Independent final editorial review — 2026 temperature-record hurdle

Review identity and revision

• Story: The final-quarter hurdle for a new temperature record. Audience: curious readers of climate-record news, climate journalists and educators seeking a concrete interpretation of the September bulletin.
• Reviewer: /root/temperature_final_evidence_review, 15 October 2026. I did not author or revise the outline, source material, calculation, charts, prose or tracker. I wrote the reader questions before opening the outline or draft. This is an independent six-dimension review using DataPressr's skills/story/references/editorial-review.md, story craft, voice guide and charting instructions, plus Factory AGENTS.md.
• Reviewed Git commit: 752bd0f13fa94fea120c3f109d091765cc2bda6d (prose and draft index). Earlier chart commit: 1ff890dad3fdc2686fcea71abe1b30737bca078e; outline commit: 44c7bcd6e42c869aa40ecb267ee8fb25a8059f81. The three-stage order is present in Git. The only tracked working-tree change during review was the coordinator's .beads/issues.jsonl.
• Earlier outline approval: stories/2026-temperature-record-review.md, independent reviewer /root/temperature_outline_review, APPROVED for outline SHA-256 c3697625987a30d03c7f850f8317b0eed89acb2bd1fcbd8dc30d9c09de3b0932. My own hash matches. Its historical staging/pending wording is preserved as evidence; the present review records the subsequent exact commit.
• Input manifest: independently calculated below, also /tmp/factory-temperature-final-manifest.json. Source versions are pinned by retrievals.json: CSV updated 3 October 2026, server Last-Modified 9 October; snapshot retrieved 15 October, containing observations only through September. Six original downloaded files match their declared SHA-256 and byte counts. I independently fetched the live CSV again: 51,706 bytes, identical SHA-256.

Reader questions — written before reading the draft

Original questions are saved unchanged in /tmp/factory-temperature-final-reader-questions.md.

1. How far behind or ahead of 2024 is 2026 through September, and what must the final quarter average? Answered. Lines 17–19 give the tie hurdle and strict exceedance condition; line 25 gives both January–September levels. Chart 2 shows the accumulated deficit. The reader can distinguish the required quarter from the annual result.
2. Has a previous final quarter been that warm, and how far above precedent is the requirement? Answered. Lead, chart 1 and line 25 identify the 2023 maximum and 0.17–0.18°C excess. Chart 1 provides the complete historical comparison, rather than selecting a convenient recent year.
3. Is 2026 warming toward that finish or cooling away, and what does September tell us? Partly answered, acceptable. Chart 2 shows the rising cumulative 2026 mean and September endpoint; the draft does not give September's standalone 0.8240°C anomaly or extrapolate a monthly trend. That omission preserves the narrow annual-arithmetic argument. The explicitly attributed El Niño context offers relevant direction without inventing a forecast from the plotted curve.
4. How does 2026 compare with leading years at the same point? Answered for the relevant comparison. Lines 21, 25 and 29 give second place and the first-place benchmark. A full league table would add bulk without changing this question's answer. My independent ranking confirms the claim against every year, not just the two plotted paths.
5. What annual average, baseline, dataset and calendars are used? Answered. Lines 19, 23 and 40 explain both weights, monthly climatologies and source; the linked calculation handles leap years correctly. No pre-industrial anomaly is substituted for the chart baseline.
6. Is this a prediction/probability, and what does it imply about a record? Answered. Lines 19, 31 and 40 separate a tie calculation, physical uncertainty, historical precedent, likelihood and official classification. The reader is not promised odds this method cannot supply.
7. How current and authoritative is the source, and could revisions matter? Partly answered, acceptable. Line 23 gives the observation cutoff; frontmatter and linked provenance give the frozen vintage and primary source. The linked catalog explains possible early-release revision. The prose does not quantify revision risk because this input supplies no physical uncertainty interval. Its snapshot qualification is adequate for a conditional calculation, not a promise of a final official record.
8. Does missing 2024's record mean warming has reversed? Answered at the required level. Line 29 says second place is already warm; line 31 separates a single year's rank from long-term warming. The historical chart also shows the longer-run context without a new causal claim.

Evidence and editorial findings

Independent numerical reproduction

My calculation: /tmp/factory-temperature-final-scan.py; complete independently generated series and results: /tmp/factory-temperature-final-numbers.json. My subsequent cross-check/link/hash script: /tmp/factory-temperature-final-checks.py. These files were not copied from the earlier review or author script. Author statistics.json was used only after my own scan, to compare the quantities actually consumed by the chart builder.

Input fields: month and supplied ano_91-20, degrees Celsius relative to each month's 1991–2020 climatology. All 1,041 monthly keys are unique and consecutive from 1940-01-01 through 2026-09-01; all five numerical columns are finite. There are exactly 86 complete October–December groups, 1940–2025. No missing-month interpolation or zero filling is used.

For equal months, Q=(12*A2024−9*S2026)/3. For day weights, Q=(365*A2024−273*S2026)/92. Each 2024 mean uses its own calendar: 366 days in the year and 274 through September; 2026 has 365 and 273. Every final quarter has 92 days. Repeating the three 2023 monthly anomalies in 2026 preserves the same October, November and December day weights.

Both conventions rank all available January–September years with 2024, 2026, 2025, 2023, 2016, 2020 as their top six. Complete-year rankings both put 2024 first, followed by 2023 and 2025. Thus the chosen annual target and partial-year rank are justified by the full dataset. The 2023 repeat inputs are October 0.8501, November 0.8541 and December 0.8450°C. September 2026 is 0.8240°C: its standalone anomaly must not be confused with the 0.62°C cumulative mean, and the chart/prose labels avoid that confusion.

All 86 historical values under each convention, every 2024/2026 cumulative point, all six scenario cumulative points and the top-six rankings agree with the JSON consumed by the charts to floating precision. The month-specific climatologies and offsets are constant by calendar month; ano_pi = ano_91-20 + offset_pi is exact in Decimal arithmetic. The maximum absolute-temperature-minus-climatology discrepancy is 0.0001°C. Using the supplied anomaly is therefore appropriate.

Rounding: nearest-0.0001 inputs allow at most 0.00005°C error in a historical weighted mean and 0.00035°C / 0.0003467391°C in the equal-month/day-weighted hurdle. Within each matching convention, conservative hurdle-minus-record lower bounds are 0.1802333333 and 0.1736406213°C. Both exceed 0.17°C. These are input-rounding checks, not physical uncertainty estimates. The two-decimal display ranges are correct; the prose expressly warns that exceeding a rounded label does not establish strict exceedance of the unrounded target.

All data numbers in reader-facing prose and alt text are on the charts: both hurdle/excess ranges, baseline years, comparison years, maximum Q4, YTD means/ranks, annual benchmark, scenario mean and shortfall. The friction note's “No exempt data numbers” statement is accurate. Metadata dates and the executable command are not unsupported story data claims.

Primary sources, coverage and rights

I opened the live C3S September bulletin and methods page, and inspected the frozen evidence. The bulletin supports second place through September, the late-2023 conditional comparison and the attributed strengthening El Niño context. The methods page defines ERA5 surface air temperature across land and ocean and distinguishes annual from monthly pre-industrial conversions. The story sensibly retains its single 1991–2020 scale. It does not claim that either averaging convention reproduces all details of the eventual official classification.

The ERA5 catalog confirms coverage from 1940 and explains possible revision of early data; the licence page supports the stated CC BY reuse. Chart attribution names C3S/ECMWF, ERA5, the snapshot and modified information. The source HTML's archival retention is distinguished from dataset reuse rights. No new unsupported external data number appears in the prose.

“Released in October” avoids making an unnecessary day-specific claim from the archived news page's differing page date and dateline. The source vintage and the last observation are not conflated with the story's date.

Visible chart and full-page inspection

All four committed PNGs were opened directly, not merely checked as SVG text:

• 2026-temperature-record-hurdle-720.png: 720×620. Complete negative and positive historical domain; zero reference visible; maximum and hurdle labels separate; no 2026 observation is invented.
• 2026-temperature-record-hurdle-360.png: 360×310. Finding and historical distribution remain discernible; no clipping or overlap. Ancillary type is small.
• 2026-temperature-record-repeat-720.png: 720×700. Actual 2024 and observed 2026 paths directly named; dashed conditional continuation and open endpoint clearly differ from actuals. September and December comparison rows provide exact values without crowded endpoint labels.
• 2026-temperature-record-repeat-360.png: 360×350. Complete figure and comparison rows visible; no label collision. Blue hypothetical label applies to 2026 only, preserving the real 2024 continuation.

I also opened /tmp/factory-temperature-full-story-desktop.png (1200×3133) and /tmp/factory-temperature-full-story-360.png (360×3271). These show the complete rendered story, including the method. The local preview HTML is now /Users/rgrp/src/datasets/factory/.runtime/factory-drafts-local.html; image source files are staged copies of the exact reviewed SVGs. CSS gives desktop content width 752px inside an 800px main element and phone content width 328px after 16px padding each side. I observed readable body text, intact title, both charts, method links and draft/voice notice. No page-level horizontal overflow or clipping appears. The shell commands use an internally scrollable code box on the phone; this does not hide prose or distort chart axes.

The chart builder uses Observable Plot; numbers are derived from verified statistics. Both historical weighting choices and the hurdle strip are honestly labelled. Sensitivity envelopes are extremely narrow because the differences really are small; they are not inflated into a false uncertainty interval. The second chart includes zero and marks the October–December region as hypothetical only for 2026. SVG accessibility descriptions and Markdown alt text name the comparisons and values.

These are local rendering checks, not verification of a deployed DataHub page. The preview coordinator reports a deployment package that excludes archived third-party HTML/text and points provenance to original sources. Its altered provenance is a packaging artifact outside this review's source manifest; source story and figures remain unchanged. Publishing must retain functional data/build/provenance links and draft status.

Reproduction and readiness checks

Commands actually run in /tmp/factory-temperature-final-build, an isolated copy of stories/:

npm ci --no-audit --no-fund
python3 2026-temperature-record-src/calculate.py
node 2026-temperature-record-make-charts.mjs
python3 2026-temperature-record-src/calculate.py
node 2026-temperature-record-make-charts.mjs

Both runs succeeded with Node v26.4.0 / npm 11.17.0. Hashes for both SVGs and statistics.json equal each other and the committed originals; evidence in /tmp/factory-temperature-final-rebuilds.json. Pinned dependencies are Observable Plot 0.6.17 and jsdom 30.0.1. git diff --check succeeds. Story/source/code were not modified by this review.

The story's method links stay within the story folder, so they do not escape a published root. The actual story directs commands from stories, correctly. The source provenance's phrase “From any working directory” before a repository-relative command is overbroad; an optional documentation tidy would replace it with “From the repository root”. It does not block reproduction using the draft's correct instructions.

Outcome

• Strongest objection: a reader might interpret “above every previous final quarter” as a forecast that 2026 cannot set a record. This is adequately addressed: the draft leads with a conditional calculation, shows how close the repeat scenario comes, explicitly rejects a historical ceiling and gives the relevant attributed El Niño context. The full-year result is separated from official classification. No blocking objection remains.
• Most valuable revision — optional: make the title identify the year and the result, for example “2026 needs a hotter final quarter than any in the record”. The current title is accurate but generic outside the story page; the more concrete title would carry the reader value into an index or link. No change is required to clear this review. Any adopted title edit still needs a narrow re-review of its wording and metadata.
• Verdict: APPROVED for the exact committed draft and hashes below, for agent-reviewed draft handoff only.
• Required corrections: none. Acceptable limitations: conditional arithmetic gives no probability or physical uncertainty; recent inputs may be revised; very small phone chart attribution is supported by ordinary-sized prose and source links; local rendering is not a deployed-site validation.
• Re-review: this is the first final-review round. No material edits were made during review. Later material changes to title, claims, charts, figures or method reopen review; changes to the argument/chart plan also reopen outline review.
• Outstanding human stage: author's personal voice pass and owner publication approval. Agent approval does not certify the author's voice, change publication status or authorise publication/messages. The complementary writing review is separately documented by its reviewer.

Independently calculated SHA-256 manifest

Paths are relative to /Users/rgrp/src/datasets/factory.

Full-page local preview evidence: