marketing: DMLDiD launch carousel (Circuit palette) + claims-sync guard - #802
Merged
Conversation
11-slide poster-mode deck for the v3.11 DMLDiD launch (ML-forward hook; user decisions 2026-08-29): cover curve-vs-line motif, the confidently-wrong receipt (linear 2.5909, reported SE 0.0743, vs true 2.2388 - nominal-inference caveats slide-local), dark ridge twist, Chang (2020) paper slide (attributed paraphrase, no verbatim pull quote - the published PDF was not cross-checked word-for-word), annotated Case-1 score slide (the UNCENTERED summand s labeled as such, centered score psi = s - ATT in the caption, orthogonality scoped to the learned nuisances g and l, arrows pixel-anchored via manual fraction composition - mathtext bboxes carry asymmetric slop), four-learner payoff chart with the double-robustness beat, honest DMLDiD-vs-CS decision split plus the learner-flexibility band (four built-ins with no extra installs; scikit-learn estimators fit the duck-typed fit()/predict() contract), dark code slide with a real sklearn swap line, production grid (aggregation/HonestDiD/sup-t, survey + clustering with the A2.3 same-target-population qualifier on the RCS card, diagnostics) with the DoubleML Case-1 parity strip (ATT diff 4.4e-16, committed spike), CTA with the single tutorial-32 mention. Every estimate is a seed-locked tutorial-32 value from the simulated truth-known example (labeled on-slide). tests/test_dml_carousel_claims.py (17 checks): notebook-surface sync with row-associated learner-table pins, arithmetic backing for the five-SE and recovery claims, constants-reach-their-slides + qualifier slide-locality enforcement, banned-absolutes list (including the retired overclaims: "in each nuisance", "confident CI"), REGISTRY sync for the DoubleML figure, dependency-free proof for the built-in-learner claim, and the documented-sklearn-contract check.
Overall Assessment: ✅ Looks goodExecutive Summary
MethodologyP3 — Documented extensions correctly scoped
Code QualityNo findings. PerformanceNo findings. MaintainabilityNo findings. Tech DebtP3 — PDF/source drift guard remains deferred
SecurityNo findings. Documentation/TestsP2 — Learner contract copy omits
P2 — Cohort synchronization assertion is effectively vacuous
|
…onstruction pin PR #802 review P2s: (1) slide 8 said any learner needs fit()/predict(), but propensity learners require predict_proba() - the card and the flexibility band now state both contracts (regressors fit()/predict(), classifiers predict_proba()), with slide-local pins; (2) the cohort sync assertion's str(g)-in-code fallback was vacuous (single digits occur everywhere) - it now requires the exact np.where(early, 4, 5) construction rendered from the deck constants. PDF regenerated; claims guard 17/17 green.
|
🔁 AI review rerun (requested by @igerber) Head SHA: Overall Assessment:
|
…t docstring guard PR #802 review round 2 P1: the slide fix missed the module docstring, which still described a bare fit()/predict() scikit-learn contract - propensity classifiers need predict_proba() (a LinearSVC would be rejected). The spine bullet and claim-discipline bullet now state both contracts, and a new ast.get_docstring test requires predict_proba() there (the slide-string guard deliberately excludes docstrings, so the docstring needs its own pin). PDF regenerated (byte-freshness only; slide content unchanged); claims guard 18/18 green.
|
🔁 AI review rerun (requested by @igerber) Head SHA: Overall Assessment: ✅ Looks goodExecutive Summary
MethodologyP3 — Documented extensions are correctly scoped
Code QualityNo findings. PerformanceNo findings. MaintainabilityNo findings. Tech DebtP3 — PDF/source drift guard is tracked
SecurityNo findings. Documentation/TestsP2 — Slide 9’s displayed fit call is invalid Python
The previous learner-contract P1 is resolved at |
…e slide PR #802 review round 3 P2: the displayed call used a positional ellipsis after keyword arguments (a SyntaxError) under a caption claiming the shown call reproduces the shown numbers. The slide now renders the tutorial's full invocation (unit/time/first_treat keywords + base_period=universal) in a slightly wider panel, and a new claims check reconstructs the snippet from the slide's token lines via AST and requires it to ast.parse, pinning base_period and the fit keywords. PDF regenerated; claims guard 19/19 green.
|
🔁 AI review rerun (requested by @igerber) Head SHA: Overall Assessment: ✅ Looks goodExecutive Summary
MethodologyP3 — Documented extensions are correctly scoped
Code QualityNo findings. PerformanceNo findings. MaintainabilityNo findings. Tech DebtP3 — PDF/source drift guard remains tracked
SecurityNo findings. Documentation/TestsP3 — Prior slide-9 finding resolved
The environment lacked |
…outs CI failure on every Python leg: test_module_docstring_states_both_contracts read the generator without the GENERATOR.exists() guard the rest of the file carries, and the isolated-install CI jobs copy only tests/ (no carousel/). Verified both layouts: full checkout 19 passed; a simulated tests-only layout 19 skipped.
|
🔁 AI review rerun (requested by @igerber) Head SHA: Overall Assessment: ✅ Looks goodExecutive Summary
MethodologyP3 — Documented extensions correctly scoped
Code QualityNo findings. PerformanceNo findings. MaintainabilityNo findings. Tech DebtP3 — PDF/source drift guard remains tracked
SecurityNo findings. No apparent secrets or unsafe file handling were introduced. Documentation/TestsP2 — CTA overstates Tutorial 32’s provenance
P3 — Prior slide-9 finding resolved
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
carousel/generate_dml_carousel.py+ committedcarousel/diff-diff-dml-carousel.pdf(11 slides, poster mode, "Circuit" palette — graphite structure with a single signal-green ML/truth accent).slabeled as such withpsi = s - ATTin the caption; orthogonality scoped to the learned nuisances g and ℓ; annotation arrows pixel-anchored via manual fraction composition), four-learner payoff chart with the double-robustness beat, honest DMLDiD-vs-CS decision split plus the learner-flexibility band (four built-ins, no extra installs; scikit-learn estimators fit the duck-typed contract), dark code slide with a real sklearn swap line, production grid with the DoubleML Case-1 parity strip and the A2.3 same-target-population qualifier on the RCS card, CTA.tests/test_dml_carousel_claims.py(17 checks): every deck estimate located on the committed tutorial-32 notebook surface (row-associated learner-table pins), arithmetic backing for the five-SE and recovery claims, constants-reach-their-slides + qualifier slide-locality enforcement, a banned-absolutes/overclaims list, REGISTRY sync for the DoubleML figure, a dependency-free proof for the built-in-learner claim, and the documented-sklearn-contract check.Methodology references (required if estimator / math changes)
Validation
tests/test_dml_carousel_claims.py(new, 17 checks, green).Security / privacy