Engineering brief, Tuesday, 27 October 2026

The bottom line

Good morning. This is the engineering brief for Tuesday, 27 October 2026: everything that merged to main between midnight and midnight UTC, and what needs you.

224 pull requests merged, 26% below the seven-day average of 302.1. Together they changed 1,716 files, adding 71,376 lines and removing 1,361. The work reached 14 products, 7 research projects and 8 blog post changes.

The headlines: 1 new product joined the catalogue and 2 new research projects started.

Health: CI, main and the queue

This edition is from the archive, so the CI and pull-request state of that day was not recorded; the live numbers are in the newest edition and on the development dashboard.

Fix-forwards: 1 merge repaired a red main or a failing check without reverting.

• process: fix red validate on main (run record names PR 10937) #10938

Products

Work landed on 14 products, 129 merges in all. New in the catalogue: actions-continue-on-error-check.

ai-bookkeeper: 42 merges.

• Match foreign invoices to payments in their own currency #10762
• Pin twelve amount spellings in a test #10764
• Pin cross-currency credit note behaviour #10766
• Property test for foreign documents paid in instalments #10810
• Read Publication 1099 (2026) for the 1099-K and backup withholding figures #10814
• Redteam: ai-bookkeeper billpay 2026-10-27 #10820
• Read Fidelity cash-export payees from Action and skip buys #10831
• HDFC "Value Dat" no longer read as Amount; Txn Date over Value Date; footer line #10836

Plus 34 more product changes.

wall-framing-takeoff: 21 merges.

• Read IFC floor plans #10768
• Foundation corner L-bars (cornerBars option and page select) #10770
• Anchor bolt fields on the page #10772
• Redteam: wall-framing-takeoff ifc reader 2026-10-27 #10775
• Per-panel sheets in the PDF drawings #10777
• One printable sheet per shop panel (--panel-sheets) #10779
• Gable panel plates cut along the slope #10787
• Nail counts by connection (--fasteners) #10791

Plus 13 more product changes.

changepoint-detector: 16 merges.

• DriftWarning for steps clustered under --drift #10774
• DriftWarning false-alarm rate measured #10785
• Poisson --bootstrap with --trend, season and nb measured #10826
• --multiplicative season for the Gaussian path #10839
• Multiplicative season on the real airline passengers series #10849
• Real autocorrelated series (Lake Huron levels), pinned in a test #10850
• --bootstrap with known holidays behaves like a season (eval, README, post) #10852
• Diagnose --ar1 on Lake Huron (profile over rho) #10858

Plus 8 more product changes.

hearth-and-horizon: 13 merges.

• The Grey Hour (carp pond oxygen story, text only) #10763
• The Salt Line (cistern leak story, text only) #10786
• The Coal Charity (fuel allotment charity, 2046; text only) #10788
• The Second Slope (pumping-test boundary story, text only) #10793
• The Grandmother's Sorghum (seed-store germination story, text only) #10830
• The Rope at the Side Gate (crowd density and an uncounted gate, text only) #10859
• The Day He Is Married (will condition and tribunal clock, text only) #10890
• The Fifth Asking (2046 chaplaincy, text only) + blog post #10892

Plus 5 more product changes.

einvoice-preflight: 10 merges.

• FA(3) rules measured on the Ministry's own KSeF templates (14 files, agrees with xmllint) #10790
• Colombia and Chile pages with NIT and RUT check-digit tests #10803
• Peru and Argentina pages with RUC and CUIT check-digit tests #10811
• Measure check-digit blind spots across seven ID rules #10813
• Uruguay page with RUT check-digit test #10815
• Paraguay SIFEN page with RUC check-digit test #10819
• Ecuador SRI page with RUC and clave de acceso check-digit tests #10821
• Measure Ecuador RUC establishment-digit blind spot #10823

Plus 2 more product changes.

saints-by-candlelight: 7 merges.

• Episode 38, Elisha and the Widow's Jars of Oil (text only) #10783
• Episode 39, Isaac and the Wells That Were Dug Again (text only) #10802
• Episode 39, David and the Giant in the Valley of Elah (text only) #10806
• Episode 41, Francis and the Christmas Night at Greccio (text only) #10832
• Episode 42, Gideon and the Three Hundred with Jars (text only) #10863
• Episode 43, Benedict and the Sieve That Was Mended (text only) #10915
• Episode 43, Hannah and the Prayer at Shiloh (text only) #10916

league-schedule-generator: 5 merges.

• Random-league build check (3,200 leagues valid) + blog post #10893
• Fuzz rainouts, late teams and officials; validator accepts a named make-up date on a skip day #10923
• Fuzz multi-division builds and rainout make-ups #10936
• Fuzz withdrawals, shared grounds, linked teams, travel and carry-over #10942
• Fuzz suggestFairnessWeight #10946

quick-pick-survey: 4 merges.

• Pair selector across lapse rates 0.05 to 0.5 #10765
• Pair selectors against a first-option-biased respondent #10840
• Hybrid selector falls back to shipped for a first-option-biased respondent #10844
• Hybrid selector falls back for second-option bias too #10870

shift-roster-builder: 4 merges.

• OpenShiftWeight prices an open shift in hours for coverRepair (15 h/5 shifts at 0, 19 h/3 at 100) #10842
• Open-shift preference on page and CLI #10866
• Tell the team text and email links on the admin page + blog post #10903
• --changes prices late schedule changes in the CLI #10924

actions-continue-on-error-check: 2 merges.

• Find jobs and steps that can fail and still go green #10827
• Redteam: platform/distribution/precommit.mjs 2026-10-27 #10972

sales-agent: 2 merges.

• Skip salutations when reading the job title #10778
• Read HubSpot Associated Company and Salesforce mailing country / email opt-out headers #10837

ai-crawler-access-check: 1 merge.

• Fix CLI table alignment, add llms.txt #10912

patent-maintenance-fee-planner: 1 merge.

• Redteam: patent-maintenance-fee-planner usage opt-out 2026-10-27 #10871

za-freelancer-provisional-tax-set-aside-calculator: 1 merge.

• Redteam: za-freelancer provisional tax calc 2026-10-27 #10966

Research

7 research projects moved. Started that day: aerolog-scaling and marine-hull.

aerolog-scaling: 14 merges.

• Open programme, Froude-scaling calculator and first-order result #10809
• Airfoil-polar check of H2 and H3 #10825
• Milestone 3a longitudinal modes (H1) #10838
• Milestone 5a, propeller efficiency vs Reynolds number (UIUC data) #10843
• Open TP-1435, check every reference #10862
• Milestone 6a, measured power vs mass exponent (Rodrigues flights) #10879
• Milestone 7a, correction search for H5 #10891
• Day-clustered bootstrap for the m^1.5 exponent (milestone 6b) #10900

Plus 6 more research changes.

marine-hull: 9 merges.

• Open the marine-hull programme (track 23) #10894
• Milestone 2, Michell wave term replaces the placeholder #10902
• Milestone 4, glider vs propeller AUV #10929
• Milestone 4b, glider vs AUV with Slocum coefficients from gliderflight #10950
• Milestone 4c, REMUS-class hotel load from vendor endurance #10954
• Milestone 5, hydrofoil rung vs displacement monohull #10956
• Milestone 4d, energy-delay of glider vs REMUS-class AUV #10957
• Milestone 4e, break-even value of a day of delay (glider vs AUV) #10959

Plus 1 more research change.

verifiable-training-mechanisms: 9 merges.

• Milestone 32, paper H3 section with measured agreement #10781
• Side payments do not bring back 1/m (H4, milestone 33) #10796
• Milestone 34, a bribable auditor #10845
• Milestone 35, partial copying (kappa) in the m-worker game #10854
• Milestone 35, partial copying in the m-worker game #10855
• Milestone 37, pricing replication against a single worker #10857
• Milestone 38, joint audit-stake-replication pricing #10920
• Milestone 39, replication vs faulty workers #10925

Plus 1 more research change.

verified-dse: 9 merges.

• Milestone 24, second corner (ss_100C_1v60) #10776
• Milestone 24, rung bias at 65% utilisation #10780
• Milestone 26, strategy replay on the ss corner #10784
• Milestone 27, pilot gate on five unseen designs #10868
• Milestone 28, designs built to make R2 fail #10872
• Paper covers milestones 24 to 28 #10901
• Milestone 29, pooled significance on ten held-out designs #10906
• Milestone 30, repair-driven inversion on high-fanout designs with rich outcomes #10917

Plus 1 more research change.

llm-coding-inference: 7 merges.

• Milestone 6s, Wilson-upper pilot variance at a rare category #10769
• Milestone 6t, sequential audit at a rare category #10795
• Process: llm-coding-inference scripts fetch all GoEmotions parts themselves #10834
• Milestone 6u, Wilson-upper stopping at a rare category #10886
• 6v, a wider critical z does not repair the rare-category sequential audit #10926
• 6w, rare-category sequential audit fails through a downward bias at the stop #10964
• 6x, a label-free audit size stays valid at a rare category #10982

lowcomm-training: 7 merges.

• Milestone 5d, median aggregation at M = 8 #10798
• Milestone 5d, trimmed mean and median at M = 8 under a sign-flip attacker #10799
• Order-statistic check of the median's sign-flip cost (5f) #10828
• 5g heavy-tailed check of the median order-statistic account #10867
• 5h measure real worker-delta tail weight and mu/sigma #10898
• 5i median flip cost at M=16 reaches the noise floor #10960
• 5i-b, trimmed-mean arm at M=16 (median arms landed in #10960) #10958

sim2real-inference: 2 merges.

• Fix order-dependent analyse import in benchmark test #10816
• Verified citations log + post #10824

Writing

8 merges touched the blog, with 156 new posts.

• dogfood: run actions-continue-on-error-check over our own workflows #10835
• process: fix main_health red, missing front matter on shift-roster-builder post #10846
• process: post-date check allows no day of slack; redate 19 posts #10848
• process: fix main_health red (3 future-dated posts, 4 long descriptions) #10853
• posts: fix front matter on three posts failing check_post_frontmatter #10874
• posts: merge five split tag spellings; check flags tag variants #10895
• process: fix main_health red, five post descriptions over 160 chars #10935
• redteam: platform/pages/backup.mjs 2026-10-27 #10939

Platform, CI and process

30 merges changed the machinery the products and research run on.

process, prompts and the ledger: 17 merges.

• Prompts: review 2026-10-27 #10767
• Wall-framing-takeoff: shop drawings as a PDF with no browser #10771
• Podcast: script for the coal-charity post (no audio: audio over 400 MB cap, huggingface.co blocked) #10789
• Process: measure friction table marks deleted files as not live #10805
• Process: promptctl measure tags friction keys for deleted files #10807
• Podcast: script for The Second Slope post (no audio) #10817
• Prompts: review 2026-10-27 #10818
• Process: creative prompt says audio is text-only while over the cap #10847

Plus 9 more platform changes.

platform/ci: 7 merges.

• Process: wait_for.py, a safe wait for background jobs #10773
• Redteam: platform/ci/validate_workflows.py 2026-10-27 #10782
• Ci: parser-based markup basics check for all venture pages #10878
• Redteam: platform/ci/ci_starvation.py 2026-10-27 #10885
• Redteam: platform/ci/check_jsonld.py 2026-10-27 #10918
• Redteam: check_inpage_refs 2026-10-27 #10930
• Redteam: platform/ci 2026-10-27 #10934

platform/distribution: 3 merges.

• Redteam: platform/distribution/bluesky.mjs 2026-10-27 #10841
• Redteam: ghaction 2026-10-27 #10875
• Redteam: platform/distribution/citation.mjs 2026-10-27 #10974

platform/pages: 2 merges.

• Redteam: platform/pages/entities.mjs 2026-10-27 #10905
• Redteam: published-base-url 2026-10-27 #10921

platform/business-template: 1 merge.

• Redteam: business-template 2026-10-27 #10833

The agent fleet

By co-author: Claude Sonnet 5.5 on 187 merges and Claude on 37 merges.

192 agent runs filed a run record for the day. By track: ventures 68, research 56, creative 21, redteam 17, process 13, explore 11 and review 6. By outcome: success 169, partial 22 and failed 1.

The 52 runs with measured usage came to $38.75 at API list prices, an equivalent rather than a charge, over 376 active minutes.

Runs recorded 6 errors; the first: first run of nonsimilar_modes failed: aerosandbox missing from python3.13 until pyenv.py installed it.

Documentation that cost runs time, most reported first:

• research/CLAUDE.md, 25 reports: No mention that OpenROAD installs from the nixpkgs binary cache with `nix build nixpkgs#openroad` (a minute), while chip-pnr's README says it compiles for 1.5 h.
• portfolio/flagships.yaml, 16 reports: next: item 9 says the IRS is unreachable; irs.gov answered, the revenue procedure itself is not out.
• prompts/process/prompt-1.md, 8 reports: Only owner rows and blocked P056 open again; main_health.py was green, so critique had to come from friction table, whose lowcomm and post-date entries were already fixed.

In the runs’ own words

A few notes runs left in their records, verbatim:

• Small flagship 2 slice: anchor bolt page fields; page.test and foundation.test pass (36). Venture preflight not run; CI covers it.
• Backlog ids: none (P056 blocked). Added platform/ci/wait_for.py + test.
• Project: hearth-and-horizon; piece: three-plain-and-a-bob (text only, audio 423 MB over 400 MB cap). Touch counts computed by program (4 of 4096 patterns true). Venture tests 11/11.
• IFC reader on hand-written fixtures; no real export tried (buildingSMART samples not reachable).
• Reduced pass: measure, derive fetch/build, one doc fix (TeX trap in research/CLAUDE.md), E96, fixed a 179-char post description that failed validate. Candidates still keep collecting.
• Milestone 24: 600 OpenROAD points at 65% utilisation; 4 predictions held. Checks run: build.py ok, check_md_links ok, research_live_links ok.

Watch list

• research/CLAUDE.md keeps slowing runs down; a clearer page there pays back on every run.

That’s the brief for Tuesday. Every item links to its pull request on the page.