feat(math): in-conv greedy floor + persistent carry (PR-E)

[jucor]

Port the two Clojure :in-conv steps the Python pipeline was missing
(conversation.clj:243-269) into the 'clojure-legacy' engine mode of
Conversation._get_in_conv_participants. Improved mode (default) is unchanged
(threshold set only, no carry, no greedy floor).

Legacy mode now, every tick:

1. CARRY: unions the vote-threshold set (>= min(7, n_cmts) votes) into the
   PERSISTENT in-conv set carried on the conv ((or (:in-conv conv) #{}),
   conversation.clj:247), so a participant, once in, never leaves.
2. GREEDY FLOOR: if fewer than 15 participants are in, greedily admits the
   top 15 - n remaining participants by vote count descending, INCLUDING
   below-threshold voters (conversation.clj:259-268), and persists them.
   The persisted set feeds base clustering (which matrix rows) and is serialized
   as the blob :in-conv (to_dict), matching the clustered rows.

self.in_conv is threaded in-memory across update_votes (deepcopy in recompute),
NOT persisted to dynamo — same lifetime as the PCA/smoother/cluster warm state.

Greedy tie rule: Clojure sorts a hash-map with (sort-by (comp - second)),
whose order among EQUAL vote counts is hash-map iteration order — inherently
non-deterministic. We break ties by matrix ROW ORDER (user-vote-counts
insertion order) via a STABLE sort — a deterministic, reproducible surrogate
for that underspecified Clojure tie case. Flagged for the integrator.

RED evidence (against PR-C HEAD, before this commit):

• test_legacy_greedy_fills_to_fifteen: got 2 (threshold only) != 15.
• test_legacy_greedy_tie_break_is_row_order: lows never admitted.
• carry / self.in_conv tests: AttributeError (no in_conv field / no greedy).
  GREEN: 9 new tests pass; full suite 520 passed / 17 skipped / 47 xfailed (was
  511 after PR-C), zero regressions. The vw cold-identity gate stays green (67
  in-conv > 15, so no greedy fires and the serialized :in-conv is unchanged).

Co-Authored-By: Claude Fable 5 noreply@anthropic.com

commit-id:a861cd87

---

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

Ports Clojure’s :in-conv semantics into the Python clojure-legacy engine path so legacy-mode clustering uses the same “persistent carry + greedy floor to 15” participant selection behavior as the Clojure math pipeline, while keeping the default improved mode unchanged.

Changes:

• Add Conversation.in_conv warm-state to persist legacy-mode in-conv membership across ticks (carry semantics).
• Implement legacy-mode greedy floor logic in _get_in_conv_participants (fill to 15 by vote-count, stable tie-break by row order, and prune banned participants).
• Add a focused test module covering improved-vs-legacy behavior, carry persistence, serialization, monotonicity, and ban-prune floor refill.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
delphi/polismath/conversation/conversation.py	Adds legacy-mode persistent in_conv state, greedy floor + prune logic in participant selection, and legacy serialization of in-conv to match clustered rows.
delphi/tests/test_in_conv_greedy_carry.py	New unit tests validating legacy carry+greedy behavior and ensuring improved mode remains threshold-only.

---

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