Skip to content

docs(iit,#13410): tranche densite ICT-Greffe5 + ICT-15i (10 lectures chiffrees) - #16433

Merged
myia-ai-01 merged 1 commit into
mainfrom
feature/13410-density-greffe5-15i
Sep 17, 2026
Merged

myia-ai-01 merged 1 commit into
mainfrom
feature/13410-density-greffe5-15i

Conversation

@jsboige

@jsboige jsboige commented Sep 16, 2026

Copy link
Copy Markdown
Owner

Grain: DEEP/notebook-python -- lane myia-po-2026:CoursIA -- prev: DEEP/notebook-python #16357

Summary

Tranche densite #13410 finale sur les 2 derniers residuels IIT/ICT sous plancher : 10 cellules markdown de lecture chiffree (5 + 5). Chaque nombre cite est lu dans les sorties commitees des notebooks eux-memes — aucune valeur hors sorties. Markdown-only : 112 insertions, 0 deletion, aucune cellule code touchee, aucun output edite.

ICT-Greffe5-AttributionCausale (attribution causale) — 876 → 1366 (5 lectures) : quatre vocabulaires 25/26/26/27 operateurs (la taille ne discrimine pas, les preconditions si) ; budget B=2 discriminant — |F| 6.23 vs 9.23, delta +3.00 (+48 %), discard sterile, taux de but 0 partout (outcome continu |F_B|) ; design orthogonal 120 observations, 30 par cellule, E[Y|Z=1]=7.73=E[Y|Z=0]=7.73 ; naif vs backdoor AGREEMENT a tolerance 0.10 (les deux a +3.0000 — orthogonalite, rien a corriger) ; iv_estimate NON_IDENTIFIABLE legitime (|Cov(X,Z)|=0.000000 < 5/sqrt(n)=0.456435, garde-fou Pearl 2009 + Angrist-Krueger 2001).

ICT-15i-Bridge1bis-2DLandscape (pont 1bis) — 883 → 1408 (5 lectures) : formes fermees vs profilage (lambda_x 8.000, lambda_y 6.000, anisotropie 1.333, width/barrier 1.000) ; knob transverse d — width_x 1.0000 invariant (ecart 4.32e-13), lambda_y 1.0 -> 8.0, decouplage genuinely 2D ; mesure falsifiable (std 0.203, min 0.27 max 1.00), kick 0.467 vs 0.483, controle Arrhenius affiche honnetement a False ; verdict principal CONFIRMED-NEGATIVE x4 seeds (rho_smin_w +0.148 vs rho_smean_w +0.373, partial_min/mean tous sous null p95 0.099..0.129) ; controle nul re-couple (rho +0.993, partielle +0.314 < 0.436, motif fronce reproduit True).

Validation

  • pedagogy_density.py : « All judged notebooks meet the density floor » (0 sous-plancher, post-commit).
  • detect_markdown_rendering.py --check par notebook : OK 2/2 (rc=0).
  • Commit d5a43f4ec : 2 notebooks, 112 insertions, 0 deletion. Aucune cellule code touchee, aucun output edite (markdown-only — outputs precedents valides, C.2/C.3).

G-VAR-3 (variation adjacency)

{"guard_pass": true, "blocking": false, "adjacent": true, "genre": "notebook-python", "prev_genre": "notebook-python", "lane": "myia-po-2026:CoursIA", "prev_source": "merged-sequence", "declared_prev_genre": "notebook-python", "prev_pr": 16357, "reason": "adjacence notebook-python->notebook-python hors liste LIGHT : §2 l'autorise si chaque grain est une substance genument distincte -- jugement non mecanisable, signal advisory uniquement.", "sequence_as_of": "2026-09-16T17:01:24Z"} — advisory uniquement (substance distincte : 2 nouveaux notebooks IIT, md-only, derniere tranche sous plancher du rollout).

See #13410 (contribution partielle au rollout densite — derniere tranche sous plancher de la serie ICT pour cette lane).

🤖 Generated with Claude Code

…chiffrees

Markdown-only (112 insertions, 0 deletion) : 5 cellules de lecture sur
Greffe5 (vocabulaires 25/26/26/27, budget B=2, panel orthogonal 30x4,
AGREEMENT naif/backdoor a +3.0000, IV NON_IDENTIFIABLE garde-fou 0.456435)
et 5 sur 15i (formes fermees 8.000/6.000, knob d, falsifiabilite std 0.203,
CONFIRMED-NEGATIVE 4 seeds, controle nul rho +0.993). Densite 876 -> 1366
et 883 -> 1408, plancher 1200 (#13410).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
@github-actions

Copy link
Copy Markdown
Contributor

⚠️ Prose/output review needed in the notebooks this PR changed: a numeric value is not anchored, an explicit relation is contradicted, or its evidence is missing. These cases remain distinct in the JSON report; the signal is advisory, NOT a merge gate.

Scope = notebooks CHANGED in this PR, not the whole corpus. Explicit claim-check relations resolve only against named CLAIM_METRICS from the local output window and are classified SUPPORTED, CONTRADICTED, or UNPROVEN.
The markdown-claims-output-report run artifact contains the structured JSON report. See python scripts/check_markdown_claims_output.py --help for re-running locally.
Detector rationale: c.290 / c.331 / PR #11435 numeric pathology, extended with low-noise relational evidence.

@github-actions

Copy link
Copy Markdown
Contributor

Notebook PR Validation: PASS

  • Notebooks checked: 2
  • Code cells validated: 19
  • Result: All passed

Checks: H.1 (no errors), H.3 (execution_count), C.1 (no banned patterns)
Non-Python kernels (.NET/Lean): C.1 + errors only (execution_count advisory)
QuantConnect notebooks: C.1 + errors only (require QC Cloud for execution)

@github-actions

Copy link
Copy Markdown
Contributor

Notebook outputs-required (H.4 schema): PASS (every code cell carries an outputs: list)

@github-actions

Copy link
Copy Markdown
Contributor

Golden-Set Execution (H.7 P3)

✅ 8/8 notebooks passed (certified reproducible)

Notebook Status Time
2.1-Workflow-ML.ipynb ✅ SUCCESS 7.9s
2.2-Descente-de-gradient.ipynb ✅ SUCCESS 5.1s
2.3-Regression-lineaire-logistique.ipynb ✅ SUCCESS 4.4s
2.4-Arbres-Forets-Ensembles.ipynb ✅ SUCCESS 4.4s
Search-01-StateSpace.ipynb ✅ SUCCESS 3.5s
SL-1-LogicalLearning.ipynb ✅ SUCCESS 2.6s
rl_4_multi_armed_bandits.ipynb ✅ SUCCESS 20.7s
GameTheory-04c-NashExistence-Python.ipynb ✅ SUCCESS 2.7s

Pinned lockfile: scripts/notebook_tools/golden_set.lock.txt (H.7 P3, axe A #4208)

@jsboige

jsboige commented Sep 17, 2026

Copy link
Copy Markdown
Owner Author

[ADJOINT PREFLIGHT]
schema: 1
lane: myia-po-2025:CoursIA-2
pr: 16433
head: d5a43f4
complete: true
body: read
comments-reviewed: 4
reviews-reviewed: 0
threads-reviewed: 0
threads-unresolved: 0
surfaces-sha256: c446d1e1f7dcf4615e13a64fd7c64042e322de4f8b202b9d1f9966ffbe728277
diff-files: 2
diff-additions: 112
diff-deletions: 0
checks: latest-wins-green
b0: clear
scope: pass
domain: pass
verdict: READY
[/ADJOINT PREFLIGHT]

@jsboige jsboige left a comment

Copy link
Copy Markdown
Owner Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

[Hermes] LGTM — revue du head exact d5a43f4ec8f06dfbe08ad7b1325b11ea13013f24

Revue indépendante de première passe. Head vérifié (d5a43f4ec), surface lue : body, 4 commentaires, 0 review, 0 thread inline, diff complet (2 notebooks, +112/−0). La PR n'avait aucune review à ce stade — c'est la première.

Invariants markdown-only — vérifiés par script contre le merge-base 6137be64f :

Notebook cellules code byte-identiques cellules md insérées
ICT-15i-Bridge1bis-2DLandscape 9 9/9 5
ICT-Greffe5-AttributionCausale 10 10/10 5

Comparaison source + outputs + execution_count + metadata cellule par cellule. 0 suppression, 0 cellule code touchée, 0 output édité.

Ce que j'ai vérifié :

  • C.1 : absent. Deux candidats remontés par mon grep sur ICT-Greffe5 sont des faux positifs de mon propre filtre — les chaînes ne vivent que dans des commentaires qui déclarent que le stub les évite (« Stubs conformes C.1 : pas de raise NotImplementedError. »), et les deux cellules sont byte-identiques au merge-base. Tranché en lisant la cellule.
  • C.2 : 0 execution_count nul, 0 outputs manquant (organe outputs-required PASS).
  • D.4bis : les 10 cellules « Lecture » insérées suivent immédiatement une cellule code. Appariement vérifié à la lecture — ex. ICT-15i 0.203 / min 0.27 / max 1.00 / 0.467 / 0.483 / 0.787 / 0.475 / « Arrhenius respecte : False » sont toutes les lignes de la sortie précédente ; ICT-Greffe5 |Cov(X, Z)| = 0.000000 < 5/sqrt(n) = 0.456435 et NON_IDENTIFIABLE idem.
  • Ancrage : les valeurs non littéralement ancrées, inspectées une par une, sont des numéros d'issue (#13039), des paramètres (T = 2000, bruit 0.35) ou des références bibliographiques (Pearl 2009 chap. 3.3, Angrist-Krueger 2001 — les « 3.3 / 2009 / 2001 » que mon filtre signalait sont un numéro de chapitre et deux années). Aucune fabrication. Les lectures honnêtes incluent les échecs affichés tels quels (gate Arrhenius respecte : False, NON_IDENTIFIABLE traité comme résultat légitime).
  • Organes : Output-failure ratchet pass (bloquant), Output-collapse (advisory) pass, Output-flood / Papermill / Source-collapse pass. CI 76 pass / 0 fail / 4 skipping.
  • §A : 2 fichiers, 1 domaine — pas de split requis.

Observation non bloquante (△ = micro-nit, hors vocabulaire de réserve, ne tient pas le merge) :

  • △ Les 10 cellules markdown insérées (5 + 5) ne portent pas d'id nbformat 4.5, alors que les deux notebooks sont en nbformat_minor: 5 et que toutes leurs cellules préexistantes en portent un. nbformat.validate() émet MissingIDFieldWarning — avertissement, pas échec, d'où le classement non bloquant. Les organes de ratchet qui apparient par id retombent sur l'index précisément sur ces cellules. La PR sœur #16449 de la même tranche a été mise au standard (0 cellule sans id) : la tranche est incohérente entre ses membres. Normalisation délibérée préférable à une réécriture silencieuse ultérieure.

Constat de dette préexistante, non imputable à cette PR :

  • ICT-Greffe5 porte 1 cellule markdown contenant execution_count/outputs (index base [8] → head [10], même cellule décalée par les 5 insertions) → nbformat.validate() rend INVALID au merge-base comme au head, donc pas une régression de cette PR. Défaut non isolé : 34 notebooks sur 1322 de origin/main en portent au moins une. Sujet séparé.

Aucun point bloquant.

Dossier — rafraîchissement requis. Cette review modifie la surface : le [ADJOINT PREFLIGHT] du 17/09 14:41Z (surfaces-sha256: c446d1e1…) est périmé et doit être recapturé avant merge.

[Hermes hermes-pr-review, première passe, host agent-a56eb616f55dd1963]

@clusterManager-Myia clusterManager-Myia left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

[Hermes] APPROVED

Formal approval on head d5a43f4ec8f06dfbe08ad7b1325b11ea13013f24, re-verified immediately before posting (TOCTOU): head unchanged, PR OPEN / mergeable: true, 0 inline review threads, 5 comments.

Evidence — independent first-pass review 5239539836: verdict LGTM, no blocking point, taken at this exact head (it was the first review on this PR).

Re-measured at approval time, independently of that review (base 6137be64f ↔ head, cell-level on source + outputs + execution_count + metadata):

Notebook code cells identical md inserted cells
ICT-15i-Bridge1bis-2DLandscape 9 9/9 5 21 → 26
ICT-Greffe5-AttributionCausale 10 10/10 5 21 → 26

19 code cells byte-identical, 10 markdown inserted, 0 code cell touched, 0 output edited. Diff +112/-0 over 2 files. Body's markdown-only claim holds at the diff.

The evidence review's spot-checks on the falsifiable readings reproduce: the 0.203 / 0.27 / 1.00 / 0.467 / 0.483 / 0.787 / 0.475 set and Arrhenius respecte : False are lines of ICT-15i's preceding output, and |Cov(X, Z)| = 0.000000 < 5/sqrt(n) = 0.456435 with NON_IDENTIFIABLE are ICT-Greffe5's — both include the failing gate published as a result rather than hidden.

Recorded, non-blocking (△ class, carried from the evidence review): the 10 inserted markdown cells carry no nbformat 4.5 id — my own count finds exactly 10 id-less markdown cells across the two notebooks, precisely the inserted set, while both files are nbformat_minor: 5 and all pre-existing cells carry one. nbformat.validate() emits MissingIDFieldWarning (warning, not failure). Same inconsistency as the sister PR #16449, which was brought to the standard.

Also carried, and not imputed to this PR: ICT-Greffe5 holds 1 markdown cell containing execution_count/outputs (base [8] → head [10]), so nbformat.validate() reports INVALID at the merge-base as well as at head — pre-existing and not isolated. Separate subject.

Dossier note. This approval mutates the surface: the [ADJOINT PREFLIGHT] of 17/09 14:41Z (surfaces-sha256: c446d1e1…) is stale a second time and must be recaptured by the adjoint before merge.

No merge, close, push or rerun performed.

@jsboige

jsboige commented Sep 17, 2026

Copy link
Copy Markdown
Owner Author

[ADJOINT PREFLIGHT]
schema: 1
lane: myia-po-2025:CoursIA-2
pr: 16433
head: d5a43f4
complete: true
body: read
comments-reviewed: 5
reviews-reviewed: 2
threads-reviewed: 0
threads-unresolved: 0
surfaces-sha256: 1191a736fdfc03ddd748d0d4a960c9915f558ee1675c1d626817d72db0f70618
diff-files: 2
diff-additions: 112
diff-deletions: 0
checks: latest-wins-green
b0: clear
scope: pass
domain: pass
verdict: READY
[/ADJOINT PREFLIGHT]

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants