feat(delphi): acceptance canonicalization — suppress cross-engine ordering artifacts in certify comparer#2648
Draft
jucor wants to merge 1 commit into
Draft
feat(delphi): acceptance canonicalization — suppress cross-engine ordering artifacts in certify comparer#2648jucor wants to merge 1 commit into
jucor wants to merge 1 commit into
Conversation
This was referenced Jul 22, 2026
Draft
docs(delphi): R1-parity goal docs, journal (2026-07-18 → 07-24 s5), quirks + divergence ledger
#2626
Draft
Draft
This was referenced Jul 24, 2026
Draft
Contributor
There was a problem hiding this comment.
Pull request overview
This PR updates the Delphi replay/certification comparer to treat cross-engine ordering differences as non-semantic during acceptance comparison. It adds a canonicalization pass that normalizes ordering (and PCA component sign) so Clojure vs Python step-0 diffs don’t get dominated by insertion/sort-order artifacts.
Changes:
- Add
canonicalize_blob()to normalize ordering oftids, tid-aligned PCA arrays, base/group cluster ordering, and list-like id sets; also deterministically canonicalize PCA component signs and flip aligned arrays accordingly. - Apply canonicalization inside
project_acceptance()so both hash-first and diff paths compare normalized acceptance blobs. - Add a focused test suite covering (a) ordering-only equality, (b) PCA sign-flip invariance, and (c) preservation of “real” divergences.
Reviewed changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.
| File | Description |
|---|---|
delphi/polismath/replay/crosslang.py |
Implements acceptance-blob canonicalization (ordering + PCA sign normalization) for cross-engine comparison. |
delphi/polismath/replay/certify.py |
Wires canonicalization into project_acceptance() so comparer/hashing operates on normalized blobs. |
delphi/tests/replay_harness/test_certify_canonicalization.py |
Adds regression tests ensuring ordering/sign artifacts are suppressed while true divergences remain visible. |
Comments suppressed due to low confidence (1)
delphi/tests/replay_harness/test_certify_canonicalization.py:15
- Docstring contradicts the tests/implementation about
votes-base: it says votes-base is permuted along withbase-clusters, but the blob comments and canonicalizer deliberately do not permutevotes-base(it’s already aligned to sorted base-cluster ids on both engines). This is misleading for future readers troubleshooting acceptance diffs.
``project_acceptance`` therefore canonicalizes both sides before hashing and
diffing: id-sets sorted, tid-aligned pca arrays re-indexed by sorted tid,
base-clusters columns re-indexed by sorted id (votes-base per-cluster lists
following the same permutation), group-clusters sorted by id with sorted
members. Real divergences (a differing pid, a differing center value for the
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+5
to
+7
| Clojure emits ``tids``/``in-conv`` (and everything positionally aligned to | ||
| them: pca.center, pca.comps rows, base-clusters columns, votes-base lists) in | ||
| hash/insertion order, while Python emits sorted order. Each blob is internally |
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
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.
Clojure emits tids/in-conv (and everything positionally aligned to them:
pca center/comps rows/comment arrays, base-clusters columns) in
hash/insertion order; Python emits sorted. Both blobs are internally
consistent, so cross-engine array order is not a semantic divergence.
canonicalize_blob() (crosslang.py) normalizes both sides inside
project_acceptance: tids sorted with tid-aligned pca arrays re-indexed;
in-conv/mod-in/mod-out/meta-tids sorted when lists (None passes through);
base-clusters columns re-indexed by sorted id, members sorted;
group-clusters sorted by id, members sorted. votes-base is deliberately
NOT permuted (already sort-by-:id-aligned on both engines by construction,
conversation.clj:593). Kills ~850 of ~1030 step-0 battery divergences as
ordering artifacts; the residual real roots (sign convention, votes-base
shape, emission gaps) are ledgered in docs/divergences.json.
Also canonicalizes run-arbitrary PCA COMPONENT SIGNS: Clojure's
first-tick power iteration has no start vectors, so its unseeded init
flips component signs between ITS OWN runs (observed on a vw single-cut
re-record: comps[1] + base-clusters.y negated vs the prior recording).
Each component's sign is fixed deterministically (max-|entry| positive,
evaluated after tid alignment) and every component-aligned array flips
with it (comps row, comment-projection row, base-clusters x/y,
group-clusters center[k]); pca.center is a data mean and never flips.
An inconsistent flip (y negated without comps[1]) still diverges.
commit-id:d1ae732b
Stack: