Methods: the temperature field that misses the daily high

Revision 2, October 4, 2026. Supporting material for Matt’s article. This is a historical measurement comparison, not a settlement or profitability study.

Scope and preserved history

The first study used O’Hare, LaGuardia and Miami. Its final result was 612 positive whole-F differences in 1,052 eligible station-days. All original deliverables remain byte-for-byte in history/v1/, with a preservation manifest. Original source files, scripts and results also remain at the evidence root. The initial, pre-audit calculation remains labeled in initial-analysis/.

After the first review, Matt selected weather and authorized this revision. The extension uses all five stations in the current PM-US weather FAQ. Four stations are new. Miami reuses the original file. No result pools the two studies. The original hypothetical two-degree bin exercise remains historical and is not used in the revised article.

The revision plan is revision-2/REVISION_PLAN.md. It was saved after source downloads and before their aggregation. Later inspection prompted the disclosed precise-temperature sensitivity, not a change to the primary eligibility rules.

Stations, source files and period

The study covers January 1 through December 31, 2025 at each station.

Station	NOAA identifier	METAR identifier	Fixed UTC offset	Eligible dates
Central Park	USW00094728	KNYC	UTC−5	344
Chicago Midway	USW00014819	KMDW	UTC−6	352
Miami International	USW00012839	KMIA	UTC−5	350
San Francisco International	USW00023234	KSFO	UTC−8	351
Los Angeles International	USW00023174	KLAX	UTC−8	352

Station selection follows the PM-US weather FAQ, read October 4. This does not establish that the same contracts or stations were offered in 2025.

The CSV inputs are NOAA’s LCDv2 bulk archive. File names follow LCD_<NOAA identifier>_2025.csv. Copies and exact URLs are bound by revision-2/source-manifest.json. Miami remains in sources/mia-2025.csv at the evidence root. Other station inputs are in revision-2/sources/.

LCDv2 combines hourly observations and daily summaries. Its documentation specifies metric units for these bulk CSV files. Both temperature columns use Celsius to tenths. Its time convention is local standard time.

Climate day and source-product limits

The October 2024 NWS instruction, applicable at the start of 2025, defines the complete CLI day as midnight-to-midnight local standard time. The June 2025 replacement retains that requirement in section 4.1.2. Both specify an overnight issuance for the previous complete day and another issuance for the current, incomplete day.

The fixed offsets above apply throughout the year. During daylight saving time, the climate day corresponds to 1 a.m. through 1 a.m. by the daylight clock. Raw-record checks nevertheless found timestamp discrepancies near the clock changes. We excluded affected dates instead of silently shifting their records.

An archived LCD summary is not a copy of the first CLI publication. NWS explains that CLI data are preliminary and NCEI conducts later quality control. We did not retrieve a historical CLI publication series or determine which revision any actual contract used. The historical measurements therefore cannot establish settlement errors or settlement outcomes.

As an archive cross-check, we downloaded GHCNd TMAX for all five stations, all 365 dates, in standard units. All 1,825 values match the corresponding LCD daily maximum after whole-F conversion. This is agreement between related NOAA archives, not independent sensor evidence. The GHCNd source flags identify 1,188 values as W, WBAN/ASOS summaries from ISD, and 637 as 1, NWS CF6 summaries. Neither flag proves identity with an original CLI. See the retained GHCNd source definitions.

Primary calculation and exclusions

For each station-date:

1. Select one unambiguous numeric DailyMaximumDryBulbTemperature from SOD records.
2. Select numeric HourlyDryBulbTemperature values from routine FM-15 reports.
3. Require at least 23 distinct local-standard hour labels among those routine readings.
4. Reject conflicting numeric temperatures at an identical timestamp across FM-15 and special FM-16 reports.
5. Reject the whole date if a routine or special raw UTC timestamp disagrees with DATE plus the fixed offset.
6. Reject the whole date if a precise raw T temperature group disagrees with its decoded temperature field.
7. Subtract the highest routine reading from the archived daily maximum, after separate whole-F conversions.

Conversion uses exact decimal arithmetic: °F = 1.8 × °C + 32, rounded to the nearest integer with ties away from zero. This is our comparison convention. It is not a claim about a contract’s rounding rule. An actual CLI value already reported in Fahrenheit is used at its published precision.

We checked 51,750 raw routine/special timestamps. Of those, 64 disagreed with the decoded timestamp. We also checked 50,104 precise temperature groups, with two disagreements. The clock discrepancies affect four dates per station. The temperature discrepancies affect Miami on July 3 and August 7.

Station	Fewer than 23 hours	Clock discrepancy	Same-time conflict	Precise-temperature discrepancy	Unique excluded dates
Central Park	17	4	1	0	21
Midway	9	4	2	0	13
Miami	9	4	1	2	15
San Francisco	10	4	2	0	14
Los Angeles	9	4	1	0	13

Reason counts overlap. We excluded 76 of 1,825 station-dates. The complete eligibility ledger and raw audit findings are in revision-2/daily-results.json and revision-2/results.json. We did not repair suspicious values or exclude retained disagreements after inspecting their direction.

Primary results

A positive gap means the archived daily high exceeds the maximum routine current-temperature field.

Station	Eligible	Positive	Equal	Negative	Positive share	All-24-hour result
Central Park	344	202	141	1	58.7%	193/330, 58.5%
Midway	352	186	166	0	52.8%	182/344, 52.9%
Miami	350	241	109	0	68.9%	237/342, 69.3%
San Francisco	351	234	116	1	66.7%	228/342, 66.7%
Los Angeles	352	233	119	0	66.2%	225/343, 65.6%

Positive-gap distributions, in ascending gap size:

• Central Park: 1°F on 180 dates, 2°F on 19, 3°F on three.
• Midway: 1°F on 158 dates, 2°F on 25, 3°F on two, 4°F on one.
• Miami: 1°F on 191 dates, 2°F on 44, 3°F on six.
• San Francisco: 1°F on 180 dates, 2°F on 42, 3°F on nine, 4°F on three.
• Los Angeles: 1°F on 181 dates, 2°F on 41, 3°F on seven, 4°F on one, 5°F on three.

Gaps of at least 2°F occurred on 22, 28, 50, 54 and 52 eligible dates, in the station order above. Those shares are 6.4%, 8.0%, 14.3%, 15.4% and 14.8%.

Central Park on March 2 has a 33°F daily maximum and a 37°F routine value. The latter comes from a report without a precise T group. San Francisco on October 10 has a 69°F daily maximum and a precise routine value of 72°F. We retain both discrepancies. Their causes are unresolved.

A sensitivity requiring at least 23 distinct hours with precise T groups yields positive counts of 196/334, 186/352, 241/349, 234/350 and 233/351, respectively. Central Park’s reverse case disappears. San Francisco’s remains. This sensitivity changes eligibility and is not the primary result.

Adding special FM-16 current-temperature observations yields positive counts of 194/344, 175/352, 235/350, 227/351 and 226/352. Reverse cases number three, zero, zero, one and three. More reports still do not create a complete daily maximum series. We make no separate causal claim from these reverse cases.

Midnight maximum remarks: matched comparison

The ASOS manual, section 3.1.2, describes the calendar-day maximum in the local-midnight METAR’s 4snTxTxTxsnTnTnTn group. Its maximum and minimum use tenths Celsius. The report normally precedes 23:59 LST and can differ from the complete daily summary.

We parsed that exact nine-digit pattern only in routine METAR remarks after RMK. For comparison, we required a primary-eligible date, a report stamped in hour 23 LST, and one unambiguous maximum. Raw clock checks therefore also apply. Observed retained report times are 23:51 at Central Park, 23:53 at Midway/Miami/LAX and 23:56 at SFO. Out-of-hour examples occur near clock changes and do not enter the paired comparison.

Station	Paired dates	Routine maximum equals daily high	Midnight remark equals daily high	Primary dates without eligible remark
Central Park	340	139	339 (99.7%)	4
Midway	351	165	348 (99.1%)	1
Miami	349	109	347 (99.4%)	1
San Francisco	351	116	350 (99.7%)	0
Los Angeles	352	119	351 (99.7%)	0

Both columns use exactly the same dates. The eight mismatches remain in the results, including extreme values:

Station and date	Archived daily high	Midnight remark
Central Park, November 25	58°F	57°F
Midway, May 9	65°F	68°F
Midway, August 20	77°F	79°F
Midway, October 9	63°F	75°F
Miami, November 11	68°F	77°F
Miami, December 30	68°F	70°F
San Francisco, February 19	64°F	55°F
Los Angeles, December 3	67°F	111°F

The retained LAX raw group is 404390100, whose maximum decodes to 43.9°C, or 111°F. We do not treat this as a credible physical high or diagnose its cause. It is a disagreement in the archived record. We did not discard it to improve agreement. Midnight remarks are not an automatic replacement for the contract’s source.

Miami example and six-hour periods

For January 20, 2025, Miami has readings in all 24 routine hours. The highest current-temperature field is 21.1°C, or 70°F. The SOD daily maximum is 22.8°C, or 73°F. At 23:53 LST, raw remark 402280144 also reports a 22.8°C maximum. The 00:53 LST report that carries the 21.1°C value is the first routine report of the climate day. The SOD record gives no occurrence time for its maximum, and we do not infer one.

At 00:53 LST that morning, raw UTC time 200553Z identifies the report associated with the 06 UTC synoptic hour. Its six-hour maximum group 10256 reports 25.6°C, or 78°F. At the synoptic hours, ASOS computes the six-hour maximum from the stored hourly maxima (ASOS manual, section 3.1.2). The window for this group therefore ends at the 00:53 LST report and begins about six hours earlier, near 18:53 LST on January 19. The 18:53 LST reading on January 19 is also 25.6°C. The group does not supply the time of its maximum. We do not assign 78°F to January 20’s climate day.

Six-hour windows follow UTC synoptic reporting times, not a universal local calendar-day boundary. NWS also notes the minutes between transmission and the nominal hour. We did not aggregate four six-hour maxima by local report date. Raw records and line numbers are in revision-2/miami-january-20-records.json.

Concrete rules-to-source example

The current PM-US FAQ identifies Miami’s KMIA station and links directly to CLIMIA. The retained NWS response has header CDUS42 KMFL 040827, identifier CLIMIA, and issue time 4:27 a.m. EDT on October 4, 2026. It covers October 3 and reports MAXIMUM 87 in its Fahrenheit table. The occurrence time is 1:27 p.m. LST. The snapshot is revision-2/sources/current-miami-cli.html, with extracted text alongside it.

The April 3 filed temperature template, Attachment A, defines strict “above” and inclusive “at least.” At a reported value of 87°F, the illustrative predicates 87 > 87 and 87 >= 87 are false and true. We did not identify or evaluate a traded 87°F contract.

The template uses the first finalized source publication, with contract terms specifying finalization where necessary. The current FAQ names CLI and describes 8 a.m. ET settlement. If the CLI reading is inconsistent with the 24-hour METAR observation for the same location, settlement may be delayed until 11 a.m. ET for review. These references demonstrate the rule/source chain. They do not settle the finalization terms of an unidentified contract. Later NCEI quality control and contract finalization are distinct processes.

The CLI URL is mutable. The retained snapshot supports the example, even after the live URL advances. A web-reader response showed an older cached September report. The direct NWS download supplied the October 4 issue retained here. No claim relies on the stale response.

Reproduction, verification and graphics

Run these scripts from any working directory with Python 3:

python3 sessions/evidence/next-field-note-2026-10-04/revision-2/analyze.py
python3 sessions/evidence/next-field-note-2026-10-04/revision-2/remarks.py
python3 sessions/evidence/next-field-note-2026-10-04/revision-2/verify.py
python3 sessions/evidence/next-field-note-2026-10-04/revision-2/verify_remarks.py

verify.py independently rebuilds raw timestamp/temperature exclusions and recomputes daily maxima with SQLite. It imports no analysis functions and matches all 1,749 eligible station-dates. verify_remarks.py uses separate token parsing and integer arithmetic. It reproduces all 1,743 paired remarks and all 1,825 archive comparisons. These are independent calculations within this worker session, not a paid external review.

revision-2/make_figures.py generates two editable text SVGs, two outlined SVGs, PNG exports and 390/350-pixel previews. The palette uses the accepted Splitrule light-theme tokens. Labels use Inter and numerals use JetBrains Mono. Sources, font assets and bounds-check results remain in the revision directory. The local reading preview links this article and methods without a deployment.

What this study establishes

The results quantify differences among fields and reporting periods at five named stations during one year. Missing hours do not explain the main pattern. Maximum remarks recover information absent from a maximum over routine current-temperature fields, but some records disagree.

The study does not measure live availability, latency, forecast skill, actual contract strikes, transaction costs, profits, or settled outcomes. It does not attribute every gap to a between-report temperature peak. Rounding, report faults and later quality control can also matter. Dates and stations are not random independent draws, so we provide descriptive counts rather than inferential confidence intervals.

The reporting distinction is established in NWS guidance and existing weather-trading explainers. Iowa State’s IEM CLI audit page shows the audit path from raw METARs and six-hour groups through DSM, CLI and CF6 reports. It also notes the local standard day. NWS Los Angeles explains that the high can occur between five-minute observations. Kalshi’s weather help page describes the same local standard day, which runs 1 a.m. to 1 a.m. by the daylight clock during daylight saving time, and a consistency check against METAR six-hour and 24-hour highs. We read all three pages on October 4, 2026. Prior research also discusses temporal sampling and the difference between hourly forecast maxima and realized daily maxima. The original RESEARCH_LOG.md retains the bounded search. We make no priority claim and no claim that a weather audience on X or Reddit has demonstrated demand for this article.