promote: staging → main (2026-10-20-2) — 9bb5265ae by fedcourtsai-dev[bot] · Pull Request #2590 · ModelMirrorAI/fedcourtsai

fedcourtsai-dev · 2026-10-20T14:23:10Z

Drafted by the promotion train at 2026-10-20T14:35Z for promote/2026-10-20-2 at 9bb5265ae over main f9bdab01a. The branch was cut from staging and never moves, so the integration evidence and an approval of this head stay valid while staging carries on. The train regenerates the block between the markers while this PR is open; write anything else outside it.

Gate status (promote/2026-10-20-2 at 9bb5265ae):

freshness: not yet — freshness: no green 'ranged-reads' integration-test run at 9bb5265;freshness: no green 'corpus-service' integration-test run at 9bb5265;freshness: no green 'stub-cascade' integration-test run at 9bb5265
registry: not yet — registry: predictor codex-baseline computes sha256:70fee158526caa6870d43ace70c3781db39f644379c86c363538ebdefa57547c, which the registry does not register; run fedcourts register-process in the PR that moved it;registry: predictor gemini-baseline computes sha256:a9033e56819e775e561b802dec24bae437c17c751e5a7f5fa4b3eeb31383951f, which the registry does not register; run fedcourts register-process in the PR that moved it;registry: evaluator codex-judge computes sha256:18a76b9301aef6770b5c5f9965dd6fcabb2a929964a56be2151405ce878e0456, which the registry does not register; run fedcourts register-process in the PR that moved it
contexts (admin read, run by hand): due — the batch changes .github/workflows/ci.yml, .github/workflows/integration-test.yml, .github/workflows/promotion-prep.yml, .github/workflows/run-analytics.yml, .github/workflows/run-backtest.yml (+9 more).
Run scripts/promotion-gate.sh contexts with an admin-read token before merging.
It reads main's workflows, so for the workflow changes run scripts/promotion-gate.sh contexts again once this merges.

Train: the integration gate is running (https://github.com/ModelMirrorAI/fedcourtsai/actions/runs/39440437340)

Next (maintainer): read the batch and approve this PR at 9bb5265ae. The train then merges it with a merge commit once the integration gate run, the required checks and the registry gate hold at this head and no predict/evaluate/back-test round is in flight. Merging it yourself instead is fine. To stop it, request changes or close it: a closed candidate is not re-cut at the same staging tip, and the next one is cut when the train is dispatched after staging moves.

The batch

docs(security): record the case-sensitive ruleset patterns and the case-variant guard #2543: docs(security): record the case-sensitive ruleset patterns and the case-variant guard
docs(security): align the token table, App grants and environment lists with the workflows #2544: docs(security): align the token table, App grants and environment lists with the workflows
fix(ci): drain the credential tripwires' grep; spend ceiling to $3,000 #2545: fix(ci): drain the credential tripwires' grep; spend ceiling to $3,000
fix(documents): bound the contact scrub's email shape; leave Ohio State Reports citations alone #2546: fix(documents): bound the contact scrub's email shape; leave Ohio State Reports citations alone
fix(summaries): withhold staged documents the contact scrub left residuals on #2547: fix(summaries): withhold staged documents the contact scrub left residuals on
test(ci): cut the gate's slowest tests without dropping coverage #2548: test(ci): cut the gate's slowest tests without dropping coverage
test(ci): build the CLI's Click tree once per test process #2549: test(ci): build the CLI's Click tree once per test process
feat(process-version): proc-v10 — codex pair → gpt-6.1-sol, gemini pair → gemini-4-argon #2550: feat(process-version): proc-v10 — codex pair → gpt-6.1-sol, gemini pair → gemini-4-argon
config(review-hold): auto-release budget to $2,500 #2556: config(review-hold): auto-release budget to $2,500
docs(security): the production read-write role's trust names corpus-write alone #2557: docs(security): the production read-write role's trust names corpus-write alone
fix(sync-staging): arm auto-merge through GraphQL with a bounded retry #2558: fix(sync-staging): arm auto-merge through GraphQL with a bounded retry
fix(live): an ancillary paper's distribution never opens the distribution moment #2559: fix(live): an ancillary paper's distribution never opens the distribution moment
feat(release): commit a release manifest the site reads from main #2560: feat(release): commit a release manifest the site reads from main
feat(registry-gate): refuse a batch that changes a row main has stamped #2561: feat(registry-gate): refuse a batch that changes a row main has stamped
fix(salience): free an interim reserve slot whose occupant's docket reads disposed #2562: fix(salience): free an interim reserve slot whose occupant's docket reads disposed
feat(summaries): withdraw summaries of a re-docketed number's earlier petition #2563: feat(summaries): withdraw summaries of a re-docketed number's earlier petition
fix(salience): hold one interim reserve slot per application number #2564: fix(salience): hold one interim reserve slot per application number
feat(set-aside): set aside scotus/9026000455 under a new pre-distribution-mint reason #2565: feat(set-aside): set aside scotus/9026000455 under a new pre-distribution-mint reason
fix(evaluate): move a judge's misfiled retrieval.md/tooling.json to the cell level at stamp time #2567: fix(evaluate): move a judge's misfiled retrieval.md/tooling.json to the cell level at stamp time
feat(schedule): publish each pending public docket's next conference #2568: feat(schedule): publish each pending public docket's next conference
feat(votes): derive merits lineups for every recorded decision from the corpus #2569: feat(votes): derive merits lineups for every recorded decision from the corpus
feat(corpus): a remote-writes switch only the corpus writer jobs set #2570: feat(corpus): a remote-writes switch only the corpus writer jobs set
fix(judgment): read summary rulings, Rule 46 decrees and per-docket clauses #2572: fix(judgment): read summary rulings, Rule 46 decrees and per-docket clauses
fix(cert): refuse partial and conditional exits, read plural grants, restore wrongly recorded exits #2573: fix(cert): refuse partial and conditional exits, read plural grants, restore wrongly recorded exits
ci(run-repair): give the corpus job's dry run the read-only role and no write credential #2574: ci(run-repair): give the corpus job's dry run the read-only role and no write credential
feat(stamp): audit a predict cell's logged calls for reads and writes outside its lane #2575: feat(stamp): audit a predict cell's logged calls for reads and writes outside its lane
ci: measure coverage on pushes to main only, not on every PR run #2576: ci: measure coverage on pushes to main only, not on every PR run
docs: map document provisioning; record the OCR sweep carve-out as decided #2580: docs: map document provisioning; record the OCR sweep carve-out as decided
feat(cert): read probable jurisdiction noted as a direct-appeal grant #2582: feat(cert): read probable jurisdiction noted as a direct-appeal grant
feat(public-update): a write-once weekly update history under metrics/updates for the site #2583: feat(public-update): a write-once weekly update history under metrics/updates for the site
ci: land data-run PRs as merge commits under their PR titles #2587: ci: land data-run PRs as merge commits under their PR titles
fix(registry-gate): digest a candidate with its own engine defaults #2589: fix(registry-gate): digest a candidate with its own engine defaults
Sync: chore: sync main into staging #2551, chore: sync main into staging #2584

Effect checks

#2546: fix(documents): bound the contact scrub's email shape; leave Ohio State Reports citations alone

uv run pytest tests/test_documents.py -k "ohio or linear_time or not_a_contact or mailing_address" is green, and uv run fedcourts process-digest --all prints the same label and six digests at the carrying merge as at its first parent.

#2547: fix(summaries): withhold staged documents the contact scrub left residuals on

On the next summarize run after promotion, the stage job's summary line reads stage check: … kept, … removed, M document(s) withheld. A case with residual-flagged documents shows ::warning::removed scotus/<docket>/<kind> from the staged records: document withheld: <n> contact residual(s). uv run pytest tests/test_summaries.py -k stage is green.

#2550: feat(process-version): proc-v10 — codex pair → gpt-6.1-sol, gemini pair → gemini-4-argon

After the carrying promotion, the next run-predict codex and gemini cells carry process_version.label: proc-v10 with digests 7ab92dd4… / bee46e92…. Their usage.json records "model": "gpt-6.1-sol" / "gemini-4-argon", with cost lines at the new rates.

To check:

git grep -l '"model": "gpt-6.1-sol"' origin/main -- data/cases | wc -l is greater than 0, and the same for gemini-4-argon.

Then re-run uv run fedcourts successor-disclosures --closed proc-v8 --successor proc-v10 for the dated follow-up entry.

#2557: docs(security): the production read-write role's trust names corpus-write alone

None at run time; this is documentation. The IAM change itself is checked by the next scheduled run-pull window succeeding under corpus-write.

#2587: ci: land data-run PRs as merge commits under their PR titles

After promotion, the first data-run PR merged by its job lands as a merge commit with the PR title as its subject:
git log origin/main --first-parent --merges --format='%s' -5
This should show predict: N prediction(s) (run …) (#N)-style subjects. It should not show Merge pull request #N from ModelMirrorAI/run/… for job-merged PRs.

No effect-check note found. Add one to the feature PR's description (the next refresh picks it up) or write it below the end marker, or say there why none applies:

docs(security): record the case-sensitive ruleset patterns and the case-variant guard #2543: docs(security): record the case-sensitive ruleset patterns and the case-variant guard
docs(security): align the token table, App grants and environment lists with the workflows #2544: docs(security): align the token table, App grants and environment lists with the workflows
fix(ci): drain the credential tripwires' grep; spend ceiling to $3,000 #2545: fix(ci): drain the credential tripwires' grep; spend ceiling to $3,000
test(ci): cut the gate's slowest tests without dropping coverage #2548: test(ci): cut the gate's slowest tests without dropping coverage
test(ci): build the CLI's Click tree once per test process #2549: test(ci): build the CLI's Click tree once per test process
config(review-hold): auto-release budget to $2,500 #2556: config(review-hold): auto-release budget to $2,500
fix(sync-staging): arm auto-merge through GraphQL with a bounded retry #2558: fix(sync-staging): arm auto-merge through GraphQL with a bounded retry
fix(live): an ancillary paper's distribution never opens the distribution moment #2559: fix(live): an ancillary paper's distribution never opens the distribution moment
feat(release): commit a release manifest the site reads from main #2560: feat(release): commit a release manifest the site reads from main
feat(registry-gate): refuse a batch that changes a row main has stamped #2561: feat(registry-gate): refuse a batch that changes a row main has stamped
fix(salience): free an interim reserve slot whose occupant's docket reads disposed #2562: fix(salience): free an interim reserve slot whose occupant's docket reads disposed
feat(summaries): withdraw summaries of a re-docketed number's earlier petition #2563: feat(summaries): withdraw summaries of a re-docketed number's earlier petition
fix(salience): hold one interim reserve slot per application number #2564: fix(salience): hold one interim reserve slot per application number
feat(set-aside): set aside scotus/9026000455 under a new pre-distribution-mint reason #2565: feat(set-aside): set aside scotus/9026000455 under a new pre-distribution-mint reason
fix(evaluate): move a judge's misfiled retrieval.md/tooling.json to the cell level at stamp time #2567: fix(evaluate): move a judge's misfiled retrieval.md/tooling.json to the cell level at stamp time
feat(schedule): publish each pending public docket's next conference #2568: feat(schedule): publish each pending public docket's next conference
feat(votes): derive merits lineups for every recorded decision from the corpus #2569: feat(votes): derive merits lineups for every recorded decision from the corpus
feat(corpus): a remote-writes switch only the corpus writer jobs set #2570: feat(corpus): a remote-writes switch only the corpus writer jobs set
fix(judgment): read summary rulings, Rule 46 decrees and per-docket clauses #2572: fix(judgment): read summary rulings, Rule 46 decrees and per-docket clauses
fix(cert): refuse partial and conditional exits, read plural grants, restore wrongly recorded exits #2573: fix(cert): refuse partial and conditional exits, read plural grants, restore wrongly recorded exits
ci(run-repair): give the corpus job's dry run the read-only role and no write credential #2574: ci(run-repair): give the corpus job's dry run the read-only role and no write credential
feat(stamp): audit a predict cell's logged calls for reads and writes outside its lane #2575: feat(stamp): audit a predict cell's logged calls for reads and writes outside its lane
ci: measure coverage on pushes to main only, not on every PR run #2576: ci: measure coverage on pushes to main only, not on every PR run
docs: map document provisioning; record the OCR sweep carve-out as decided #2580: docs: map document provisioning; record the OCR sweep carve-out as decided
feat(cert): read probable jurisdiction noted as a direct-appeal grant #2582: feat(cert): read probable jurisdiction noted as a direct-appeal grant
feat(public-update): a write-once weekly update history under metrics/updates for the site #2583: feat(public-update): a write-once weekly update history under metrics/updates for the site
fix(registry-gate): digest a candidate with its own engine defaults #2589: fix(registry-gate): digest a candidate with its own engine defaults

After the merge

Tag promotion/<YYYY-MM-DD> on the merge commit (-2 for a same-day second batch).
Fill the freeze-record placeholders this batch owes:
- <FILL: census evaluations digest=a8610eec,6a88d3dc>
- <FILL: census predictions digest=7ab92dd4,bee46e92>
- <FILL: merge sha>
- <FILL: merged-at>
- <FILL: promotion tag>
Run the effect checks above and record what they printed.

Re-run promotion-gate: the integration gate suite went green at 9bb5265 in run 39440437340 (15:00Z), after the check last ran.

Generated by Claude Code

…se-variant guard Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019NtSiNeMSRGjEKSx7yTfdu

docs(security): record the case-sensitive ruleset patterns and the case-variant guard

…ts with the workflows Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019NtSiNeMSRGjEKSx7yTfdu

docs(security): align the token table, App grants and environment lists with the workflows

Under pipefail, `grep -q` exits on its first match without draining its input, so the upstream writer can take SIGPIPE and the pipeline reads as no-match. The 13 `env | grep -E '^AWS_' | grep -qvE ...` tripwires could therefore pass on a leaked credential when the environment is large. They now drain with `grep -vE ... > /dev/null`, which keeps the exit status. The engine watchdog's piped `grep -Eq` takes a here-string instead, and a new test refuses a piped quiet grep anywhere in the workflows, actions and scripts. spend.ceiling_usd goes from $4,000 to $3,000 (the derived $2,500 plus $500 of headroom), with the comments in tracking.yaml, review-hold.yaml and docs/pipeline.md brought in step. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LCyqNnoryARN5qnhghLJDd

…State Reports citations alone The plain email pattern ran unbounded local-part and label runs, so a long run of address characters with an @ and no domain behind it took quadratic time (about 29 s at the 150,000-character document cap). Every run is now bounded at the RFC limits and the domain at nine labels. The street shape and the residual detector read '123 Ohio St. 3d 456' as an address; a negative lookahead on the final name token leaves the reporter citation (optional 2d/3d series and a 1-4 digit page) alone while an Ohio Street address is still withheld. Registers the scrub-ohio-reporter-citation harness boundary pending, with its freeze-record entry. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TEb9wVQZk1u4uf8Stwn1kq

…duals on summary-stage-check now removes, from a kept case, every staged document whose manifest row carries a non-zero (or non-integer) contact_residuals count, together with its manifest row, so a contact detail the scrub missed on a filing's opening or closing does not reach the public one-day summary-stage artifact. The rule keys on the residual count, not contact_scrubbed: false, which every represented docket's documents also read. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TEb9wVQZk1u4uf8Stwn1kq

…s full effect Review round: a series-less Ohio St. citation now needs a pin cite or a parenthesis after its page, and a series one a page standing alone and not followed by a capitalized word, so '350 Ohio St. 200', '350 Ohio St. 4B' and '1200 W Ohio St. 3d Floor' stay withheld; the Law Journal spelling is read as a citation. The freeze entry states the manifest flip, the past-bound email strings, the counting windows, the evaluator side and the unsized exposure. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TEb9wVQZk1u4uf8Stwn1kq

… documents apart Review round: a manifest row with no contact_residuals key is withheld (nothing says it is clean); a manifest kind that is not a file stem removes the case before anything is unlinked; the stage-check line counts withheld documents apart from removed cases; case-summaries.md says the summary is written from the rest of the record under the plan's digest. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TEb9wVQZk1u4uf8Stwn1kq

fix(ci): drain the credential tripwires' grep; spend ceiling to $3,000

…ir → gemini-4-argon Moves the codex and gemini engine defaults, prices both models at their published rates, registers proc-v10 pending (four digests move, the claude pair carries forward), and adds its freeze-record entry. Tests that read the committed registry now tolerate a pending label and date their fixtures off the last counting instant. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XrszNR5q9CFWAXN6Ln6A8E

The repair-parity prerequisites test re-parsed run-repair.yml ~500 times; the workflow is now parsed once per process and each pass's steps are selected once. The every-command help render is one case per command, so xdist spreads it. The ledger scan (validate data, ops-report) parses YAML with libyaml's safe loader through serialize.load_yaml and replaces Path.relative_to with the same lexical prefix test minus the parents walk. Refs #2528 Claude-Session: https://claude.ai/code/session_01FP36E1sR8BzuBwdFwEmPsM Co-authored-by: Claude <noreply@anthropic.com>

typer.testing.CliRunner.invoke converts the Typer app to its Click group on every call, which for this CLI walks all 166 commands' signatures. The tree depends only on the app's registrations, so conftest now routes the runner's conversion through a per-app cache (tests/cli_build_cache.py), with tests that the cache is installed, keyed per app, and on invoke's path. Refs #2528 Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FP36E1sR8BzuBwdFwEmPsM

Anchor the promotion-record fixtures to a fixed past instant so they pass once the merge stamps the label, let the registry goldens through later stamps without loosening known rows, narrow the pending-close relaxation, and correct the freeze-record disclosures, the spend caveat and the budget projection the reviewers flagged. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XrszNR5q9CFWAXN6Ln6A8E

test(ci): build the CLI's Click tree once per test process

chore: sync main into staging

The $2,000 budget sits under the trailing 30-day spend ($2,302 on 2026-10-19) while the long-conference campaign is still in the window, so every predict, evaluate and summarize round waits on a human release. At $2,500, $500 under the $3,000 spend ceiling, routine rounds release again. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LCyqNnoryARN5qnhghLJDd

feat(process-version): proc-v10 — codex pair → gpt-6.1-sol, gemini pair → gemini-4-argon

config(review-hold): auto-release budget to $2,500

Resolve the freeze-record append conflict (the proc-v10 entry first, then this entry) and restate the entry's digest line and counting windows against the proc-v10 label now pending on staging. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TEb9wVQZk1u4uf8Stwn1kq

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TEb9wVQZk1u4uf8Stwn1kq

…rite alone The maintainer removed environment:prod from the production read-write corpus role's trust, so SECURITY.md and docs/security.md state the narrowed trust as the current design rather than a pending IAM step; the provisioning order stays as the rule for re-provisioning. The Claude GitHub App exchange stays the stated accepted residual. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TEb9wVQZk1u4uf8Stwn1kq

Review round: security.md's staging comparison, its runbook step and its role table, data-pipeline.md's credential-split note, and the comments in run-predict, run-evaluate, run-pull and staging-corpus-refresh no longer describe the read-write role's narrowing to corpus-write as a future step. The re-provisioning order names the subject it keeps and what skipping its last step leaves open. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TEb9wVQZk1u4uf8Stwn1kq

…ail-bound-ohio-cite fix(documents): bound the contact scrub's email shape; leave Ohio State Reports citations alone

…op-residuals fix(summaries): withhold staged documents the contact scrub left residuals on

…t-app-residual docs(security): the production read-write role's trust names corpus-write alone

`gh pr merge --auto` ran seconds after `gh pr create`, while the PR's required checks were queued but unreported; gh read the PR as mergeable, called mergePullRequest instead of arming, and the ruleset refused. Arm with enablePullRequestAutoMerge(MERGE), fall back to a direct merge, retry for a few minutes until the checks register, and fail the run when neither takes. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DfjaLBEcsn9UBUSe33fpwE

…PR body wording Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DfjaLBEcsn9UBUSe33fpwE

…er's current petition A summary filed under a docket id whose number the Clerk re-docketed describes the earlier petition. summarize-plan now lists every committed summary whose snapshot day precedes the newest record's DocketedDate as withdrawn, compares the case against its newest summary left, and the publish job deletes the withdrawn files on the reviewed refresh PR, the jail admitting only those deletions. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DfjaLBEcsn9UBUSe33fpwE

Step 0 gains check 8: every label, blessing, window and boundary stamped on main reaches the merge field for field and in main's order, the merge-stamping declaration stays put, and evaluator blessing labels only grow by a label new to main. The comparison is three-way against every merge base, so a stamp not yet synced back passes. register-process's pending close and blessing drop pass, and so does a revocation of a closed window dated at or after the head. The golden harness boundaries now pin the five rows stamped on 2026-10-16 and 2026-10-18. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TEb9wVQZk1u4uf8Stwn1kq

…pdates for the site run-analytics gains a public-updates mode and a Monday 12:57 UTC cron that adds each missing week's public-update render to metrics/updates/<week>.md (fedcourts archive-public-updates) and lands it as an unattended PR on metrics/public-updates, bounded by a new add-only path jail (assert-update-paths) that the required paths check runs. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01EzhLezuGW2VqKEbvVCjtsx

Refuse an incomplete --through week, validate week numbers in the jail, collapse caption whitespace, fence metrics/updates out of the back-test replay, move the cron to 12:59, guard git add on an absent directory, and bring the docs and isolation tests in step. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01EzhLezuGW2VqKEbvVCjtsx

chore: sync main into staging

feat(public-update): a write-once weekly update history under metrics/updates for the site

The unattended data-run PRs to main (collect, the big-case board, the update history) switch from squash to merge commits, so every merge on both branches is a merge commit and main's rulesets can drop squash once this is promoted. Each merge commit takes "<PR title> (#N)" as its subject, on merge-when-green.sh's own merge and on the auto-merge arm, so main's first-parent log reads one line per run and keeps the auto-released mark the review hold counts its daily cap from, dated when the PR lands. Tests pin the counter against merge-commit history (merge commit and ready commit both reachable, either alone in the window, the platform's default subject), the merge subject and method, and the arm string. The promotion-prep walk for the last promotion merge is sized for the data merges it now passes. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013fZbpKTGZGjiioGWJu3ZzR

ci: land data-run PRs as merge commits under their PR titles

The promotion train runs the base's code, and the registry gate copied only the candidate's config and prompts, taking pricing.DEFAULT_MODELS and ENGINE_RETRIEVAL from the running package. A batch that bumps an engine default was therefore digested with the old model and refused. Parse both literals from the candidate's source (never importing it) and thread them through digest_for_actor; refuse a source the gate cannot read as exactly one literal. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GCZXYYDuAk85KKTdV8eMBA

…2589)

github-actions · 2026-10-20T15:26:02Z

promotion-record — `7d4f2c6` (#2590): success

Run: https://github.com/ModelMirrorAI/fedcourtsai/actions/runs/39448152455

tag: promotion/2026-10-20, prereg/proc-v10
filled: 7 placeholder(s); stamped: 13 registration(s)
stale placeholders an earlier merge carried (not filled):
- line 10886: refresh commit
- line 10887: ot2026-cell pool
- line 10888: ot2025-cell pool

claude and others added 30 commits October 18, 2026 18:33

docs(security): record the case-sensitive ruleset patterns and the ca…

be5cccb

…se-variant guard Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019NtSiNeMSRGjEKSx7yTfdu

Merge pull request #2543 from ModelMirrorAI/docs/ruleset-case-variants

7bf0726

docs(security): record the case-sensitive ruleset patterns and the case-variant guard

docs(security): align the token table, App grants and environment lis…

4b16c0d

…ts with the workflows Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019NtSiNeMSRGjEKSx7yTfdu

Merge pull request #2544 from ModelMirrorAI/docs/security-token-table

e5a6822

docs(security): align the token table, App grants and environment lists with the workflows

Merge pull request #2545 from ModelMirrorAI/claude/project-thread-cohl1p

881eac3

fix(ci): drain the credential tripwires' grep; spend ceiling to $3,000

Merge pull request #2549 from ModelMirrorAI/claude/project-thread-ytom7o

f6aec3d

test(ci): build the CLI's Click tree once per test process

Merge pull request #2551 from ModelMirrorAI/main

4405bc4

chore: sync main into staging

Merge pull request #2550 from ModelMirrorAI/claude/project-thread-nxtzac

6940b11

feat(process-version): proc-v10 — codex pair → gpt-6.1-sol, gemini pair → gemini-4-argon

Merge pull request #2556 from ModelMirrorAI/claude/project-thread-cohl1p

e7a7e57

config(review-hold): auto-release budget to $2,500

Merge origin/staging into fix/2512-summary-stage-drop-residuals

71dfa3a

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01TEb9wVQZk1u4uf8Stwn1kq

Merge pull request #2546 from ModelMirrorAI/fix/2513-contact-scrub-em…

fdc9f95

…ail-bound-ohio-cite fix(documents): bound the contact scrub's email shape; leave Ohio State Reports citations alone

Merge pull request #2547 from ModelMirrorAI/fix/2512-summary-stage-dr…

fc59c38

…op-residuals fix(summaries): withhold staged documents the contact scrub left residuals on

Merge pull request #2557 from ModelMirrorAI/docs/2511-write-role-trus…

bc8ecd5

…t-app-residual docs(security): the production read-write role's trust names corpus-write alone

fix(sync-staging): report a direct merge of an open sync PR; neutral …

61348eb

…PR body wording Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DfjaLBEcsn9UBUSe33fpwE

claude and others added 8 commits October 19, 2026 22:58

Merge pull request #2584 from ModelMirrorAI/main

1517420

chore: sync main into staging

Merge pull request #2583 from ModelMirrorAI/claude/project-thread-jjkl5m

fe84bff

feat(public-update): a write-once weekly update history under metrics/updates for the site

Merge pull request #2587 from ModelMirrorAI/claude/project-thread-4xql86

58c6d0a

ci: land data-run PRs as merge commits under their PR titles

fix(registry-gate): digest a candidate with its own engine defaults (#…

9bb5265

…2589)

modelmirror deployed to staging October 20, 2026 14:34 — with GitHub Actions Active

modelmirror marked this pull request as ready for review October 20, 2026 15:18

modelmirror requested a review from a team October 20, 2026 15:18

modelmirror self-requested a review as a code owner October 20, 2026 15:18

modelmirror approved these changes Oct 20, 2026

View reviewed changes

modelmirror enabled auto-merge October 20, 2026 15:18

modelmirror merged commit 7d4f2c6 into main Oct 20, 2026
34 of 36 checks passed

modelmirror deleted the promote/2026-10-20-2 branch October 20, 2026 15:24

This was referenced Oct 20, 2026

codex pair → GPT-6.1 Sol: predict/evaluate model, pricing and re-bless (proc-v10 candidate) #2386

Closed

gemini pair → Gemini 4 Argon once it is served in the API (proc-v9 candidate) #2193

Closed

Navigation Menu

Repository navigation

promote: staging → main (2026-10-20-2) — 9bb5265ae - #2590

promote: staging → main (2026-10-20-2) — 9bb5265ae#2590
modelmirror merged 79 commits into
mainfrom
promote/2026-10-20-2

fedcourtsai-dev Bot commented Oct 20, 2026 •

edited by modelmirror-agent

Loading

Uh oh!

Uh oh!

github-actions Bot commented Oct 20, 2026

Uh oh!

This branch was successfully deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

fedcourtsai-dev Bot commented Oct 20, 2026 • edited by modelmirror-agent Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

The batch

Effect checks

#2546: fix(documents): bound the contact scrub's email shape; leave Ohio State Reports citations alone

#2547: fix(summaries): withhold staged documents the contact scrub left residuals on

#2550: feat(process-version): proc-v10 — codex pair → gpt-6.1-sol, gemini pair → gemini-4-argon

#2557: docs(security): the production read-write role's trust names corpus-write alone

#2587: ci: land data-run PRs as merge commits under their PR titles

After the merge

Uh oh!

Uh oh!

github-actions Bot commented Oct 20, 2026

promotion-record — 7d4f2c6 (#2590): success

Uh oh!

This branch was successfully deployed

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

fedcourtsai-dev Bot commented Oct 20, 2026 •

edited by modelmirror-agent

Loading

promotion-record — `7d4f2c6` (#2590): success