated in force 2026-06-02 REPORTED (ILO; The Peninsula Qatar 2026-06-02; ≥4 sources)
WBGT_ref = 56.7 − 11.5·log₁₀M / 59.9 − 14.1·log₁₀M °C 567/10, 23/2, 599/10, 141/10 ISO 7243:2017 reference limits, acclimatized/unacclimatized, 115 < M < 520 W — the 2017 formulas, not the superseded 1989 class table VERIFIED (ISO 7243:2017(E) Annex A read directly)
WBGT = 90 °F 290/9 °C US military black-flag suspension REPORTED (DoD HPRC)

The WBGT rows are scoped, not smuggled: WBGT is natural wet-bulb + globe + dry-bulb — a different observable from thermodynamic T_W, and ISO 7243 says so itself, Annex D, verbatim: "The natural wet bulb temperature is thus different from the thermo-dynamic temperature determined with a psychrometer" (VERIFIED). The v1 flip census runs at the two T_W thresholds only. The WBGT thresholds enter a census only if and when a published WBGT-estimation formula from (T, T_d, P)-class inputs is itself named as a float adversary and frozen — the flip is then between exact and float computations of the same published estimator, with Qatar's 321/10 as the statutory stake. No silent T_W-for-WBGT substitution occurs anywhere in this law.

The flip census, defined. For each admitted hour h, adversary a ∈ {STULL, ITER}, threshold θ ∈ {350/10, 611/20}: a flip is sealed iff the exact verdict is EXCEEDS and a's is BELOW, or the reverse — ON_BOUNDARY, DOMAIN_REFUSED, QC_EXCLUDED, ELEMENT_MISSING, PRECISION_EXHAUSTED, INVERTED_DEWPOINT rows are their own censuses, never flips or agreements. A sealed flip row carries: station id, ISO-8601 timestamp, the three wire integers with QC characters, the exact rational bracket, a's binary64 output as a printed decimal string, θ, a, and the libm identity.

Success criteria, each an integer decision:

  • S1 — Always-green and absence controls. The three admitting control hours (Ulsan and Pohang winter, Abu Dhabi hot): identical verdicts from the exact key and every arm at every θ; any disagreement halts the study on an instrument defect. The Gimhae absence control must print ELEMENT_MISSING and never a verdict — a scorer that renders any verdict on it is defective and halts the study the same way. DOMAIN_REFUSED must fire on a constructed out-of-domain probe (T = −25 °C) and never on the admitting controls.
  • S2 — Replication band. Over all admitted, in-domain hours of the frozen calibration slice (the 2025 files of the pinned station list), at least 99 of every 100 must print q(T_w,STULL) − q(T_w,exact) within [−10, +7] tenths, q being floor-to-integer-tenths on the printed value and on the bracket midpoint — Stull's own −1.0/+0.65 °C adopted as a frozen integer window (REPORTED band, declared as such). If the measured deviation does not live inside the instrument's own confession, the pipeline is defective and every downstream criterion is void.
  • S3 — Existence (retires loss (i) or seals it). |FlipSet(θ = 611/20)| ≥ 1 over the frozen Gulf-block stations' admitted summer hours. The Gulf's top-0.1% wet-bulb climatology (REPORTED — Raymond et al., Science Advances 6:eaaw1838, 2020: station data show "some coastal subtropical locations have already reported a TW of 35 °C", and the top 0.1% of Gulf shoreline readings exceed 31 °C) is why near-threshold hours are expected at 611/20. No flip requirement is placed at 350/10 — hour-scale 35 °C crossings are rare on the record, and a criterion demanding them would be gamed by its own denominator. An empty flip set at 611/20 prints LOSS (i) with the census totals.
  • S4 — Casualty-window intersection (retires loss (ii) or seals it). At least 1 sealed flip-hour inside a frozen REPORTED 2026 window: Japan 2026-07-13…19, Japan 2026-07-01…31, Korea 2026-07-12, Korea 2026-07-31…2026-08-02 (window sources dated above). Standing gate: UNGRADEABLE until NCEI publishes access/2026/; graded then, with this law unmoved. An empty intersection when graded prints LOSS (ii).
  • S5 — Discrimination (retires loss (iii) or seals it). The Section-zero integers as law, with one consequence per event: a breach of the agreement floor — any exact-versus-float disagreement on an hour farther than G = 15 tenths from every θ — or of the flip ceiling — flip fraction within G at a θ exceeding 1/2 per arm — prints LOSS (iii): the census fires everywhere, it is a turn counter, and it is retired on the page. A sealed flip at distance > G from its θ is not a loss and never prints one: it is the Section-zero S1-class instrument defect — a beyond-G flip indicts the scorer, not the float band — and its frozen consequence is halt, publish, fix, restart the seal.
  • S6 — Byte identity. Every scored file pinned by URL, byte count, and the S3 LastModified where the mirror serves it (2025/47152099999.csv = 1,906,289 bytes, 2025-10-04T13:28:53Z — VERIFIED); the QC admission set and coefficient rationals frozen; the WMO psychrometer pair re-read against WMO-No. 8 Annex 4.B and frozen verbatim (DONE 2026-08-27 — formula 4.B.8 quoted above, document sha256 pinned); the second admitting winter control pinned (DONE 2026-08-27 — Pohang row quoted above); the station list and the ROTHFUSZ-act thresholds pinned before the gate runs. A stranger with curl reproduces the corpus byte-exact or the seal does not stand.

All three losses are graded as losses and published as losses. There is no outcome in which this page quietly becomes about something else.

Act 1 — the gate, run 2026-08-27

The engine is one Swift source file, courts/study-28-wetbulb/study28-court.swift in the uum8dSolarResearch repository, compiled with swiftc -O and nothing else. The exact arm computes every verdict in big-integer rational arithmetic — addition, subtraction, multiplication, and cross-multiplied comparison only; no division, no GCD, no float anywhere in a verdict — with the sign-correct truncated-Taylor exp intervals of Crossing B. The float arms run in IEEE-754 binary64 with the platform libm exactly as frozen; libm identity recorded in every run: Apple libSystem, macOS Version 27.0 (Build 26A5416b). The full run log is courts/study-28-wetbulb/court-run-2.log.

The corpus, pinned byte-exact (S3 mirror noaa-global-hourly-pds.s3.amazonaws.com/2025/<station>.csv, all fetched and HEAD-receipted 2026-08-27):

Station Name Bytes S3 Last-Modified
47152099999 Ulsan, KS 1,906,289 Sat, 04 Oct 2025 13:28:53 GMT
47153099999 Gimhae Intl, KS 2,224,690 Sat, 04 Oct 2025 13:33:40 GMT
41217099999 Abu Dhabi Intl, AE 2,602,046 Sat, 04 Oct 2025 13:32:04 GMT
41715099999 Jacobabad, PK 673,693 Sat, 04 Oct 2025 13:30:52 GMT
41184099999 Ras Al Khaimah Intl, AE 2,330,979 Sat, 04 Oct 2025 13:26:43 GMT
41024099999 Jeddah King Abdulaziz Intl, SA 2,466,290 Sat, 04 Oct 2025 13:29:12 GMT
41150099999 Dhahran, SA 2,518,015 Sat, 04 Oct 2025 13:33:53 GMT
47138099999 Pohang, KS 1,943,227 Sat, 04 Oct 2025 13:33:02 GMT
47165099999 Mokpo, KS 1,968,665 Sat, 04 Oct 2025 13:30:12 GMT

Jacobabad is excluded by the court's own law and published as such: its served ELEVATION is 56.38 m, outside the frozen |elevation| ≤ 50 m bound — STATION_ELEV_EXCLUDED, zero of its hours scored. The station most famous in the humid-heat literature is the first casualty of the SLP-honesty rule, and that is the rule working.

S1 — GREEN. All three admitting controls render identical verdicts on every arm at both thresholds — Ulsan winter BELOW/BELOW/BELOW, Pohang winter BELOW/BELOW/BELOW, Abu Dhabi hot BELOW/BELOW/BELOW at both 611/20 and 350/10. The constructed out-of-domain probe (T = −25 °C) prints DOMAIN_REFUSED from the STULL arm and it never fires on a control. The Gimhae absence-control row prints ELEMENT_MISSING before any arithmetic, and its year census is 6,397 ELEMENT_MISSING of 6,412 rows.

S2 — GREEN. Over every admitted, in-domain hour of the calibration slice: 23,620 of 23,741 STULL deviations land inside the frozen [−10, +7]-tenths window — as an exact integer inequality, 23,620 × 100 = 2,362,000 ≥ 99 × 23,741 = 2,350,359. The measured deviation of the deployed formula lives inside the formula's own published confession, so the pipeline is licensed to grade it.

S5 — GREEN, and the census is not a turn counter. Agreement-floor breaches — any exact-versus-float disagreement farther than G = 15 tenths from every threshold — zero, out of tens of thousands of far-from-threshold verdicts. Near-threshold flip fractions per station at 611/20, each an exact ratio ≤ 1/2: Abu Dhabi 4/43, Ras Al Khaimah 3/24, Dhahran 10/124, Jeddah 0/35. The instrument can fail, was given every chance to fire everywhere, and did not.

S3 — SEALED. Loss (i) is retired: the flip set at the survivability line is not empty. Seventeen station-hours, all sealed:

Station Hour (UTC) TMP DEW SLP exact T_W (tenths) exact STULL
41217099999 Abu Dhabi 2025-06-17T09 +406 +271 9951 301 BELOW EXCEEDS
41217099999 Abu Dhabi 2025-07-20T09 +390 +277 9936 302 BELOW EXCEEDS
41217099999 Abu Dhabi 2025-07-25T12 +398 +277 9908 303 BELOW EXCEEDS
41217099999 Abu Dhabi 2025-08-03T12 +393 +279 9960 304 BELOW EXCEEDS
41184099999 Ras Al Khaimah 2025-06-27T09 +402 +273 9953 302 BELOW EXCEEDS
41184099999 Ras Al Khaimah 2025-07-05T09 +380 +284 9955 305 BELOW EXCEEDS
41184099999 Ras Al Khaimah 2025-07-05T12 +371 +284 9946 303 BELOW EXCEEDS
41150099999 Dhahran 2025-08-05T12 +363 +289 9987 305 BELOW EXCEEDS
41150099999 Dhahran 2025-08-07T09 +385 +278 9995 301 BELOW EXCEEDS
41150099999 Dhahran 2025-08-10T09 +379 +282 10016 303 BELOW EXCEEDS
41150099999 Dhahran 2025-08-10T12 +377 +283 10003 303 BELOW EXCEEDS
41150099999 Dhahran 2025-08-11T21 +350 +293 10007 305 BELOW EXCEEDS
41150099999 Dhahran 2025-08-15T15 +358 +291 9956 305 BELOW EXCEEDS
41150099999 Dhahran 2025-08-17T12 +374 +286 9969 305 BELOW EXCEEDS
41150099999 Dhahran 2025-08-17T15 +354 +292 9971 305 BELOW EXCEEDS
41150099999 Dhahran 2025-08-21T09 +368 +285 10003 303 BELOW EXCEEDS
41150099999 Dhahran 2025-08-22T06 +361 +289 10018 304 BELOW EXCEEDS

Every flip is one-sided — the fitted surface prints EXCEEDS where the exact key prints BELOW, on hours whose exact wet-bulb sits 0.5 to 5.4 tenths under the 30.55 °C line — which is exactly the shape of Stull's published +0.65 °C upper error. There are no reverse flips on this slice. And the sharpest measured separation is the one the charter predicted: the ITER arm — the same identity, iterated in binary64 to 10⁻⁶ — never flipped once. At hour-verdict granularity on this slice, the float defect lives entirely in the regression's shape, not in the last bits of an iteration on the true identity. That is a statement about instruments, measured, not asserted.

The full census, with every denominator published (per station: admitted / QC_EXCLUDED / ELEMENT_MISSING / INVERTED_DEWPOINT / saturated T=T_d / ON_BOUNDARY / PRECISION_EXHAUSTED): Ulsan 5,558/19/3/0/0/0/0 · Gimhae 15/0/6,397/0/1/0/0 · Abu Dhabi 1,834/15/5,802/0/0/0/0 · Ras Al Khaimah 1,711/31/5,456/0/0/0/0 · Jeddah 1,823/28/5,632/0/0/0/0 · Dhahran 1,833/11/6,290/0/0/0/0 · Pohang 5,567/9/2/0/0/0/0 · Mokpo 5,570/7/1/0/10/0/0. The large Gulf ELEMENT_MISSING counts are the METAR-mix feeds carrying altimeter instead of SLP on their sub-hourly rows — absence published as absence, never substituted. No hour anywhere exhausted precision; no hour printed ON_BOUNDARY.

S4 — STANDING, ungradeable, unmoved. The 2026 archive still does not exist (re-measured this session); the casualty-window criterion waits for access/2026/ with this law frozen. Nothing from 2025 was substituted into it.

What this protects — people, ecology, and the planet

Workers under midday-ban regimes. Every Gulf state runs a summer outdoor-work ban, and all but one are calendar-and-clock rules: UAE June 15–Sept 15, 12:30–15:00; Saudi Arabia June 15–Sept 15; Kuwait June 1–Aug 31, 11:00–16:00; Bahrain June 15–Aug 31 (REPORTED — MoHRE via Gulf News; Morgan Lewis GCC survey 2026-06). Qatar alone adds a weather-triggered rule: under Ministerial Decision No. 17 of 2021, all work stops whenever worksite WBGT exceeds 321/10 °C, any hour — restated in force 2026-06-02 (REPORTED, ≥4 independent sources). Human Rights Watch, from interviews with 20 migrant workers February–May 2026, reports the clock-based bans "fall short" and the region's one WBGT trigger high and unevenly enforced (VERIFIED — hrw.org, 2026-06-23, fetched live); the finding that extreme Gulf humid heat now peaks in the evening and persists overnight (REPORTED — Communications Earth & Environment 5, s43247-024-01763-3) undercuts fixed midday hours structurally. A threshold statute is only as good as the number computed at it; a court that seals the hours where the deployed formula and exact arithmetic sit on opposite sides of the line is the accounting layer under every such rule.

Heat-warning integrity. The 2026 season produced, in one region, both failure modes a threshold-gated warning system can have. Korea overhauled its alert system in June 2026 — the new top tier fires on a one-day forecast of apparent temperature ≥ 38 °C or actual ≥ 39 °C, replacing a two-consecutive-day rule — and issued it for the first time on 2026-07-12 (REPORTED — Japan Times 2026-07-12; Seoul Economic Daily). Japan's top-tier Special Heatstroke Alert, gated on forecast WBGT ≥ 35 at every target point in a prefecture, had never been issued as of its April 2026 revision, and no source found says it fired during July 2026 — a month with 41,174 transports and a record 43 prefectures under the ordinary WBGT ≥ 33 alert on July 20 (REPORTED — Japan Times 2026-04-20; TechTimes 2026-07-20; the non-activation is absence of evidence across targeted searches, stated as exactly that). Whether a tier fires is a comparison between a computed index and a frozen number. This census has the same comparison shape — a computed index against a frozen line — with the arithmetic made exact and public; it is not the same comparison, because Japan's tiers fire on WBGT and Korea's on apparent temperature, and the scoping paragraph above keeps that distinction load-bearing.

The habitability record. Raymond et al. 2020 found T_W 35 °C exceedances at least 14 times since 1979 in station data, concentrated at Jacobabad, Ras al-Khaimah, Jeddah, and Dhahran, and a doubling of extreme humid-heat frequency since 1979 (REPORTED — Science Advances 6:eaaw1838; the study used HadISD, a QC'd derivative of the same NOAA ISD archive this court reads). Every such record-keeping claim passes through a wet-bulb computation; a record whose crossings depend on which of five coefficient sets and 21 index algorithms the analyst picked has a soft edge. An exact rational key gives the same archive one hard edge, replayable by anyone.

The law-written-twice cost, at its sourced strength. At least 21 heat-index algorithms in documented research use, the source's own conclusion being that "methods to calculate heat index are inconsistent across studies" (REPORTED — Anderson et al. 2013); five saturation-curve coefficient sets in simultaneous common use and "dozens" in the literature (VERIFIED — tc-2023-8); 13 metrics in one climate model's index module (REPORTED — Buzan et al. 2015); a psychrometer coefficient spanning 6.24×10⁻⁴ to 6.7×10⁻⁴ (REPORTED). No census of app-level implementations exists, and this page invents none. What can be said exactly: every duplicate implementation is a fresh chance for two instruments to render opposite verdicts at the same statutory line on the same observed air, and this court finds those hours and publishes them.

What this court does not judge

  • Physiology. The court computes a thermodynamic property of air from three archived integers. It does not model sweating, acclimatization, clothing, workload, age, or any body's response. The claim class, verbatim and load-bearing: a presented station-hour is verified; a body's response is not modelled.
  • Heat-stress science. Nothing here claims exact arithmetic replaces heat-stress modelling, chamber studies, or epidemiology. The 35 °C ceiling is Sherwood & Huber's; the 30.55 °C mean is Vecellio's, carried with its authors' own humid-regime caveat and their finding that hot-dry critical values are far lower still (25.75–27.82 °C at low vapor pressures — VERIFIED). The court grades instruments at the field's published lines; it neither moves nor endorses them.
  • Medical or occupational advice. This page recommends nothing to any worker, employer, agency, or patient. It is research output: a bookkeeping study on public archived weather bytes.
  • WBGT as observed. Natural wet-bulb plus globe is an instrument reading with siting, wind, and radiation in it, separated from psychrometric T_W by ISO 7243's own annex (VERIFIED, quoted above). The court grades computations, and only ever the computation it names.

Honest limits

  • The 2026 archive does not exist yet, and the charter says so instead of working around it. access/2026/ 404s; the S3 mirror holds zero 2026 keys against a passing 2025 control; the archive edge is 2025-08-24T21:00Z (all VERIFIED 2026-08-27). Loss (ii) is frozen but ungradeable until NCEI publishes 2026. A criterion graded on substituted data would be a different criterion.
  • Two seed claims failed verification and are corrected on this page. The "≥79 dead, >41,000 hospitalizations" figures are Japan-only FDMA July-2026 numbers, not a Japan+Korea total (Korea's KDCA counts are separate and smaller, cited above). And the Gulf top-0.1% T_W > 31 °C finding belongs to Raymond et al. 2020 in Science Advances, not Nature Communications Earth & Environment; the CE&E 2024 paper contributes the distinct evening-maxima finding. An archive fetched beats a number remembered.
  • The Stull band is REPORTED, not quoted. The formula is VERIFIED from the author's own textbook page; the −1.0/+0.65 °C band and its domain are consistently stated across independent secondary sources, but the AMS journal text 403-blocked automated fetch this session. S2 declares the band's provenance in its own text.
  • The WMO psychrometer pair was the one REPORTED coefficient set inside the identity, and is now VERIFIED. 653/1000000 and 118/125000 were re-read against WMO-No. 8 (2024 edition) Annex 4.B formula (4.B.8) on 2026-08-27 and match verbatim; the fetched document's sha256 is pinned above. The psychrometer coefficient is plural in the literature — the Guide itself records 6.50–6.78×10⁻⁴ for screen configurations — and exactly one is now frozen here.
  • SLP is not station pressure, and the exact key's P is therefore itself a convention away from sea level. ISD's SLP field is reduced to sea level, so at any elevated site the exact arm reads a convention, not the station's own air. The frozen |elevation| ≤ 50 m admission bound keeps the SLP-for-station-pressure approximation bounded, and the corpus-stage upgrade path is named with it: read ISD's MA1 station-pressure field where present, and grade from the station's own air.
  • Psychrometric T_W is itself a convention. The psychrometer equation's root and the adiabatic-saturation temperature are close but not identical objects; the exact key is exactly the frozen identity above, no more. The adversaries are graded against the same identity (ITER) or their own published definitions (STULL) — no arm faces a moving target.
  • ISD QC and siting are inherited, not adjudicated. The admission set is frozen and every exclusion published with its characters, but upstream QC, station siting, instrument changes, and archive revisions are NOAA's; HadISD exists precisely because raw ISD needs care (REPORTED). The court pins bytes as served, so an upstream revision is a new fact beside the old seal, never a silent move.
  • ROTHFUSZ cannot flip at a T_W threshold and is not pretended to. It outputs a heat index in °F; its census is a deferred act with thresholds frozen before scoring. What stands now is its documentary role: the operational instrument's own statement that no exact equation exists.
  • A flip is not a harm. A sealed flip-hour proves the instruments disagree at the line on that hour's archived air — nothing about what any person experienced or decided. S4's window intersection establishes stake, not causation, and the page will not conflate them.

Cross-links

Status: ACT 1 SEALED — section-zero gate GREEN (S1 controls, S2 replication 23,620/23,741, S5 discrimination with zero far-field breaches); 17 flips sealed at the 611/20 survivability line on the pinned 2025 Gulf slice; loss (ii)/S4 stands UNGRADEABLE until access/2026/ posts. The law above is frozen as run.

⚡ Paradigm

✅ Sealed — live court PROVEN

🔴 Live claims — standing, not sealed

☀️🌑 Eclipse 2026 — Study 01 DATA SEALED

🌊 Chartered — corpus pending

Clone this wiki locally