Issues

• backtest: decide whether party dispositions (dismissed/withdrawn) belong in the cert back-test set

  ModelMirrorAI/fedcourtsai#2217 · Filter by author modelmirror opened on Oct 3, 2026 · Back-test program

• runner: gemini INVALID_STREAM in a CI cell gets no in-run retry (the transient-retry set lives in the runner the cell workflows bypass)

  ModelMirrorAI/fedcourtsai#2214 · Filter by author modelmirror opened on Oct 3, 2026 · CI & engine hygiene

• documents: the brief-in-opposition arm misses 'Brief for the respondent(s)'; a lead docket's own selection can take a member's filing

  data-quality
  ModelMirrorAI/fedcourtsai#2212 · Filter by author modelmirror opened on Oct 3, 2026 · Provisioning & document coverage

• moments: a consolidated member's briefed merits moment never opens (the respondent's brief is on the lead docket)

  ModelMirrorAI/fedcourtsai#2211 · Filter by author modelmirror opened on Oct 3, 2026 · Post-conference re-bless: proc-v9 / sal-v5 candidates

• corpus: write helpers mirror to the configured content store even when the connection is a temp database

  ModelMirrorAI/fedcourtsai#2199 · Filter by author modelmirror opened on Oct 2, 2026 · CI & engine hygiene

• run-repair: retire finished passes, move recurring fills to the walker, and consolidate the rest around a pass registry

  ModelMirrorAI/fedcourtsai#2194 · Filter by author modelmirror opened on Oct 2, 2026 · CI & engine hygiene

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

  ModelMirrorAI/fedcourtsai#2193 · Filter by author modelmirror opened on Oct 2, 2026 · Post-conference re-bless: proc-v9 / sal-v5 candidates

• votes: a corpus-side target for opinion-votes, so decided cases the pipeline never forecast get lineups

  ModelMirrorAI/fedcourtsai#2191 · Filter by author modelmirror opened on Oct 2, 2026 · Analytics expansion

• schedule: export each open case's next conference (and argument / response dates) for the website timeline; track the Court calendar

  ModelMirrorAI/fedcourtsai#2190 · Filter by author modelmirror opened on Oct 2, 2026 · Analytics expansion

• predict: gemini writes an invented created_at; stamp it harness-side

  ModelMirrorAI/fedcourtsai#2188 · Filter by author modelmirror opened on Oct 2, 2026 · Agent observability

• documents: OCR recovery runs only on dispatch, so a scanned application reaches its forward cell with no text

  data-quality
  ModelMirrorAI/fedcourtsai#2187 · Filter by author modelmirror opened on Oct 2, 2026 · Provisioning & document coverage
  1

• run-repair: a compromised parse job could poison the Actions cache that credentialed writer jobs restore

  ModelMirrorAI/fedcourtsai#2179 · Filter by author modelmirror opened on Oct 1, 2026 · CI & engine hygiene

• run-repair: the shared repair job mints the App token, assumes the read-write role and takes the write lock even on dry runs (14 older passes)

  ModelMirrorAI/fedcourtsai#2173 · Filter by author modelmirror opened on Oct 1, 2026 · CI & engine hygiene
  1

• corpus: read-only commands migrate the pulled blob in place, so it stops matching its pointer (corpus-info included)

  data-quality
  ModelMirrorAI/fedcourtsai#2167 · Filter by author modelmirror opened on Oct 1, 2026 · CI & engine hygiene
  1

• backtest/query: the replay clock still passes a prior's other last-pull columns (citations, opinion, date_decided)

  data-quality
  ModelMirrorAI/fedcourtsai#2163 · Filter by author modelmirror opened on Oct 1, 2026

• merits parser: 10 decided cases stuck at judgment-issued, 4 stale grants, and consolidated dockets inheriting the lead disposition

  data-quality
  ModelMirrorAI/fedcourtsai#2161 · Filter by author modelmirror opened on Oct 1, 2026 · Analytics expansion
  1

• corpus: cert record errors on granted rows (24-813 recorded dismissed on its grant date; 25-1083/25-1084/24-109 without a disposition or grant latch)

  data-quality
  ModelMirrorAI/fedcourtsai#2160 · Filter by author modelmirror opened on Oct 1, 2026 · Provisioning & document coverage

• evaluate: judges see the predictors' big_case_score before forming their stakes read — reword the prompt and caveat the reading rule (no blinding)

  ModelMirrorAI/fedcourtsai#2154 · Filter by author modelmirror opened on Oct 1, 2026 · Post-conference re-bless: proc-v9 / sal-v5 candidates
  1

• corpus: four OT2024 application numbers are stored as two rows each, double-counting two grants in the interim base rate

  data-quality
  ModelMirrorAI/fedcourtsai#2152 · Filter by author modelmirror opened on Oct 1, 2026 · Provisioning & document coverage

• corpus: 24A885 / 24A886 (granted federal applications) classify as application_kind unknown, as do 49 other OT2024 rows

  data-quality
  ModelMirrorAI/fedcourtsai#2151 · Filter by author modelmirror opened on Oct 1, 2026 · Provisioning & document coverage

• statpack: the interim anchor has no applicant-class cut (e.g. state applications to vacate stays of execution)

  ModelMirrorAI/fedcourtsai#2150 · Filter by author modelmirror opened on Oct 1, 2026 · Analytics expansion

• events: committed event.yaml files that predate moments drift from the corpus projection, and cells see conflicting dates

  ModelMirrorAI/fedcourtsai#2149 · Filter by author modelmirror opened on Oct 1, 2026 · CI & engine hygiene

• evaluate: gemini-judge writes its judge-level retrieval.md / tooling.json into each candidate directory, and nothing catches it

  ModelMirrorAI/fedcourtsai#2148 · Filter by author modelmirror opened on Oct 1, 2026 · Agent observability

• stat pack: historical cert-stage completeness back to OT2015

  ModelMirrorAI/fedcourtsai#2143 · Filter by author modelmirror opened on Sep 30, 2026 · Analytics expansion
  1

• stat pack: interim docket completeness, OT2015–OT2025 (every application, category, applicant type, noted votes)

  ModelMirrorAI/fedcourtsai#2142 · Filter by author modelmirror opened on Sep 30, 2026 · Analytics expansion
  1