feat(math): small-N degenerate parity — run the real math on tiny matrices in clojure-legacy mode

[jucor]

Clojure has NO small-dim guards: a 1x1 single-vote conversation computes
center = the vote, a rank-capped zero component, a repness entry (the
best-agree guarantee holds for one comment; rest-stats over zero other
groups fall back to the (0+1)/(0+2) prior) and a consensus selection —
pinned exactly by the vw every-vote step-0 recording (journal 2026-07-22
'every-vote step-0 diagnosis COMPLETE'). Python short-circuited to zeros/
empties below 2x2, which made the every-vote battery entry diverge from
step 0 (4650/4683 steps).

Legacy-gated changes (improved keeps every historical guard):

• _compute_pca + pca_project_dataframe: <2 guards only fire on truly
  EMPTY dims in legacy; the powerit path runs at any size (comps already
  rank-capped; zero-variance yields Clojure's zero vector). Projections
  zero-pad to 2-D (Clojure's [pc1 pc2] destructure zero-fills a missing
  second component), as does the emitted comment-projection.
• conv_repness: the shape<2 early return yields to the real computation.

TDD: single-vote fixtures with exact Clojure-derived values (center -1.0,
p-success 2/3, repness 4/3, best-agree), RED observed; improved-mode
guards pinned unchanged.

commit-id:00d8bc52

---

Stack:

• docs(delphi): overnight pre-cutover audit (s6) — CUTOVER_RUNBOOK + journal #2659
• feat(delphi): zid-sharding for the math poller — one shard = one process #2658
• ci: run postgres integration tests #2637
• feat(delphi): poller-equivalence harness (goal condition 2) — PASSES live on vw + pc-meta-02 #2657
• feat(math): moderation + restart replay parity — mod_update, restart-seam restore, Q13-Q18 #2656
• docs(delphi): R1-parity goal docs, journal (2026-07-18 → 07-24 s5), quirks + divergence ledger #2626
• feat(delphi): Q12 carve-out — pin the cold-tick PCA start to ones in both replay drivers #2655
• feat(math): Q11 — replicate vectorz distance cancellation in legacy kmeans #2654
• feat(math): small-N degenerate parity — run the real math on tiny matrices in clojure-legacy mode #2653 ⬅
• feat(delphi): Q10 carve-out — pin large-conv cutoffs to certify the deterministic full-PCA path #2652
• feat(delphi): replay loader resolves private datasets under real_data/.local #2651
• feat(math): Clojure hash-map iteration order for the in-conv greedy tie-break (legacy) #2650
• feat(math): clojure-legacy blob emission + selection parity — Clojure-exact math_main at step 0 #2649
• feat(delphi): acceptance canonicalization — suppress cross-engine ordering artifacts in certify comparer #2648
• feat(delphi): certify.py battery runner + focuser + fingerprint ledger (Spec A, Phase 0) #2647
• feat(delphi): prodclone extractor — feature-filtered dataset export (Spec B) #2646
• feat(math): un-mirror priority_metric (#2571 resolved) + Q2 prev-tick group-votes #2645
• feat(math): participant-ban leak replication in clojure-legacy mode (Q1) #2644
• feat(delphi): Clojure timing probe (Spec C) #2643
• feat(math): degenerate-tick group-clusterings overwrite + <2-participants edge (legacy mode) #2642
• refactor(math): move impl-flag resolver to polismath/utils/env_flags.py #2641
• feat(delphi): Python math poller service (phase 1) — Clojure math-container replacement #2625
• fix(delphi): Clojure-parity math writers + committing write path in PostgresClient #2624
• feat(math): in-conv greedy floor + persistent carry (PR-E) #2623
• feat(math): Clojure-exact base-cluster lineage + warm start (PR-C) #2622
• feat(math): replay harness H-B — Clojure Mode A driver (dev/replay.clj) + cross-language comparer shim #2621
• feat(math): group-k-smoother in clojure-legacy engine mode (PR-D smoother) #2620
• feat(math): PCA warm start in clojure-legacy engine mode (PR-B) #2640
• feat(math): add POLISMATH_ENGINE_MODE flag + prev-tick warm-start scaffolding (PR-A) #2618
• feat(delphi): replay harness H-A — schedule spec+slicer, Python driver, store, step comparer, CLI #2639
• fix(delphi): cold-start generator copies FULL revote history (no DISTINCT ON dedup) #2638
• fix(math): correct comment-extremity sign in pca_project_cmnts (D1b) #2615
• docs(delphi): add math-algorithm history & routing-regression reference #2613

---

⚠️ Part of a stack created by spr. Do not merge manually using the UI - doing so may have unexpected results.

[Copilot]

Pull request overview

This PR improves Clojure parity for degenerate small-N conversations in clojure-legacy engine mode by allowing the real PCA/repness math to run on tiny (e.g., 1×1) vote matrices, matching the Clojure behavior used by the step-0 “every-vote” oracle. It keeps the historical “short-circuit to zeros/empties” behavior unchanged in improved mode.

Changes:

• Relax PCA and repness “<2 rows/cols” guards in legacy mode so 1×1 conversations compute real center/comps/repness/consensus instead of returning empties.
• Zero-pad legacy PCA projections to keep the emitted shapes consistent with Clojure (notably 2D projections / 2-row comment-projection emission).
• Add targeted tests that pin exact legacy outputs for a single-vote fixture while asserting improved-mode behavior is unchanged.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
delphi/tests/test_legacy_blob_shape.py	Adds single-vote degenerate fixtures asserting legacy runs real math and improved mode keeps existing guards.
delphi/polismath/pca_kmeans_rep/repness.py	Makes repness/consensus run for 1×1 in clojure-legacy while preserving the historical guard in improved.
delphi/polismath/pca_kmeans_rep/pca.py	Adjusts PCA small-dim guard behavior for the require_powerit (legacy) path and pads projections for legacy parity.
delphi/polismath/conversation/conversation.py	Updates PCA guard logic by engine mode and pads legacy comment-projection output to match Clojure’s always-2-row emission.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.