Skip to content

docs(iit,#13410): densite ICT-16 (1079->1306) + ICT-17 (1163->1294) -- 5 lectures chiffrees - #16446

Merged
myia-ai-01 merged 1 commit into
mainfrom
feature/13410-density-ict16
Sep 17, 2026
Merged

myia-ai-01 merged 1 commit into
mainfrom
feature/13410-density-ict16

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

Livrable

Tranche densite #13410 : ICT-16-MDLTwoPartCode (1079 → 1306) + ICT-17-EpsilonMachine (1163 → 1294), plancher 1200. Markdown-only : +35 lignes (5 cellules md par ancres), 0 cellule code touchee, 0 output edite. Branche dediee feature/13410-density-ict16 depuis origin/main 5a1989a92, commit d0dc2905c.

Les 5 lectures (chaque nombre lu dans les sorties committes)

ICT-16 (3 lectures) :

  1. Codelengths KT (§1) : uniform −23,359 / diag asym −30,555 / eye −42,379 bits — l'ordre lu comme hierarchie de sparsite (19 bits d'economie entre uniform et eye), le signe negatif explique (likelihood marginale KT).
  2. Code deux parties (§2) : residu/n cycle 0,0836 vs markov+10 % 0,7011 vs iid 2,0528 bits (proche du maximum log2 4 = 2) — ce que le modele ne capture pas passe au residu ; la symetrie modele/residu lue sur les totaux (−6,722 bits de modele pour ~2 bits/symbole d'economie).
  3. Gate 6, tableau ligne par ligne : tau Kendall = +0,786 (p = 0,006, attendu ≥ 0,5) ; 8 trajectoires det k=2→k=5 (11,88→−29,83 bits vs zlib 136→168, k-gain jusqu'a 0,9357), markov p=.7/.5/.3 (513,85→776,41 vs 1520→2184), iid 4/8 (852,07/1216,54 vs 2312/3096, k-gain 0,0170/0,0031) — l'ordre relatif concorde, l'echelle absolue non (rang, pas bits).

ICT-17 (2 lectures) :
4. Cycle deterministe (§2) : 88 passes → 3 etats causaux, distribution stationnaire 0,341/0,33/0,33, C_mu = 1,5848 et E = 1,5850 = log2(3), serie E_k plate (1,5848 ± 0,0001 sur k = 1..4) — representation minimale et prediction complete coincident.
5. Gate 9 : plafond E = 0,5458 bit vs plug-in I = 0,5230 (ecart 0,0229) vs uniforme I = 0,0018 (ecart 0,5441) — gate PASSED, l'uniforme ne recolte que 0,3 % de l'information.

Validation

  • pedagogy_density.py <2 notebooks> → « All judged notebooks meet the density floor » (1306 / 1294).
  • detect_markdown_rendering.py --check → OK sur chacun.
  • Pre-commit : tous les hooks Passed (gitleaks, H.3, Aucun garde ne parse la source des cellules : une cellule non compilable portant une sortie traverse les 60+ checks #13326, rendering bloc).
  • Incident d'edition repris : un remplacement texte via heredoc a casse un escape JSON (\log a 1 backslash) — detecte par le guard densite (parse FAIL), repare par re-parse json.dumps integral, re-verifie (source rendue $\log_2 3$ correcte aux 2 occurrences). Lesson connue (heredoc-backslash-collapse).

Preflight

Par chemins exacts : 0 PR open sur ICT-16-MDLTwoPartCode et ICT-17-EpsilonMachine (#16278 mentionne ICT-17 en renvoi de prose mais ne touche que ICT-37/38 — verifie sur ses fichiers).

See #13410

🤖 Generated with Claude Code

…- 5 lectures chiffrees

ICT-16 : 3 lectures (codelengths KT uniform/eye/diag -23,359/-42,379/-30,555
lus en ordre de sparsite ; residu/n 0,0836 vs 0,7011 vs 2,0528 = ce que le
modele ne capture pas ; Gate 6 tau=+0,786 p=0,006, tableau 8 trajectoires
11,88->1216,54 bits vs zlib 136->3096, k-gain 0,9357->0,0031). ICT-17 :
2 lectures (cycle 0,1,2 : 3 etats causaux, C_mu=1,5848, E=1,5850=log2(3),
E_k plate ; Gate 9 : plafond E=0,5458 vs plug-in 0,5230 (ecart 0,0229) vs
uniforme 0,0018, PASSED). Md-only : 0 cellule code touchee, 0 output edite.
Densite floor post-commit : 1306 et 1297, rendering guard OK 2/2.

See #13410

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

✅ No prose/output mismatch detected in the notebooks this PR changed.

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

github-actions Bot commented Sep 16, 2026 •

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 5.2s
2.2-Descente-de-gradient.ipynb ✅ SUCCESS 6.2s
2.3-Regression-lineaire-logistique.ipynb ✅ SUCCESS 6.7s
2.4-Arbres-Forets-Ensembles.ipynb ✅ SUCCESS 5.5s
Search-01-StateSpace.ipynb ✅ SUCCESS 5.2s
SL-1-LogicalLearning.ipynb ✅ SUCCESS 2.7s
rl_4_multi_armed_bandits.ipynb ✅ SUCCESS 32.4s
GameTheory-04c-NashExistence-Python.ipynb ✅ SUCCESS 3.5s

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

@github-actions

Copy link
Copy Markdown
Contributor

Notebook PR Validation: PASS

  • Notebooks checked: 2
  • Code cells validated: 23
  • 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)

@myia-ai-01 myia-ai-01 changed the title docs(iit,#13410): densite ICT-16 (1079->1306) + ICT-17 (1163->1297) -- 5 lectures chiffrees docs(iit,#13410): densite ICT-16 (1079->1306) + ICT-17 (1163->1294) -- 5 lectures chiffrees Sep 16, 2026
@jsboige

jsboige commented Sep 17, 2026

Copy link
Copy Markdown
Owner Author

[ADJOINT PREFLIGHT]
schema: 1
lane: myia-po-2025:CoursIA-2
pr: 16446
head: d0dc290
complete: true
body: read
comments-reviewed: 4
reviews-reviewed: 0
threads-reviewed: 0
threads-unresolved: 0
surfaces-sha256: 96ef165ebe5adede088c63eb0e27a9b6eb294ffa499953831b6c74d969fa589e
diff-files: 2
diff-additions: 35
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 d0dc2905c21a193d7fbcd703832ea4fa82bf6726

Revue indépendante de première passe. Head vérifié (d0dc2905c), surface lue : body, 4 commentaires, 0 review, 0 thread inline, diff complet (2 notebooks, +35/−0). Première review de cette PR.

Invariants markdown-only — vérifiés par script contre le merge-base 5a1989a92 :

Notebook cellules code byte-identiques cellules md insérées
ICT-16-MDLTwoPartCode 12 12/12 3
ICT-17-EpsilonMachine 11 11/11 2

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 : aucun motif banni dans les 2 notebooks.
  • C.2 : 0 execution_count nul, 0 outputs manquant (organe outputs-required PASS).
  • D.4bis : les 5 cellules « Lecture » insérées (3 + 2) suivent immédiatement une cellule code. Les valeurs du body sont toutes ancrées dans la sortie de la cellule précédente : ICT-16 0,0836 / 0,7011 / 2,0528 (§2 code deux parties), tau Kendall = +0,786, k-gain 0,9357 (gate 6) ; ICT-17 0,341/0,33/0,33, C_mu = 1,5848, E = 1,5850 = log2(3), 0,5458 / 0,5230 / 0,0018 (gate 9). Le seul token non ancré est 0001 (fragment de date).
  • Exécution réelle : execution_count non nuls partout, séquences cohérentes, pas de sortie factice.
  • 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.
  • Signal honnête : le body documente un incident d'édition (escape JSON cassé par heredoc, \log à 1 backslash) détecté par le guard densité et réparé, avec re-vérification. Le signaler fait partie du contrat (la dérive a été trouvée par un organe, pas par chance).

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

  • △ Les 5 cellules markdown insérées (3 + 2) 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 sur ces cellules. La PR sœur #16449 de la même tranche a été mise au standard (0 cellule sans id) : normalisation délibérée préférable à une réécriture silencieuse ultérieure.

Aucun point bloquant. Aucune dette préexistante relevée sur ces 2 notebooks (nbformat.validate() VALID au merge-base comme au head, 0 cellule markdown porteuse d'outputs).

Dossier — rafraîchissement requis. Cette review modifie la surface : le [ADJOINT PREFLIGHT] du 17/09 15:27Z (surfaces-sha256: 96ef165e…) 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 d0dc2905c21a193d7fbcd703832ea4fa82bf6726, re-verified immediately before posting (TOCTOU): head unchanged, PR OPEN / mergeable: true, 0 inline review threads, 5 comments.

Evidence — independent first-pass review 5239540080: verdict LGTM, no blocking point, taken at this exact head (first review on this PR).

Re-measured at approval time, independently of that review (base 5a1989a92 ↔ head, cell-level on source + outputs + execution_count + metadata):

Notebook code cells identical md inserted cells
ICT-16-MDLTwoPartCode 12 12/12 3 25 → 28
ICT-17-EpsilonMachine 11 11/11 2 25 → 27

23 code cells byte-identical, 5 markdown inserted, 0 code cell touched, 0 output edited. Diff +35/-0 over 2 files. Body's markdown-only claim holds at the diff.

The evidence review's anchoring spot-checks reproduce: ICT-16's 0,0836 / 0,7011 / 2,0528, tau Kendall = +0,786, k-gain 0,9357, and ICT-17's 0,341 / 0,33, C_mu = 1,5848, E = 1,5850 = log2(3), 0,5458 / 0,5230 / 0,0018 are all values of the preceding code cells' committed outputs.

Recorded, non-blocking (△ class, carried from the evidence review): the 5 inserted markdown cells carry no nbformat 4.5 id — my own count finds exactly 5 id-less markdown cells, precisely the inserted set, while both files are nbformat_minor: 5 and all pre-existing cells carry one. MissingIDFieldWarning only; same inconsistency as sister PR #16449, which was brought to the standard.

Notable and in the PR's favour: the body documents an editing incident (JSON escape broken by heredoc, \log with a single backslash) that was detected by the density guard and repaired, with re-verification. Reporting the drift found by an organ is the behaviour the contract asks for. No pre-existing debt on these two notebooks (nbformat.validate() valid at merge-base and at head).

Dossier note. This approval mutates the surface: the [ADJOINT PREFLIGHT] of 17/09 15:27Z (surfaces-sha256: 96ef165e…) 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: 16446
head: d0dc290
complete: true
body: read
comments-reviewed: 5
reviews-reviewed: 2
threads-reviewed: 0
threads-unresolved: 0
surfaces-sha256: 5b614879ec530e7ea9a1f842bf58864d8fcd61b38e5f18c212ddcc2c1d5cfa80
diff-files: 2
diff-additions: 35
diff-deletions: 0
checks: latest-wins-green
b0: clear
scope: pass
domain: pass
verdict: READY
[/ADJOINT PREFLIGHT]

@myia-ai-01
myia-ai-01 merged commit 4e60839 into main Sep 17, 2026
99 of 100 checks passed
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