feat(math): degenerate-tick group-clusterings overwrite + <2-participants edge (legacy mode) by jucor · Pull Request #2642 · compdemocracy/polis

jucor · 2026-07-22T00:51:07Z

Port of the approved 2026-07-21 Opus verdict (CLOJURE_QUIRKS Q4+Q5). Clojure
recomputes :group-clusterings EVERY tick unconditionally: max-k-fn >= 2 always
(conversation.clj:274-279), so a 1-base-cluster tick runs k=2 kmeans on one
point (clean-start caps at distinct-point count -> 1 cluster, lineage id),
OVERWRITES :group-clusterings with that degenerate value, and the recovery
tick warm-starts from it, minting split ids via (inc max-id)
(clusters.clj:267). Clojure also has no <2-in-conv-participants guard past the
truly-empty short-circuit.

clojure-legacy mode now skips both early returns and falls through to the
normal legacy path, which reproduces all of the above exactly: max_k
arithmetic yields k range [2]; calculate_silhouette_sklearn returns the 0.0
sentinel for singleton clusterings (= Clojure's singleton rule,
clusters.clj:350-353), so the P6a smoother advance is preserved with the same
{2: 0.0} input via the main path (the branch-local P6a advance block is
removed as dead). The 0-participant early return stays in both modes
(unreachable past the empty short-circuit given the PR-E greedy floor).
improved mode keeps both guards byte-for-byte.

TDD: tests/test_degenerate_tick_parity.py RED on the old code for the pinned
divergences (stale 2-cluster group_clusterings after a collapse tick; empty
structures on a single-participant tick), GREEN after; existing P6a smoother
tests unchanged and green (24 passed with test_group_k_smoother.py; 37 passed
with lineage/greedy-carry/pca-warm-start/engine-mode neighbors).

commit-id:bbb74f71

Stack:

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

…ants edge (legacy mode) Port of the approved 2026-07-21 Opus verdict (CLOJURE_QUIRKS Q4+Q5). Clojure recomputes :group-clusterings EVERY tick unconditionally: max-k-fn >= 2 always (conversation.clj:274-279), so a 1-base-cluster tick runs k=2 kmeans on one point (clean-start caps at distinct-point count -> 1 cluster, lineage id), OVERWRITES :group-clusterings with that degenerate value, and the recovery tick warm-starts from it, minting split ids via (inc max-id) (clusters.clj:267). Clojure also has no <2-in-conv-participants guard past the truly-empty short-circuit. clojure-legacy mode now skips both early returns and falls through to the normal legacy path, which reproduces all of the above exactly: max_k arithmetic yields k range [2]; calculate_silhouette_sklearn returns the 0.0 sentinel for singleton clusterings (= Clojure's singleton rule, clusters.clj:350-353), so the P6a smoother advance is preserved with the same {2: 0.0} input via the main path (the branch-local P6a advance block is removed as dead). The 0-participant early return stays in both modes (unreachable past the empty short-circuit given the PR-E greedy floor). improved mode keeps both guards byte-for-byte. TDD: tests/test_degenerate_tick_parity.py RED on the old code for the pinned divergences (stale 2-cluster group_clusterings after a collapse tick; empty structures on a single-participant tick), GREEN after; existing P6a smoother tests unchanged and green (24 passed with test_group_k_smoother.py; 37 passed with lineage/greedy-carry/pca-warm-start/engine-mode neighbors). commit-id:bbb74f71

jucor marked this pull request as draft July 22, 2026 00:51