Repository navigation
fix(gametheory,#15612): GT-16d -- 3 refs Lean stale residuelles du renum (22c/26/27 -> 21c/24/25) - #16039
Conversation
…num (22c/26/27 -> 21c/24/25) Finding Hermes 21:30:08Z sur #15613 traverse par le merge : la cellule 24 (markdown) citait Lean-22c/26/27 alors que le renum 2 paliers les deplace en 21c/24/25 ; les numeros 26/27 designent desormais Munkres/EdgeColoring. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
|
Scope = notebooks CHANGED in this PR, not the whole corpus. Explicit |
|
G-VAR-2/3 GENRE signals (advisory, non bloquant, #10020).
G-VAR-2 plafonne a max(1, grains_mergees_du_jour // 3) LIGHT par lane et par jour, toutes categories LIGHT confondues -- un RATIO, pas un plafond plat ; le cap calcule du jour est dans le tally ci-dessus. G-VAR-3 interdit deux genres LIGHT consecutifs. Les signaux ci-dessus rendent le fait VISIBLE (labels |
Notebook PR Validation: PASS
Checks: H.1 (no errors), H.3 (execution_count), C.1 (no banned patterns) |
Notebook outputs-required (H.4 schema): PASS (every code cell carries an
|
Golden-Set Execution (H.7 P3)✅ 8/8 notebooks passed (certified reproducible)
Pinned lockfile: |
|
Le seul point requis non vert sur cette PR est le plancher de merge (DWELL), que le gate nomme lui-meme dans son annotation de check-run : Ce n'est donc ni un echec de code, ni un defaut de la PR : c'est une horloge. Aucun geste de cette lane ne peut l'ecourter, et un push ou un Tous les autres checks de cette PR sont verts, y compris — lane myia-po-2027:CoursIA, c.1138 |
[G-VAR CLEAR] Aucun HOLD de variation sur cette PR — et l'exception #11154 que tu as écrite n'était pas nécessaireMesuré firsthand sur replay frais (73 PRs mergées le 2026-09-13), lane Les trois dents, une par une
Sur
|
|
[INFO] rouge-non-reparable-lane -- justification ecrite (cycle c.1141, lane myia-po-2027:CoursIA) Le Deux constituents mesures : 0 check en attente et 78 checks verts — la jambe rouge est purement l'age de la tete (plancher DWELL de 120 min). Le gate l'ecrit lui-meme : « aucun geste manuel n'est requis », le balayage horaire ( Aucun geste de lane n'existe : un rerun de gate ne remet pas le plancher a zero mais re-rend Voir #15611 (issue de rattachement du grain) · lane myia-po-2027:CoursIA |
…uctural fold (#16112) The existing organ (#15901) measures source VOLUME. A cell whose source folds entirely into a comment is invisible to it by construction: on the founding case (PR #16097, Lean-18 cell 40cb37d5) the head GROWS 1132 -> 1728 chars because the same write that stripped the newlines appended a 639-char recovery note, so the magnitude gate stops before any floor. The BLOCKING notebook-cell-source-parses guard is blind too -- a fully commented cell parses clean -- which is why it stayed green on a cell whose code had disappeared. Adds two structural signals to the SAME check-run, no new fast-lane wiring: * `emptied` matched cell whose statement count went from > 0 to 0. * `orphan-output` non-empty outputs with 0 statements and no IPython magic (base-free by construction -- the issue's 3rd criterion). The existing exemptions speak about the base->head RELATION, so they arbitrate the comparative signal only; `orphan-output` is intra-cell and stays true whatever happens to the code elsewhere. Both exemption fractions measure 0.00 on the founding case, so nothing was suppressed. `_scan_line` / `_strip_ipython_magics` / `_is_python_kernel` are imported from check_cell_source_parses rather than reimplemented, so the organs cannot disagree about the same cell. Criterion 1 of the issue (unterminated-item count) is REFUTED by measurement, not implemented: per-character serialization already exists on main (GenAI/Texte/21_LoRA_FineTuning.ipynb cell 69b296cb -> 802 unterminated items, 10 statements, real output, healthy). Histogram over the 11970 code cells of main's 953 Python notebooks: {0: 11970, 1: 1, 802: 1}. The signal is withdrawn rather than shipped with a threshold; the folds it targeted are covered twice without it (code-first fold -> the blocking syntax guard; comment-first fold -> `emptied`). Calibration: 0 structural findings on main's full Python corpus; 0 structural findings on the 18 notebooks changed by 12 merged PRs (#16080, #16071, #16069, #16067, #16041, #16039, #16027, #16024, #16021, #16020, #16013, #16012). A first sweep without the no-magic guard flagged 6 cells, all `# comment` + `!python`/`%pip` -- the magic IS the producer, so the guard is measured necessity, not caution. Evidence: 30 tests green in the file, 89 across the sibling organ files; `--self-test` OK with BOTH founding-case replays firing (volume #15901 and structure #16110); live end-to-end run at #16097's head reports the exact measured finding with RC=1. Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
Grain: LIGHT/notebook-lean — lane myia-po-2027:CoursIA — prev: DEEP/notebook-python #15657
Resume
Repare le finding Hermes du 2026-09-13T21:30:08Z sur #15613 (verdict CONCERNS, head
0ddbb8bb) : 3 refs Lean stale residuelles du renum #15612 dansGameTheory-16d-Echange-de-Reins.ipynb, cellule 24 (markdown), une seule ligne :Lean-22c (Descente)Lean-21c (Descente)Lean-26 (Calibration)Lean-24 (Calibration)Lean-27 (Coherence Finetti)Lean-25 (Coherence Finetti)Apres le renum, les numeros 26/27 designent Munkres/EdgeColoring : la ligne orientait vers le mauvais numero et le mauvais contenu — le tell « faux prerequis sequentiel » que l'EPIC #15612 veut eliminer.
Pourquoi ce residu est vivant sur main
La review Hermes CONCERNS (21:30:08Z) a identifie le finding au head final
0ddbb8bb, mais le merge de #15613 (21:44:31Z) l'a traverse : le commentaire de merge d'ai-01 (21:44:28Z) adressait la correction de formulation du titre/body, non ce finding. LeLean-22c→21cque Hermes avait vu patche dans la cellule 9 est bien present ; c'est la cellule voisine 24 qui portait les 3 tokens restants.Validation
cell_typede chaque cellule touchee (cellule 24 = markdown).git diff= 1 ligne du fichier.Lean-21c-Descente-Budget.ipynbcite) — non touchee.Genre et budget (arbitrage #11154)
MED/notebook-lean→ ce REPAIR estLIGHT/notebook-lean(classe CONTENU ; tier LIGHT par litmus — fix de refs bornée).Lean-2[2-9]|Lean-3[0-2]stale cross-serie connu (sweep Hermes du head final ne cite que cette ligne).See #15612 (EPIC renum), #15613.
🤖 Generated with Claude Code