feat(delphi): Q12 carve-out — pin the cold-tick PCA start to ones in both replay drivers#2655
Draft
jucor wants to merge 1 commit into
Draft
feat(delphi): Q12 carve-out — pin the cold-tick PCA start to ones in both replay drivers#2655jucor wants to merge 1 commit into
jucor wants to merge 1 commit into
Conversation
This was referenced Jul 22, 2026
docs(delphi): R1-parity goal docs, journal (2026-07-18 → 07-24 s5), quirks + divergence ledger
#2626
Draft
Draft
…both replay drivers Production Clojure draws an UNSEEDED random PCA start vector on the cold tick (rand-starting-vec, pca.clj:79-82 — the original author's own 'should really throw a [seeded] random number generator in the equation here... XXX'); with a small eigengap the fixed 100 power iterations keep a start-dependent residual (~4e-4 on pc-smallmix-01 step 0), so even two Clojure runs differ on cold ticks. No deterministic reference exists. Both replay drivers now pin the cold start to the ONES vector — the exact value power-iteration already pads new-comment columns with (pca.clj:46-49 / pca.py _power_iteration) — by seeding the conv with a single-element [1.0] start that the padding expands to any width (dev/replay.clj certify-cold-start-pca; polismath/replay/driver.py). Warm ticks take the real previous comps from tick 2 exactly as before. Logged per run. CLOJURE_QUIRKS.md Q12. Battery effect: pc-smallmix-01 and pc-midmix-01 went DIVERGENCE -> MATCH; pc-revote-01's step 0 cleared (residual from step 1 = split-loop extraction order on tie-dense 15-comment geometry, diagnosed separately). 14/15 MATCH with the five prodclone extractions in the battery. commit-id:6b40e1bf
jucor
changed the title
jucor
changed the title
jucor
changed the title
jucor
changed the title
This was referenced Jul 24, 2026
Draft
Contributor
There was a problem hiding this comment.
Pull request overview
This PR makes the replay/certification harness deterministic on the cold tick by forcing the initial PCA power-iteration start vectors to expand to an all-ones vector, matching the padding behavior already used when new comments are appended. This reduces run-to-run divergence between Clojure replays and provides a stable reference point for cross-engine certification.
Changes:
- Seed the Clojure replay driver’s initial conversation with single-element PCA comps to pin the cold-start power-iteration start vector to ones.
- Mirror the same cold-start PCA seeding in the Python replay driver by initializing
conv.pcabefore the first recompute. - Adjust
proj_probe.cljto avoid unconditional auto-run and add a batch-mode probe helper.
Reviewed changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.
| File | Description |
|---|---|
| math/dev/replay.clj | Seeds replay conv with a deterministic cold-start PCA comps vector for certification. |
| math/dev/proj_probe.clj | Prevents unconditional script auto-run; adds a batch-mode probe that reuses replay helpers. |
| delphi/polismath/replay/driver.py | Pins Python replay cold-start PCA by seeding conv.pca with ones-based comps. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
+131
to
+134
| ;; Auto-run only when invoked with args (clojure -M dev/proj_probe.clj ...); | ||
| ;; library-style load-file (batch-probe callers) skips it. | ||
| (when (seq *command-line-args*) | ||
| (apply -main *command-line-args*)) |
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.
Production Clojure draws an UNSEEDED random PCA start vector on the cold
tick (rand-starting-vec, pca.clj:79-82 — the original author's own
'should really throw a [seeded] random number generator in the equation
here... XXX'); with a small eigengap the fixed 100 power iterations keep
a start-dependent residual (~4e-4 on pc-smallmix-01 step 0), so even two
Clojure runs differ on cold ticks. No deterministic reference exists.
Both replay drivers now pin the cold start to the ONES vector — the
exact value power-iteration already pads new-comment columns with
(pca.clj:46-49 / pca.py _power_iteration) — by seeding the conv with a
single-element [1.0] start that the padding expands to any width
(dev/replay.clj certify-cold-start-pca; polismath/replay/driver.py).
Warm ticks take the real previous comps from tick 2 exactly as before.
Logged per run. CLOJURE_QUIRKS.md Q12.
Battery effect: pc-smallmix-01 and pc-midmix-01 went DIVERGENCE -> MATCH;
pc-revote-01's step 0 cleared (residual from step 1 = split-loop
extraction order on tie-dense 15-comment geometry, diagnosed separately).
14/15 MATCH with the five prodclone extractions in the battery.
commit-id:6b40e1bf
Stack: