bruno/homemaker-layout
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

homemaker-py-sd3: fix vacuous 94g keep-better guard (collapse_insearch leak)

Browse source 
driver.collapse_best built its evaluator with _fitness_for's default
collapse_insearch=True, so collapse_finish's base_fails/cand_fails were
both measured through score_with_fails' own auto-collapse pass -- base
silently equalled collapsed on 5/5 probed files, making the "keep only
if fails don't increase" safety guard vacuous and understating 94g's
real effect in logs. fitness.collapse_finish now forces canonical
(collapse_insearch=False) scoring for its own measurement regardless of
self's config; collapse_best now builds its evaluator canonically too
(matching what homemaker-fitness reports for the written .dom) and
threads max_share/conn_grade through. Same-family fix in
search_annealed's no-polish-budget rescore branch, which silently
defaulted to collapse_insearch=True via _evaluate's default.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014uCyZJCh5mBuA2yEFxgQpo
This commit is contained in:
Bruno Postle 2026-08-05 07:52:44 +01:00
parent 30adbf4948
commit 35adcd4b4d
5 changed files with 95 additions and 27 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

38
.beads/issues.jsonl
View file

@ -31,7 +31,7 @@
{"id":"homemaker-py-wkh","title":"DP-exact hard pre-filter: replace/augment predicted_shape_fails with shapecurve's boolean infeasibility","description":"homemaker-py-6xh item 1 (DESIGN.md §37.2/§37.4). The shapecurve DP (src/homemaker_layout/shapecurve.py, promoted from experiments/shapecurve_spike.py) gives an EXACT feasible/infeasible verdict for the size/width/proportion family, currently wired only as an NM warm-start (safe: never prunes). operators.predicted_shape_fails' threshold-based prune (driver._evaluate, feasibility_max_shape_fails/best_n_fails) still uses the older heuristic-count proxy. Using shapecurve.solve's infeasible verdict as an ADDITIONAL/replacement hard-prune signal would be stronger (exact, not a graduated heuristic) but riskier: unlike a bad warm-start, a wrong prune permanently discards a topology that could have beaten the incumbent. DESIGN.md §37.2 measured 0/200 false negatives (DP infeasible, NM reaches 0 anyway) on harbor-house-l0, but that is not a proven bound (the rectangle-vs-skew-quad approximation is a known ~7-12% error source, §37.2). Needs: (a) a design for how the DP's boolean signal composes with the existing pred\u003ethreshold\u0026\u0026pred\u003e=best_n_fails guard, (b) a false-negative-risk validation before enabling by default (a larger/less-rectangular topology sweep than the 200-topology harbor-house-l0 one), (c) a driver.search A/B (evals-to-N-hard-fails) against today's predicted_shape_fails-only filter.","notes":"2026-08-03: Shipped the DP-exact hard pre-filter, DESIGN.md §37.5. Full\ndetails there; summary:\n\n- shapecurve.is_feasible() (new, non-mutating refactor of solve()'s check\n phase) + shapecurve_prune flag in driver._evaluate/search, threaded\n through to `homemaker-evolve --shapecurve-prune` (default off, mirrors\n --shapecurve-warmstart). Composition: DP-feasible vetoes a heuristic\n prune outright (skips predicted_shape_fails entirely); DP-infeasible only\n hard-prunes when the incumbent already has 0 total fails (exact, since\n infeasible proves the shape-fail floor \u003e=1); otherwise defers unchanged\n to the existing predicted_shape_fails threshold. Conservative by design\n per the bead's own risk framing (a wrong prune is unrecoverable, unlike a\n bad warm-start).\n- Validation (bead item b): pointed experiments/validate_shapecurve.py at\n the promoted product module (was still validating the frozen spike) and\n gave it a programme_dir CLI arg; ran the same 200-topology protocol\n against examples/programme-house (a genuinely skewed, non-axis-aligned\n plot, not just a rotated harbor-house-l0): 200/200 agreement, 0 false\n positives, 0 false negatives, 87.4x speedup. Combined with §37.2's\n original 200 on harbor-house-l0: 0/400 false negatives across two\n structurally distinct plots.\n- A/B (bead item c): experiments/ab_shapecurve_prune.py, same protocol as\n 6xh's warm-start A/B (harbor-house-l0, budget=2000, seeds 0-4). Result:\n byte-identical off/on across all 5 seeds -- NULL, not a regression.\n Instrumented root cause: on this benchmark predicted_shape_fails itself\n (pre-existing 9gp.1, not this bead's code) rarely reaches best_n_fails\n organically -- tests/test_driver.py's own test_feasibility_filter_\n prunes_cheaply already had to force it to 999 to observe any real prune\n -- so neither the veto nor the exact-prune branch had an opening to fire\n (spied: 17/17 DP checks infeasible, incumbent total fails never reached\n 0). Not a wkh defect; 9gp.1 is documented as a \"scaling lever\", expected\n to matter at larger programmes/leaf counts than this benchmark, not here.\n\nTests: tests/test_shapecurve.py (+1), tests/test_driver.py (+4). Full\nsuite: 393 passed.\n\nFollow-up (not blocking this close, noted in DESIGN.md §37.5): re-run the\nA/B at a scale where predicted_shape_fails organically prunes to see wkh's\nmarginal value -- the more direct route there is homemaker-py-koo\n(multi-storey) and homemaker-py-tym (leaf-sharing), since today's DP\neligibility already excludes the real \u003e=2-storey, leaf-sharing-default\nprogrammes (programme-house, harbor-house) this would need to be measured\non.","status":"closed","priority":2,"issue_type":"feature","assignee":"Bruno Postle","owner":"bruno@postle.net","created_at":"2026-08-03T17:29:13Z","created_by":"Bruno Postle","updated_at":"2026-08-03T20:08:37Z","started_at":"2026-08-03T18:16:35Z","closed_at":"2026-08-03T20:08:37Z","close_reason":"DP-exact hard prune shipped + validated (0/400 false negatives across 2 plots); A/B measured NULL on harbor-house-l0 for a root-caused, pre-existing reason (9gp.1 rarely engages organically at this scale)","dependencies":[{"issue_id":"homemaker-py-wkh","depends_on_id":"homemaker-py-2g7","type":"parent-child","created_at":"2026-08-03T18:31:46Z","created_by":"Bruno Postle","metadata":"{}"}],"dependency_count":0,"dependent_count":0,"comment_count":0}
{"id":"homemaker-py-6xh","title":"Wire shapecurve DP prototype into driver.py as a real pre-filter + NM warm-start","description":"homemaker-py-2g7.4's prototype (experiments/shapecurve_spike.py, DESIGN.md\n§37.2) validated PASS on harbor-house-l0 (99% agreement vs shape-fail-only NM\nover 200 random topologies, 93.6x speedup at grid_n=150, 0 false negatives).\nIt is not yet wired into the product — it's a reference spike only, same\nstatus as experiments/autodiff_spike.py (§34).\n\nTo productionise per the original plan (DESIGN.md §37 point 2):\n- Replace/augment operators.predicted_shape_fails with the DP as driver.py's\n real per-child pre-filter (a single-sample heuristic today; the DP gives an\n exact yes/no plus a realizing ratio point).\n- Warm-start innerloop.optimise's NM from the DP's realised ratios instead of\n (or in addition to) the current proportion-aware target-geometry seed.\n- Multi-storey support: the DP only walks one level's leaves currently;\n below-linked nodes (wall-stacking across storeys) aren't modelled.\n- leaf_sharing/co_type target-adjustment: not modelled in leaf_constraints,\n needed for any programme that uses either (harbor-house-l0 doesn't).\n- Consider replacing the bounding-box leaf approximation with true skew-quad\n polygon algebra to remove the ~7-12% area approximation error identified\n as the root cause of both measured false positives (§37.2) -- or at least\n characterise it on a LESS rectangular plot than harbor-house-l0's\n near-rectangular trapezoid, where the error is likely worse.\n- A/B against the real driver.search: does DP-pre-filter + warm-start beat\n today's predicted_shape_fails + cold/proportion-aware start on wall-clock\n to N hard fails, on harbor-house (full) and/or a less-rectangular plot?","notes":"2026-08-03: Shipped NM warm-start (item 2) + a scoped A/B (item 5), left\nin_progress -- 3 of 5 description items deliberately deferred to new\ntracked beads (see below). Full details + measured numbers: DESIGN.md\n§37.4.\n\nWhat shipped: promoted experiments/shapecurve_spike.py into\nsrc/homemaker_layout/shapecurve.py (fixed a latent numpy.float64-in-division\nbug caught by round-tripping through dom.dumps in the new tests -- the spike\nnever round-tripped and so never caught it). Added shapecurve.eligible()\n(single storey, no leaf_sharing/superpose/max_share/multi_use). Wired into\ndriver._evaluate as an NM warm-start only (never a prune) behind\nshapecurve_warmstart=False default, threaded through driver.search and\nexposed as `homemaker-evolve --shapecurve-warmstart`. A/B\n(experiments/ab_shapecurve_warmstart.py) on harbor-house-l0, budget=2000,\n5 seeds: mean total-fails 16.6 (on) vs 19.6 (off), ~3.5x mean fitness\nimprovement; mean hard-fail count alone was a noise-level wash (4.6 vs 4.4\nat n=5). Tests: tests/test_shapecurve.py (4), tests/test_driver.py (+3).\nFull suite 388 passed.\n\nDeferred to new tracked beads (children of 2g7, per the epic's own\ndependency ordering):\n- homemaker-py-wkh: DP-exact hard pre-filter (item 1) -- replacing\n predicted_shape_fails' heuristic threshold with the DP's exact\n infeasibility verdict. Needed to actually chase \"evals to N hard fails\"\n rather than just improve soft-fail/fitness convergence.\n- homemaker-py-koo: multi-storey (below-link) DP support (item 3) --\n without this, the warm-start never fires on programme-house\n (storey_minimum=2) or full harbor-house, only the purpose-built\n single-storey harbor-house-l0.\n- homemaker-py-tym: leaf_sharing/co_type modelling (item 4) -- without this,\n the warm-start never fires when leaf_sharing=True, which is\n driver.search's own default.\n- homemaker-py-ekc: true skew-quad polygon algebra (the §37.2-quantified\n ~7-12% rectangle-approximation error) -- not a new bead-description item,\n but the explicit \"consider replacing the bounding-box leaf approximation\"\n bullet.\n\nNet: 6xh's own acceptance (a real evals-to-N-hard-fails win over\npredicted_shape_fails + cold/proportion-aware start) is NOT yet met --\ntoday's result is a safe, positive-but-partial step (soft-fail/fitness\nconvergence, not hard-fail convergence, and only on the single-storey\nno-sharing envelope). Keeping 6xh in_progress rather than closing it.","status":"in_progress","priority":2,"issue_type":"feature","assignee":"Bruno Postle","owner":"bruno@postle.net","created_at":"2026-08-02T22:39:46Z","created_by":"Bruno Postle","updated_at":"2026-08-03T17:40:52Z","started_at":"2026-08-03T15:51:29Z","dependency_count":0,"dependent_count":0,"comment_count":0}
{"id":"homemaker-py-2g7.9","title":"Parallel best-of-N + racing harness (use all cores, kill stragglers early)","description":"§14 measured islands \u003c= best-of-N, and the 3M runs used workers=1-2 on a 4-core box — independent seeds are the proven shape and we are not even using the local machine. Build a harness: launch N independent search_staged seeds across all cores (processes, not threads — mind the cvw id()-keyed cache bug), checkpoint fail-counts periodically, successively halve (hyperband-style: kill runs above median hard-fail count at each rung, reallocate budget to survivors). Fix/respect homemaker-py-b8g (parallel non-determinism) and homemaker-py-cvw first or work around with process isolation. This multiplies whatever eval cost the shape-curve DP issue achieves; on its own it is a free 4x locally and scales to any box. Report best + variance across seeds (the seed-variance in §12-§13 tables is huge — 78 vs 97 same config — so best-of-N is worth several levers combined).","acceptance_criteria":"harness runs N=16 seeds on 4 cores with racing; at equal total native-eval budget beats the single-seed mean on harbor by at least the observed seed spread; deterministic per-seed replay","status":"open","priority":2,"issue_type":"task","owner":"bruno@postle.net","created_at":"2026-08-02T09:15:58Z","created_by":"Bruno Postle","updated_at":"2026-08-02T09:15:58Z","dependencies":[{"issue_id":"homemaker-py-2g7.9","depends_on_id":"homemaker-py-2g7","type":"parent-child","created_at":"2026-08-02T10:15:58Z","created_by":"Bruno Postle","metadata":"{}"},{"issue_id":"homemaker-py-2g7.9","depends_on_id":"homemaker-py-b8g","type":"blocks","created_at":"2026-08-02T10:16:16Z","created_by":"Bruno Postle","metadata":"{}"},{"issue_id":"homemaker-py-2g7.9","depends_on_id":"homemaker-py-cvw","type":"blocks","created_at":"2026-08-02T10:16:15Z","created_by":"Bruno Postle","metadata":"{}"}],"dependency_count":2,"dependent_count":0,"comment_count":0}
{"id":"homemaker-py-2g7.7","title":"LLM repair operator at stagnation (dom+fails -\u003e targeted compound edits)","description":"Generalize the §4.10 lesson: deceptive valleys are crossed by COMPOUND edits (move room + re-home displaced room + fix ratios atomically), which we currently hand-code one per valley (mutate_level_compound_fix). Our fail messages are semantically rich and localized ('me1 on wrong level', 'level 1 not connected', '0/rlrlr proportion') and the .dom is readable — ideal LLM input. Loop: on stagnation (no fail-tier improvement for N evals), serialize best individual + .fails + programme summary -\u003e LLM proposes 3-5 multi-step repairs as structured edit scripts (a small DSL over existing operator primitives: swap/divide/retype/rotate with explicit paths — NOT freeform dom text, so proposals are always well-formed) -\u003e apply, inner-loop, lex-accept as usual. Native fitness disposes; a bad proposal costs one child budget. Cost discipline: one LLM call ~ thousands of native evals, so plateau-only, cache by (signature, fails) key. Benchmark: the 3M-run best sat on 'level 0 not connected' + 'me1 on wrong level' for \u003e1M evals — moves a plan-reader fixes in one edit. Use claude via API (see claude-api skill); temperature\u003e0 for diverse proposals. Later extension (separate issue): AlphaEvolve-style operator-code synthesis using our existing A/B harness as the evaluator.","acceptance_criteria":"on the evolved-3M-nols-3 15-fail plateau seed: repair loop reduces hard-fail count where 1M+ blind evals did not, within \u003c=20 LLM calls; edit-DSL rejects malformed proposals; A/B at equal native-eval budget shows strictly better final fails on \u003e=2/3 seeds","status":"open","priority":2,"issue_type":"feature","owner":"bruno@postle.net","created_at":"2026-08-02T09:15:54Z","created_by":"Bruno Postle","updated_at":"2026-08-02T09:15:54Z","dependencies":[{"issue_id":"homemaker-py-2g7.7","depends_on_id":"homemaker-py-2g7","type":"parent-child","created_at":"2026-08-02T10:15:53Z","created_by":"Bruno Postle","metadata":"{}"}],"dependency_count":0,"dependent_count":1,"comment_count":0}
{"id":"homemaker-py-2g7.7","title":"LLM repair operator at stagnation (dom+fails -\u003e targeted compound edits)","description":"Generalize the §4.10 lesson: deceptive valleys are crossed by COMPOUND edits (move room + re-home displaced room + fix ratios atomically), which we currently hand-code one per valley (mutate_level_compound_fix). Our fail messages are semantically rich and localized ('me1 on wrong level', 'level 1 not connected', '0/rlrlr proportion') and the .dom is readable — ideal LLM input. Loop: on stagnation (no fail-tier improvement for N evals), serialize best individual + .fails + programme summary -\u003e LLM proposes 3-5 multi-step repairs as structured edit scripts (a small DSL over existing operator primitives: swap/divide/retype/rotate with explicit paths — NOT freeform dom text, so proposals are always well-formed) -\u003e apply, inner-loop, lex-accept as usual. Native fitness disposes; a bad proposal costs one child budget. Cost discipline: one LLM call ~ thousands of native evals, so plateau-only, cache by (signature, fails) key. Benchmark: the 3M-run best sat on 'level 0 not connected' + 'me1 on wrong level' for \u003e1M evals — moves a plan-reader fixes in one edit. Use claude via API (see claude-api skill); temperature\u003e0 for diverse proposals. Later extension (separate issue): AlphaEvolve-style operator-code synthesis using our existing A/B harness as the evaluator.","acceptance_criteria":"on the evolved-3M-nols-3 15-fail plateau seed: repair loop reduces hard-fail count where 1M+ blind evals did not, within \u003c=20 LLM calls; edit-DSL rejects malformed proposals; A/B at equal native-eval budget shows strictly better final fails on \u003e=2/3 seeds","status":"open","priority":2,"issue_type":"feature","assignee":"Bruno Postle","owner":"bruno@postle.net","created_at":"2026-08-02T09:15:54Z","created_by":"Bruno Postle","updated_at":"2026-08-04T23:40:49Z","started_at":"2026-08-04T23:31:21Z","dependencies":[{"issue_id":"homemaker-py-2g7.7","depends_on_id":"homemaker-py-2g7","type":"parent-child","created_at":"2026-08-02T10:15:53Z","created_by":"Bruno Postle","metadata":"{}"}],"dependency_count":0,"dependent_count":1,"comment_count":0}
{"id":"homemaker-py-2g7.6","title":"Spike: graph-first construction — adjacency-realizing slicing trees / rectangular dualization","description":"Research spike, timeboxed. Literature: rectangular dualization (planar triangulated graph -\u003e rectangular floorplan) and characterizations of slicible adjacency graphs. Our programme already IS an adjacency graph (every room wants c, plus secondary pairs); instead of mutating trees hoping adjacency emerges, construct trees that realize the required adjacency by construction — the direction §11.6/§11.7 crawled toward greedily. Deliverable is a WRITTEN assessment (DESIGN.md section): can harbor's programme graph (16 rooms + spine, 2 storeys with stacking constraint) be dualized into slicing trees, how many, and is enumeration of realizing trees tractable? Prototype only if the answer is clearly yes. Watch for: multi-storey Below-inheritance constrains both floors' trees jointly; circulation spine is a connected dominating set requirement, not a simple adjacency.","acceptance_criteria":"DESIGN.md section with go/no-go verdict, the relevant algorithms named, and complexity estimate for harbor-scale programmes","status":"closed","priority":2,"issue_type":"task","assignee":"Bruno Postle","owner":"bruno@postle.net","created_at":"2026-08-02T09:15:07Z","created_by":"Bruno Postle","updated_at":"2026-08-04T23:21:32Z","started_at":"2026-08-04T23:08:59Z","closed_at":"2026-08-04T23:21:32Z","close_reason":"NO-GO verdict in DESIGN.md §37.8: rectangular dualization assumes one-vertex-one-rectangle; harbor's circulation hub is an emergent-shape multi-leaf region (§11.6 CDS seeding), breaking that assumption where it matters. Room-only graph is a trivial 3-edge matching, already fully satisfied by §11.7 seeding (0 secondary-adjacency fails on the 2g7.7 plateau benchmark). No multi-storey dualization precedent in the literature. Not prototyping.","dependencies":[{"issue_id":"homemaker-py-2g7.6","depends_on_id":"homemaker-py-2g7","type":"parent-child","created_at":"2026-08-02T10:15:07Z","created_by":"Bruno Postle","metadata":"{}"}],"dependency_count":0,"dependent_count":0,"comment_count":0}
{"id":"homemaker-py-2g7.5","title":"CP-SAT type assignment for a fixed tree (replace swap/retype random walk)","description":"For a FIXED topology, assigning room codes to leaves subject to counts, required levels, adjacency-to-circulation-spine, secondary adjacencies (k1-da1, da1-o...), and share grouping is a small discrete problem (~30-70 leaves, ~16-26 codes) — well within OR-Tools CP-SAT range, solvable optimally in milliseconds. Today swap/retype/level_retype random-walk this space; §11.6/§11.7's greedy constructive assignment was the single biggest fail-count win of Phase 6, and CP-SAT is its exact big brother. Plan: model leaf-graph adjacency (geometry.leaf_graph) as fixed at seed geometry; objective = weighted satisfied adjacencies + level compliance; use as (a) seeder replacing the greedy _assign_adjacency_aware, (b) periodic 'reassign' operator inside search (the assignment analogue of ruin_recreate §23), (c) post-collapse repair. Note the §11.2 lesson: assignment quality at SEED geometry can shift after the inner loop moves ratios — re-run assignment after geometry settles (alternating minimization).","acceptance_criteria":"A/B vs greedy seeder (harbor+maple, 3 seeds, 20k evals): adjacency+access seed fails strictly lower; end-to-end mean fails no worse; reassign operator fires and is accepted at least once per run","notes":"2026-08-04: Ran the bead's own full acceptance-criteria A/B (harbor+maple,\n3 seeds, budget=20000, experiments/ab_cpsat_assign.py) to completion --\n~12h wall clock, 18 driver.search runs total. Raw log:\nexperiments/results/ab_cpsat_assign_20k_harbor_maple.log.\n\nResults (mean hard/soft fails over 3 seeds):\n harbor-house: greedy 9.3/35.3 | cpsat 13.0/32.0 | reassign 8.3/35.3\n maple-court: greedy 22.7/72.3 | cpsat 22.3/68.3 | reassign 35.3/77.7\n reassign_fired (mean/3 seeds): harbor 0.0, maple 0.3 (fired in just\n 1 of 18 runs total, i.e. 1 of 6 reassign-arm runs).\n\nVerdict: acceptance criteria NOT met.\n - \"adjacency+access seed fails strictly lower\" -- cpsat is WORSE than\n greedy on harbor-house hard fails (13.0 vs 9.3); roughly a wash on\n maple-court (22.3 vs 22.7). No consistent win.\n - \"end-to-end mean fails no worse\" -- reassign arm is much worse on\n maple-court (35.3 vs 22.7 hard).\n - \"reassign fires and is accepted at least once per run\" -- fired in\n only 1/6 reassign-arm runs, 0/3 on harbor-house entirely. Confirms\n the earlier pilot's diagnosis: even at 20k budget the uniform-weight\n operator draw rate is too low for it to matter in a single run.\n\nDecision: assign_solver stays default \"greedy\", enable_reassign stays\ndefault False. The seeder-level win documented in\ntest_assign_cpsat_matches_or_beats_greedy_secondary_adjacency (isolated,\nseed-geometry-only, secondary-adjacency-only metric) does not survive\ncontact with a full driver.search run across two programmes -- likely\nbecause CP-SAT's exact optimum at seed geometry doesn't stay optimal once\nthe inner loop moves ratios (the bead's own §11.2 lesson), and any gain\ngets swamped by search noise at this budget. Not pursuing a higher budget\nor more seeds -- the direction (no clear win, one programme regresses)\nis consistent enough with the pilot to close this out rather than keep\nspending compute chasing it.\n\nThis bead's acceptance criteria are now fully evaluated (both shipped\nsub-items (a)/(b) AND this final A/B). Closing homemaker-py-2g7.5.\nhomemaker-py-5bv (item (c), post-collapse repair) remains open as a\nseparate child bead.","status":"closed","priority":2,"issue_type":"feature","assignee":"Bruno Postle","owner":"bruno@postle.net","created_at":"2026-08-02T09:15:06Z","created_by":"Bruno Postle","updated_at":"2026-08-04T22:05:06Z","started_at":"2026-08-03T22:44:01Z","closed_at":"2026-08-04T22:05:06Z","close_reason":"Closed","dependencies":[{"issue_id":"homemaker-py-2g7.5","depends_on_id":"homemaker-py-2g7","type":"parent-child","created_at":"2026-08-02T10:15:05Z","created_by":"Bruno Postle","metadata":"{}"}],"dependency_count":0,"dependent_count":0,"comment_count":0}
{"id":"homemaker-py-cvw","title":"Parallel staged runs: substrate_readiness reads stale id()-keyed geometry cache in the parent process","description":"Found by the homemaker-py-zrx expert review. geometry._cache is keyed by (id(node), idx) and relies on every reader being preceded by clear_cache(). In driver.search_staged stage 1 with n_workers\u003e1 that contract breaks: the PARENT process never runs score_with_fails (children are scored in the pool workers), so its cache is never cleared, yet _rank_fitness -\u003e rank_bonus_fn -\u003e graph.substrate_readiness(ind.root) reads geometry.area()/coordinate() in the parent on every tournament/admit comparison. Evicted individuals are eventually gc'd (Node trees are parent\u003c-\u003echild reference cycles, freed by the cycle collector) while their cache entries linger; freshly unpickled worker results reuse those addresses, and substrate_readiness then serves another (dead) tree's coordinates.\n\nVerified with a probe simulating the parent's allocation pattern (unpickle jittered harbor-house trees, pop-16 eviction churn, periodic gc.collect): 24/300 readiness computations returned a corrupted value, worst absolute error 0.999 on the [0,1] readiness scale (i.e. completely wrong), and the parent cache grew without bound (38k entries after 300 children — it is never cleared for the whole run). Serial staged runs are safe (every in-process score_with_fails clears the cache between children, and live/dead id coexistence prevents collisions).\n\nImpact: silently biases stage-1 substrate selection in every parallel staged run (run_staged_search.py with WORKERS\u003e1 — the default experimental harness), and makes the bias address-dependent, i.e. NON-DETERMINISTIC across byte-identical re-runs. This is a concrete, static-read-visible candidate mechanism for part of homemaker-py-b8g's irreproducibility (it is not BLAS): it perturbs the stage-1 trajectory, not a single fixed-genome score. Reported fitness numbers are unaffected (the bonus only reorders the comparator).\n\nRecommended fix: geometry.clear_cache() at substrate_readiness entry (cheap: the readiness read is a handful of areas on the base level; serial-mode behaviour is unchanged because the cache there is already cold at that point). The durable fix for the whole bug class — also covering the (unobserved but real) gc-timing hazard in collapse_finish's cand deepcopy, probed 0/6 today only because cyclic trees outlive the deepcopy window — is to cache on the Node object itself (as Urb does, per geometry.py's own comment) or key by a per-tree epoch, so a recycled address can never alias. Also add a defensive geometry.clear_cache() at collapse_global entry (one line, zero practical cost: finish-time it is one-shot, in-search the cache was just cleared by _evaluate_full).","status":"closed","priority":2,"issue_type":"bug","assignee":"Bruno Postle","owner":"bruno@postle.net","created_at":"2026-08-02T08:19:18Z","created_by":"Bruno Postle","updated_at":"2026-08-02T09:52:49Z","started_at":"2026-08-02T09:52:09Z","closed_at":"2026-08-02T09:52:49Z","close_reason":"Fixed: geometry.clear_cache() added at substrate_readiness (graph.py) and collapse_global (fitness.py) entry points; commit 2f26f46. Full test suite (338 tests) passes.","dependency_count":0,"dependent_count":1,"comment_count":0}
@ -85,7 +85,7 @@
{"id":"homemaker-py-2g7.10","title":"MAP-Elites archive over (hard-fail profile, leaf count, circulation fraction)","description":"Quality-diversity as the population-level answer to the §4.10 deceptive-valley problem: an archive keeps the elite per behavior niche, so 'transiently worse but structurally different' stepping stones survive — exactly what lex selection provably discards (§11.4's own analysis). DISTINCT from the failed §11.5/§11.8 niching: that kept diverse individuals under ONE selection pressure; MAP-Elites keeps the BEST individual per niche with no cross-niche competition. Descriptors to try: hard-fail category histogram (bucketed), total leaf count, circulation area fraction, storey balance. Emit from the existing genome.signature/score_with_grade machinery (kept default-off for exactly this reuse, §11.4 verdict). Blocked on the shape-curve DP: archive-filling needs cheap evals to be meaningful. Gate honestly per the ledger discipline: 3 seeds, control = current default stack.","acceptance_criteria":"A/B at equal budget (harbor+maple, 3 seeds): archive best hard-fails \u003c= default-stack best on mean; archive demonstrably contains the stepping stone for at least one accepted valley-crossing (traceable lineage)","status":"open","priority":3,"issue_type":"feature","owner":"bruno@postle.net","created_at":"2026-08-02T09:16:00Z","created_by":"Bruno Postle","updated_at":"2026-08-02T09:16:00Z","dependencies":[{"issue_id":"homemaker-py-2g7.10","depends_on_id":"homemaker-py-2g7","type":"parent-child","created_at":"2026-08-02T10:15:59Z","created_by":"Bruno Postle","metadata":"{}"},{"issue_id":"homemaker-py-2g7.10","depends_on_id":"homemaker-py-2g7.4","type":"blocks","created_at":"2026-08-02T10:15:59Z","created_by":"Bruno Postle","metadata":"{}"}],"dependency_count":1,"dependent_count":0,"comment_count":0}
{"id":"homemaker-py-2g7.8","title":"LLM operator synthesis (AlphaEvolve-style): evolve mutation-operator code against the A/B harness","description":"Second LLM role, after the repair operator proves the plumbing: let the LLM propose new OPERATOR CODE (python functions with the mutate_* signature) and evaluate candidates with the exact experiment discipline DESIGN.md already enforces (control reproduces baseline, 3 seeds, 20k evals, verdict). The project's ledger of 20+ operator experiments with verdicts is unusually good few-shot material: feed it the §11-§13 history so it learns what already failed (niching, grading, annealing...) and why. Sandbox the generated code; acceptance purely empirical via the harness. This is compute-hungry — schedule after the shape-curve DP lands so each A/B is cheap.","acceptance_criteria":"one synthesized operator survives the standard 3-seed A/B gate on harbor or maple (mean fails strictly better, control reproduces baseline)","status":"open","priority":3,"issue_type":"feature","owner":"bruno@postle.net","created_at":"2026-08-02T09:15:56Z","created_by":"Bruno Postle","updated_at":"2026-08-02T09:15:56Z","dependencies":[{"issue_id":"homemaker-py-2g7.8","depends_on_id":"homemaker-py-2g7","type":"parent-child","created_at":"2026-08-02T10:15:56Z","created_by":"Bruno Postle","metadata":"{}"},{"issue_id":"homemaker-py-2g7.8","depends_on_id":"homemaker-py-2g7.7","type":"blocks","created_at":"2026-08-02T10:15:56Z","created_by":"Bruno Postle","metadata":"{}"}],"dependency_count":1,"dependent_count":0,"comment_count":0}
{"id":"homemaker-py-pek","title":"fitness.py: delete the dead first process_storey definition (silently shadowed)","description":"Found by the homemaker-py-zrx expert review. class Fitness defines process_storey TWICE: the original gnw-scope version at fitness.py:1146 and the extended hgg version at fitness.py:1452. Python keeps only the second; the first ~45 lines are dead code that still reads as live. This is a silent-bug vector: an edit to the first definition (e.g. a fix to the covered-outside failure emission, which is duplicated verbatim in both) changes nothing at runtime and no test would notice. Delete the first definition (its docstring notes are preserved in the second). No behaviour change; run the suite to confirm 337 pass.","status":"open","priority":3,"issue_type":"task","owner":"bruno@postle.net","created_at":"2026-08-02T08:19:56Z","created_by":"Bruno Postle","updated_at":"2026-08-02T08:19:56Z","dependency_count":0,"dependent_count":0,"comment_count":0}
{"id":"homemaker-py-sd3","title":"driver.collapse_best bakes collapse_insearch=True into its finish evaluator, making the 94g keep-better guard vacuous","description":"Found by the homemaker-py-zrx expert review; same family as homemaker-py-7ua but in the PRODUCT (driver.py), not the experiment script. driver.collapse_best builds its evaluator as _fitness_for(str(programme_dir), leaf_sharing, superpose, multi_use=multi_use) — so collapse_insearch silently takes _fitness_for's default True. collapse_best has no collapse_insearch parameter, so evolve.py cannot thread the run's --collapse-insearch flag through even if it wanted to.\n\nConsequences, verified on 5 evolved harbor-house trees today: (1) the keep-better guard of collapse_finish is VACUOUS — base_fails is measured on a deepcopy that _evaluate_full re-collapses in-eval, so base == collapsed on 5/5 files (e.g. evolved-3M-nols-3.dom logs '12 -\u003e 12 (applied)' where the canonical evaluator shows the collapse actually did 15 -\u003e 12). The 94g safety property 'kept only if the fail count does not increase' is therefore not being checked against the true pre-collapse tree: a canonically fail-INCREASING collapse would be silently applied (collapse_global is 'monotone on harbor-house but not proven in general' per its own docstring — the guard exists precisely for that case). (2) The '[finish] collapse: N -\u003e M' log line under-reports the collapse's real effect (experiment logs quoting it understate 94g's contribution). (3) In a --no-collapse-insearch run the finish evaluator contradicts the run's objective outright — the deterministic 7ua mechanism, now in the default pipeline. (4) Minor: max_share and conn_grade are also not forwarded (matters for kpu/anneal and qi6 runs). Same pattern in search_annealed's final rescore branch: _evaluate(..., leaf_sharing=False, superpose=superpose) leaves _evaluate's collapse_insearch default True, and search_annealed has no way to pass the flag to it.\n\nOn the 5 probed files the returned tree's canonical fails happened to equal the reported number (the tree is a collapse fixpoint after iters=6 + 2-opt, so the extra in-eval collapse found nothing) — but that is not guaranteed, and the vacuous guard + misleading log line are unconditional.\n\nRecommended fix: add a collapse_insearch (and max_share/conn_grade) parameter to collapse_best, thread it from evolve.py, and make collapse_finish's keep-better measurement use a CANONICAL (collapse_insearch=False) evaluator regardless — the guard's job is to protect the canonical fail count of the written .dom, which homemaker-fitness scores with the on-disk config (no insearch override). Decide explicitly which objective the final 'best: N fails' report should quote (canonical is what the .dom.fails sidecar will say).","status":"open","priority":3,"issue_type":"bug","owner":"bruno@postle.net","created_at":"2026-08-02T08:19:41Z","created_by":"Bruno Postle","updated_at":"2026-08-02T08:19:41Z","dependency_count":0,"dependent_count":0,"comment_count":0}
{"id":"homemaker-py-sd3","title":"driver.collapse_best bakes collapse_insearch=True into its finish evaluator, making the 94g keep-better guard vacuous","description":"Found by the homemaker-py-zrx expert review; same family as homemaker-py-7ua but in the PRODUCT (driver.py), not the experiment script. driver.collapse_best builds its evaluator as _fitness_for(str(programme_dir), leaf_sharing, superpose, multi_use=multi_use) — so collapse_insearch silently takes _fitness_for's default True. collapse_best has no collapse_insearch parameter, so evolve.py cannot thread the run's --collapse-insearch flag through even if it wanted to.\n\nConsequences, verified on 5 evolved harbor-house trees today: (1) the keep-better guard of collapse_finish is VACUOUS — base_fails is measured on a deepcopy that _evaluate_full re-collapses in-eval, so base == collapsed on 5/5 files (e.g. evolved-3M-nols-3.dom logs '12 -\u003e 12 (applied)' where the canonical evaluator shows the collapse actually did 15 -\u003e 12). The 94g safety property 'kept only if the fail count does not increase' is therefore not being checked against the true pre-collapse tree: a canonically fail-INCREASING collapse would be silently applied (collapse_global is 'monotone on harbor-house but not proven in general' per its own docstring — the guard exists precisely for that case). (2) The '[finish] collapse: N -\u003e M' log line under-reports the collapse's real effect (experiment logs quoting it understate 94g's contribution). (3) In a --no-collapse-insearch run the finish evaluator contradicts the run's objective outright — the deterministic 7ua mechanism, now in the default pipeline. (4) Minor: max_share and conn_grade are also not forwarded (matters for kpu/anneal and qi6 runs). Same pattern in search_annealed's final rescore branch: _evaluate(..., leaf_sharing=False, superpose=superpose) leaves _evaluate's collapse_insearch default True, and search_annealed has no way to pass the flag to it.\n\nOn the 5 probed files the returned tree's canonical fails happened to equal the reported number (the tree is a collapse fixpoint after iters=6 + 2-opt, so the extra in-eval collapse found nothing) — but that is not guaranteed, and the vacuous guard + misleading log line are unconditional.\n\nRecommended fix: add a collapse_insearch (and max_share/conn_grade) parameter to collapse_best, thread it from evolve.py, and make collapse_finish's keep-better measurement use a CANONICAL (collapse_insearch=False) evaluator regardless — the guard's job is to protect the canonical fail count of the written .dom, which homemaker-fitness scores with the on-disk config (no insearch override). Decide explicitly which objective the final 'best: N fails' report should quote (canonical is what the .dom.fails sidecar will say).","notes":"Fixed. Two changes: (1) fitness.collapse_finish now forces collapse_insearch=False (canonical) for its own base_fails/cand_fails measurement, saving/restoring self._collapse_insearch around the two score_with_fails calls -- regardless of how the Fitness instance itself was configured, so the guard can never again compare a pre-collapsed base against a pre-collapsed candidate. (2) driver.collapse_best now builds its evaluator with collapse_insearch=False explicitly (hardcoded, not threaded -- canonical is always the right objective for the 94g guard and the final reported fail count, matching what homemaker-fitness reports for the written .dom with no override), and threads max_share/conn_grade through to _fitness_for for config parity. Also fixed the same-family bug in search_annealed's no-polish-budget final rescore branch (was silently defaulting to collapse_insearch=True via _evaluate's default; now reads collapse_insearch/multi_use from search_kw). evolve.py forwards conn_grade to collapse_best. Added a regression test (test_collapse_finish_guard_is_canonical_even_with_insearch_collapse_on) that builds a Fitness with collapse_insearch=True in conf and asserts base_f still reflects the true pre-collapse fail count. Full suite: 405 passed (same 5 pre-existing CP-SAT/reassign failures, confirmed present on main before this change, unrelated).","status":"in_progress","priority":3,"issue_type":"bug","assignee":"Bruno Postle","owner":"bruno@postle.net","created_at":"2026-08-02T08:19:41Z","created_by":"Bruno Postle","updated_at":"2026-08-05T06:46:54Z","started_at":"2026-08-04T23:41:55Z","dependency_count":0,"dependent_count":0,"comment_count":0}
{"id":"homemaker-py-d86","title":"Rigorously re-verify qpk/1ph historical numbers against the homemaker-py-iio fix","description":"homemaker-py-iio (fixed 2026-08-02) found a stale-leaf-share metadata leak\nin Fitness._collapse_value/_usage_quality that could corrupt one cell of\ncollapse_global's Hungarian assignment during any leaf_sharing+collapse\nrun -- i.e. essentially the entire \"full default stack\" used from\nhomemaker-py-x3b (leaf_sharing default-on) onward, including the very\nstudies that justified defaulting collapse_insearch on (94g, qpk/1ph, 8sh).\n\nA same-codebase fix-vs-no-fix re-run of the qpk protocol (harbor-house,\nbudget 2500, seeds 1-3) confirmed the bug demonstrably perturbs real\nper-seed outcomes under collapse_insearch=ON (2/3 seeds diverged by 5-8\nfails, non-directionally) -- see DESIGN.md §35 for full details. That\nre-run used TODAY's codebase, not the actual historical commit, and only 3\nharbor-house seeds, not the original seed sets -- so it establishes the bug\nwas real and non-trivial but does NOT establish whether 1ph's aggregate\nN=20 programme-house verdict (mean 7.95-\u003e7.10, paired t-test p~=0.028)\nwould have changed under the fix.\n\nThis issue is to do the rigorous version: check out the codebase near the\n1ph commit (~2026-07-24, \"post-qpk commits through 161\"), backport the iio\nfix there in an isolated worktree, and re-run the ACTUAL historical seed\nsets (programme-house N=20 seeds 1-20, harbor-house N=3 seeds 1-3) at the\n1ph protocol's exact parameters, comparing per-seed and aggregate results\nagainst the published numbers. Low priority: the qualitative direction of\nthe qpk/1ph conclusion is probably still right (noise is non-directional\nand the N=20 statistical margin is comfortably above the observed per-seed\nswing), this is about tightening confidence, not expecting a reversal.","notes":"homemaker-py-r5a (fixed 2026-08-02) also affects this: it is the COMMIT-door companion to iio (a leaf relabelled back to its own stale share_type resurrects a stale multiplicity credit). Any re-verification run here should use the codebase state after BOTH iio and r5a, not iio alone.","status":"open","priority":3,"issue_type":"task","owner":"bruno@postle.net","created_at":"2026-08-02T06:53:51Z","created_by":"Bruno Postle","updated_at":"2026-08-02T09:44:35Z","dependency_count":0,"dependent_count":0,"comment_count":0}
{"id":"homemaker-py-7ua","title":"run_staged_search.py final rescore omits collapse_insearch override, causing false MISMATCH under leaf-sharing","description":"experiments/run_staged_search.py's _native_score() (used for the final 're-scored (native): ... -\u003e OK/MISMATCH' sanity line) calls fitness.load_config(programme_dir) with NO overrides, but driver.search_staged's internal evaluator always runs with collapse_insearch=True (baked into driver.search's default, search_staged has no param to disable it). The script's monkeypatched fitness.load_config only injects leaf_sharing/share_edge_cap/multi_use, not collapse_insearch, so the final rescore conf silently diverges from the search-time conf whenever leaf_sharing is on (the current default stack). Observed during homemaker-py-91f: a WORKERS=4 budget=2000 harbor-house run reported best fails=38 during search but re-scored fails=34 -\u003e MISMATCH (partly parallel non-determinism per homemaker-py-b8g, but the missing collapse_insearch override is a separate, deterministic contributor). Fix: add collapse_insearch=True to the monkeypatched conf alongside leaf_sharing/share_edge_cap.","status":"open","priority":3,"issue_type":"bug","owner":"bruno@postle.net","created_at":"2026-08-01T11:32:58Z","created_by":"Bruno Postle","updated_at":"2026-08-01T11:32:58Z","dependency_count":0,"dependent_count":0,"comment_count":0}
{"id":"homemaker-py-b8g","title":"Investigate parallel/BLAS non-determinism noise source in n_workers\u003e1 runs","description":"DESIGN.md §14 (psk, island-model experiment) flagged a real, uninvestigated noise source: 'Phase A is unaffected by the probe, yet harbor seed 2 scored 71 then 73 on byte-identical re-runs -- parallel/BLAS non-determinism, the same +/-2-3 effect §12.4 flagged.' This is DISTINCT from the homemaker-py-xcy bug (ProcessPoolExecutor as_completed ordering), which was fixed and made same-worker-count parallel runs reproducible for the SEARCH TRAJECTORY. This remaining noise is at the SCORING level (a single fitness eval on a fixed genome apparently returning different fail counts across runs), plausibly numpy/scipy BLAS thread nondeterminism in the geometry/inner-loop math. It was never root-caused or fixed, and it widens the error bars on every A/B in this log run at n_workers\u003e1 (the great majority of them, since serial sweeps are expensive). Investigate: reproduce minimally (score the same frozen .dom N times under workers\u003e1), bisect whether it's BLAS threading (try OMP_NUM_THREADS=1/OPENBLAS_NUM_THREADS=1), floating-point summation order, or something else; fix or document a mitigation (e.g. pin thread count in worker processes).","design":"Reference: DESIGN.md §14 'Noise caveat (carry forward)', §12.4 (homemaker-py-xcy, the related-but-distinct trajectory-ordering bug already fixed). If the cause is BLAS thread count, the fix is likely a one-line env pin in the worker pool initializer (driver.py's ProcessPoolExecutor setup).","notes":"homemaker-py-zrx review (2026-08-02) found a concrete, non-BLAS candidate mechanism for part of this noise in PARALLEL STAGED runs: homemaker-py-cvw — substrate_readiness in the parent process reads stale id()-keyed geometry cache entries (24/300 corrupted in a churn probe, worst error ~1.0), perturbing stage-1 selection address-dependently across byte-identical re-runs. Does not explain fixed-genome single-eval divergence (if that was ever actually isolated); re-test after cvw lands before chasing BLAS.","status":"open","priority":3,"issue_type":"bug","owner":"bruno@postle.net","created_at":"2026-08-01T10:07:45Z","created_by":"Bruno Postle","updated_at":"2026-08-02T08:20:13Z","dependency_count":0,"dependent_count":1,"comment_count":0}
@ -129,27 +129,27 @@
{"id":"homemaker-py-erc.6","title":"Experiment: inner-loop slack-expansion objective term","description":"Inner-loop counterpart to plot-fill construction. If Diagnostic B shows the inner loop has room to expand leaves into slack but no objective gradient to do so (the scalar rewards hitting target area but not exceeding it where slack exists), add a term/incentive so the ratio optimiser pushes leaf boundaries out to consume neighbouring slack and satisfy size, rather than parking at target.\n\nCONDITIONAL on Diagnostic B: build this only if B localizes the gap to the inner loop (room to expand, no gradient); if B shows construction targets too-small dims, prefer the plot-fill construction sibling. Must preserve the §5.4 inner-loop cliff / §4.9 lexicographic protection — the term sits where it cannot displace the fail-count ordering. A/B vs §12.2 baseline, seeds 0/1/2, 20000 evals, staged, default-OFF. Record DESIGN.md §13.6.","notes":"DEPRIORITISED by Diagnostic B (§13.2). B shows the inner loop CANNOT repair undersize: the slack is depth-driven maldistribution baked into the frozen topology, and the equal-offset ratio DOF cannot shrink a 14x leaf to feed a starved one without trading into shape fails (0.5^n cliff). Wrong DOF and wrong direction — the blocker is slicing POSITION, not a missing expansion reward. Fix belongs upstream in construction/topology (erc.4 re-scoped, erc.3). Keep as a low-priority follow-up only if a depth-balanced construction still leaves a residual size gradient the inner loop could pick up.","status":"closed","priority":4,"issue_type":"feature","owner":"bruno@postle.net","created_at":"2026-06-22T23:16:24Z","created_by":"Bruno Postle","updated_at":"2026-06-28T13:22:22Z","closed_at":"2026-06-28T13:22:22Z","close_reason":"wont-fix (DESIGN §13.7): Diag B (§13.2) showed the inner loop cannot repair undersize (wrong DOF — slicing position, frozen-topology ratios). Superseded by depth-balanced construction (erc.4). Condition unmet.","dependencies":[{"issue_id":"homemaker-py-erc.6","depends_on_id":"homemaker-py-erc","type":"parent-child","created_at":"2026-06-23T00:16:23Z","created_by":"Bruno Postle","metadata":"{}"},{"issue_id":"homemaker-py-erc.6","depends_on_id":"homemaker-py-erc.2","type":"blocks","created_at":"2026-06-23T00:16:47Z","created_by":"Bruno Postle","metadata":"{}"}],"dependency_count":1,"dependent_count":0,"comment_count":0}
{"id":"homemaker-py-erc.5","title":"Experiment: compactness-aware cuts (minimize leaf perimeter/area)","description":"Attacks the #1 factor, crinkliness (346) — a per-leaf perimeter/area property DISTINCT from proportion (aspect ratio). Proportion-aware seeding (leu.2) sizes splits but does not bias toward balanced, square-ish subdivision. Add a KD-tree-style 'keep both children compact' cut rule (prefer the cut orientation/position that minimises summed child perimeter/area) in construction.\n\nCONDITIONAL on Diagnostic A: if A shows per-leaf shape-fail is FLAT across densities (floor intrinsic to slicing density), better cuts at the same leaf count will not pay → this should be closed wont-fix in favour of leaf-sharing. Only build if A shows shape-fail RISES with density. A/B vs §12.2 baseline, seeds 0/1/2, 20000 evals, staged, default-OFF. Record DESIGN.md §13.5.","notes":"DEPRIORITISED by erc.1 verdict (§13.1): per-leaf shape-fail flat vs slicing density and cuts already squarest (_size_divisions_from_targets picks squarest rotation) yet still ~1.8 fails/leaf =\u003e little compactness headroom at fixed leaf count. Floor is intrinsic to leaf COUNT, not cut quality. Revisit only if leaf-sharing (erc.3) underdelivers.","status":"closed","priority":4,"issue_type":"feature","owner":"bruno@postle.net","created_at":"2026-06-22T23:16:21Z","created_by":"Bruno Postle","updated_at":"2026-06-28T13:22:17Z","closed_at":"2026-06-28T13:22:17Z","close_reason":"wont-fix (DESIGN §13.7): Diag A (§13.1) showed the floor is intrinsic to leaf COUNT not cut quality; revisit condition was 'only if leaf-sharing underdelivers' but leaf-sharing OVER-delivered (−32…−39%, §13.3). Condition unmet.","dependencies":[{"issue_id":"homemaker-py-erc.5","depends_on_id":"homemaker-py-erc","type":"parent-child","created_at":"2026-06-23T00:16:21Z","created_by":"Bruno Postle","metadata":"{}"},{"issue_id":"homemaker-py-erc.5","depends_on_id":"homemaker-py-erc.1","type":"blocks","created_at":"2026-06-23T00:16:43Z","created_by":"Bruno Postle","metadata":"{}"}],"dependency_count":1,"dependent_count":0,"comment_count":0}
{"id":"homemaker-py-2g5","title":"Rebuild occlusion/daylight/sun subsystem in Python (post-Phase-5, after optimisation fully native)","description":"DESIGN.md §6 port scope — a whole subsystem, not a term. quality_daylight (Leaf.pm:281-296) needs Urb::Misc::Sun + Urb::Field::Occlusion (+CIESky); quality_uncrinkliness also takes the occlusion object. Indoor spaces return 1 for daylight; cost is outdoor spaces + crinkliness. Port Sun_horizontal (262980-minute normalisation) and the occlusion wall set from Dom-\u003eWalls.","acceptance_criteria":"Daylight and crinkliness factors match Perl (float tolerance) across the corpus, including multi-storey cases","notes":"Re-scoped 2026-06-12: occlusion disabled in the Urb oracle instead of ported (see homemaker-py-gp2). Native fitness ships with simple crinkliness (illumination factor = 1, in homemaker-py-gnw). This issue is now the eventual Python occlusion rebuild, only after optimisation works entirely in Python. Restores outdoor-daylight and shaded-wall selection pressure.\nReframed 2026-06-17: orthogonal to epic homemaker-py-c4c. This is fitness FIDELITY (restoring daylight + shaded-wall selection pressure to match Perl), not search CAPABILITY — it changes what 'good' means, not the search's ability to find good. It will NOT improve final designs in the sense currently sought. Stays P4, deferred until the topology-search-quality epic lands and optimisation is fully native.","status":"open","priority":4,"issue_type":"feature","owner":"bruno@postle.net","created_at":"2026-06-11T23:38:25Z","created_by":"Bruno Postle","updated_at":"2026-06-17T19:14:48Z","dependency_count":0,"dependent_count":0,"comment_count":0}
{"_type":"memory","key":"proportion-aware-constructive-seeding-leu-2-12-2","value":"Proportion-aware constructive seeding (leu.2/§12.2): sizing seed cuts from target AREAS only regresses (thin slivers wreck aspect); you must ALSO pick each cut's rotation for child squareness. It is a convergence ACCELERATOR via a deeper local optimum around the constructed topology: wins where that topology is roughly right and budget is scarce (harbor -13%, maple -10% at 20k evals) but DELAYS small programmes where the seed must be restructured by undivide (programme-house regresses at fixed budget, yet reaches the floor given budget - speed, not asymptote). Default-on. Also: n_storeys must honour storey_minimum, not just level: keys (programme-house storey_minimum:2, all rooms level:0 - was seeded 1 storey short; cq1)."}
{"_type":"memory","key":"homemaker-py-3l6-fix-leaf-sharing-evolve-runs","value":"homemaker-py-3l6 fix: leaf-sharing evolve runs now auto-finish before write via driver.polish_finish — unfold_shared_leaves() then a warm-started leaf_sharing=False polish search (--polish-budget, default budget//2). Makes the written .dom honest under canonical homemaker-fitness (internal==canonical when leaf_sharing off). Interrupt path forces polish_budget=0 (unfold+rescore only). This is yaa's unfold-then-polish, made automatic; Schedule B annealing is still kpu."}
{"_type":"memory","key":"island-model-psk-14-is-a-null-priming","value":"Island model (psk, §14) is a NULL: priming a population from N converged independent elites + crossover-heavy migration does not beat best-of-N at equal total budget (maple island 124 vs control 116). The child_probe instrument shows WHY: area-matched crossover across independently-converged elites almost never synthesizes (1-3 of ~64 children beat the better parent, max drop 2-5) because the slicing encoding is non-canonical (9gp), so splices are disruptive not combinatorial. Search-machinery null #3 after graded-objective and niching/restarts; residual stays geometry/shape-bound."}
{"_type":"memory","key":"user-preference-bruno-this-is-a-fedora-system","value":"User preference (Bruno): this is a Fedora system — NEVER install Python packages via pip without asking first; always ask whether to install the rpm via dnf (e.g. python3-cma) before considering pip. Applies to any dependency additions."}
{"_type":"memory","key":"collapse-global-94g-and-any-label-usage-optimisation","value":"collapse_global (94g) and any label/usage optimisation CANNOT fix geometry-intrinsic fails. The harbor-house 15-fail best layout contains long-thin cells that are useless whatever room usage is assigned — their width/proportion/crinkliness fails are shape-bound, not label slack. Two consequences: (1) do not over-claim collapse gains — only ~2-3 of that layout's fails are reclaimable relabel slack, the rest are geometry- or building-level bound; (2) the threshold objective must not be tuned to 'pass' a degenerate cell via a permissive room type — a metric-pass on a physically useless space is gaming, not a fix. Real remedies for these are geometry/topology search (cell shape) and circulation placement, filed separately, not the collapse."}
{"_type":"memory","key":"collapse-global-s-jacobi-adjacency-relaxation-homemaker-py","value":"collapse_global's Jacobi adjacency relaxation (homemaker-py-94g) is a synchronous per-round linear-assignment re-solve, which can 2-cycle indefinitely between two labellings that each satisfy ZERO adjacency requirements even though a permutation satisfying ALL of them exists -- proven on a minimal 4-cell chain (p1-q1-p2-q2, two disjoint adjacency pairs p1\u003c-\u003ep2/q1\u003c-\u003eq2) in test_two_opt_polish_escapes_jacobi_plateau. homemaker-py-9wi added Fitness._two_opt_adjacency_polish: a same-level pairwise-swap local search run after the Jacobi fixpoint, gated behind collapse_global(local_search=True) (default off, exposed as homemaker-collapse --local-search). Monotone by construction (a swap is kept only if it strictly increases total reward). Empirically on the 11 harbor-house evolved-*.dom/3m.dom/materialised-3M.dom layouts: 10 matched Jacobi-only exactly, 0 regressed, and evolved-anneal-3M.dom improved 21-\u003e19 fails (fixed a genuine mutual da1\u003c-\u003ek1 adjacency miss the Jacobi loop couldn't reach)."}
{"_type":"memory","key":"multi-storey-staircase-consistency-when-dividing-or-retyping","value":"Multi-storey staircase consistency: when dividing or retyping a circulation (C) leaf at one level, the same structural change should be propagated to the matching leaf on ALL other storeys so the stair core path is maintained. The optimizer cannot fix staircase disruptions through trial-and-error geometry alone — it requires a synchronized multi-level operator that applies the same topology change to every storey simultaneously."}
{"_type":"memory","key":"urb-oracle-nondeterminism-urb-fitness-pl-output-varies","value":"Urb oracle nondeterminism: urb-fitness.pl output varies run-to-run from Perl hash-order randomisation — .fails line ORDER shuffles (compare sorted, use oracle.Score.fail_lines) and the score float can flip by ~1 ULP (compare with math.isclose rel_tol=1e-12, never ==). Not a batching artifact; affects single runs too. Matters for the Phase 3 native-fitness parity gate (homemaker-py-uxz)."}
{"_type":"memory","key":"9o5-multi-use-leaves-is-path-a-superposition","value":"9o5 multi-use leaves is path (a) — superposition as SEARCH RELAXATION that COLLAPSES to specific usage at the end, NOT path (b) loose-fit/no-collapse. Bruno's intent: codes with SIMILAR leaf requirements form an interchangeable equivalence class; during evolution the solver doesn't commit which leaf serves which specific usage (smoother landscape, no fighting over exact leaf usage); at the end the layout is CONDENSED to specific usages by brute-forcing the in-class assignment (3 interchangeable usages over 3 leaves = 3! = 6 combinations to check, pick best). 'Derive automatically' compatibility = requirement-similarity grouping. This reverses the issue's stated 'path b preferred' note."}
{"_type":"memory","key":"homemaker-py-pythonpath-set-pythonpath-home-bruno-src","value":"homemaker-layout PYTHONPATH: package installed as 'homemaker-layout' via pip install -e . so 'import homemaker_layout' works from anywhere without PYTHONPATH. For running tests use 'python -m pytest' from project root /home/bruno/src/homemaker-layout (pyproject.toml adds src/ automatically). Never try pip show homemaker — that's the old homemaker-addon conflict."}
{"_type":"memory","key":"ld2-13-6-interior-o-seed-diagnostic-all","value":"ld2/§13.6 interior-O seed diagnostic: ALL crinkliness fails in the constructed bal+share seed are UNDER-exposed (crink\u003c0.62, landlocked rooms with no facade + no uncovered-O neighbour) — zero over-exposed sliver fails. So the erc crinkliness residual is genuine under-daylighting, validating the interior light-well premise. Default outside_divisor=6 was too sparse (null: harbor 147-\u003e142, crinkliness even rose). odiv=3 is the seed-optimal joint setting: harbor seed fails 147-\u003e129 (-18), maple 219-\u003e206 (-14), landlocked fails drop, at cost of more leaves (harbor +4, maple +8). Because it ADDS leaves it carries the §13.4 wash-out risk; A/B to convergence pending."}
{"_type":"memory","key":"unfold-strategy-for-shared-leaves-homemaker-py-8iv","value":"Unfold strategy for shared leaves (homemaker-py-8iv, resolved 2026-07-16): use the BALANCED GRID (operators._grow_balanced/_size_subtree_equal), NOT circulation-aware slicing. Slicing a shared leaf perpendicular to its access edge so every child touches the corridor was implemented + A/B-tested and LOST decisively (150k-eval warm-start polish from evolved-3M: slice 41 fails/3.5e-14 vs grid 25 fails/2.4e-09, grid ahead at every milestone). Reason: k rooms all touching one wall are intrinsically thin slices; that geometric debt (proportion/long/width) is unfixable without topology change, while the grid's squarer children let local search re-route access cheaply via level_retype/place_missing/level_fix. Lesson: at the sharing-\u003eno-sharing transition, prioritise squarer children and leave access to local search; do not reintroduce slicing in Schedule B (kpu)."}
{"_type":"memory","key":"urb-fitness-bug-found-fixed-2026-06-12","value":"Urb fitness bug found+fixed 2026-06-12 (patch in /home/bruno/src/urb, uncommitted): ProgrammeDriven.pm ratio_o/ratio_type grepped case-insensitively over the ratios hash and took the FIRST key — nondeterministic (x4.5 score swings) for designs with mixed-case type classes (both 'c' circulation and 'C' covered). Fixed to SUM the class (matches Is_Circulation//Is_Outside semantics); 35/35 corpus scores unchanged. CRITICAL for homemaker-py-3y7/gnw: the native port must implement class-SUM ratios. Building.pm has the same unpatched pattern (site-driven path, not used by our oracle). Also: the memetic search reward-hacked this bug before the fix — search results predating it are noise artifacts."}
{"_type":"memory","key":"adjacency-in-binary-slicing-tree-is-structural-not","value":"Adjacency in binary slicing tree is structural, not geometric: the inner-loop NM cannot fix topological adjacency failures. Two paths exist: (1) tree-sibling adjacency — a node is adjacent to its sibling in the tree; (2) cross-zone geometric adjacency — leaves from different subtrees that happen to share a boundary. Staircase/adjacency fails require a topology mutation that changes which nodes are siblings or which zones touch. This was proved empirically on programme-house: staircase fail from rot=0 layout could not be fixed by NM but was fixed by level_retype creating a two-C topology (2026-06-14/15)."}
{"_type":"memory","key":"cli-tool-style-prefer-python-m-homemaker-module","value":"CLI tool style: prefer python -m homemaker.module --parameters pattern, installable via pip install -e . with pyproject.toml entry_points. Not standalone bin/ scripts."}
{"_type":"memory","key":"deceptive-valleys-in-topology-search-when-every-single","value":"Deceptive valleys in topology search: when every single-step mutation from a target state passes through a high-fail intermediary (e.g. level_fix displaces a room into 5+ new fails), a compound operator that atomically applies two coordinated changes can escape. Design compound operators to land on the low-fail state directly, bypassing the deceptive gradient. Programme-house example: level_compound_fix atomically moves the level-constrained room AND re-inserts the displaced room adjacent to C in one step (operators.py, 2026-06-14)."}
{"_type":"memory","key":"never-use-corpus-filenames-candidate-001-dom-candidate","value":"Never use corpus filenames (candidate-001.dom, candidate-002.dom, generated.dom, init.dom, etc.) as --output targets when running experiments. These are test fixtures. Always write experimental outputs to scratch/ or a timestamped path. Lesson from 2026-06-14: warm-start runs overwrote candidate-001/002.dom and broke graph tests."}
{"_type":"memory","key":"experiment-harness-gotcha-the-leaf-sharing-relaxed-objective","value":"Experiment harness gotcha: the leaf-sharing RELAXED objective (§13.3) is injected ONLY by monkeypatching fitness.load_config in the parent process (run_staged_search.py / probe scripts). This is parent-process-only and does NOT propagate into ProcessPoolExecutor workers (n_workers\u003e1), which re-import fitness fresh and score under the STRICT on-disk patterns.config -\u003e r.n_fails MISMATCH (worker strict vs parent relaxed re-score). ALL §13.x floor runs were therefore SERIAL. Any future PARALLEL leaf-sharing experiment will silently mis-score until leaf_sharing lives on disk/CLI (tracked: homemaker-py-x3b). The parallel driver itself is correct; both paths score via load_config(programme_dir)."}
{"_type":"memory","key":"run-to-run-reproducibility-in-homemaker-layout-serial","value":"Run-to-run reproducibility in homemaker-layout: serial search (workers=1) is byte-for-byte deterministic; parallel (workers\u003e1) is now deterministic too AFTER fixing driver._run_batch to admit futures in submission order (was as_completed/completion order, bug xcy). Reproducibility holds only for a FIXED worker count — serial vs parallel differ because children-per-iteration is 1 vs n_workers (different batch granularity), which is expected, not a bug. The constructive seeder was NEVER nondeterministic: _assign_adjacency_aware has unique idx tiebreaks; comparing topologies with Python builtin hash() of the signature STRING is invalid (PYTHONHASHSEED salts str hashing per process) — use a stable hash (sha1) or genome.signature equality."}
{"_type":"memory","key":"strategy-decision-2026-06-12-bruno-occlusion-daylight","value":"Strategy decision 2026-06-12 (Bruno): occlusion/daylight is ORTHOGONAL to building a scalable optimiser. Disable it in Urb (env flag, homemaker-py-gp2) rather than port it; native fitness uses simple crinkliness (illumination factor = 1); rebuild occlusion in Python only after optimisation is fully native (homemaker-py-2g5, now P4). Consequence: all scores change when the flag flips — re-baseline corpus/.score, DESIGN \\$4.5 gains, gate bars at one clean boundary AFTER homemaker-py-1p0 closes; Phase-2 urb-evolve benchmark must run with the same flag."}
{"_type":"memory","key":"urb-oracle-nondeterminism-urb-fitness-pl-output-varies","value":"Urb oracle nondeterminism: urb-fitness.pl output varies run-to-run from Perl hash-order randomisation — .fails line ORDER shuffles (compare sorted, use oracle.Score.fail_lines) and the score float can flip by ~1 ULP (compare with math.isclose rel_tol=1e-12, never ==). Not a batching artifact; affects single runs too. Matters for the Phase 3 native-fitness parity gate (homemaker-py-uxz)."}
{"_type":"memory","key":"user-preference-bruno-this-is-a-fedora-system","value":"User preference (Bruno): this is a Fedora system — NEVER install Python packages via pip without asking first; always ask whether to install the rpm via dnf (e.g. python3-cma) before considering pip. Applies to any dependency additions."}
{"_type":"memory","key":"ld2-13-6-interior-o-seed-diagnostic-all","value":"ld2/§13.6 interior-O seed diagnostic: ALL crinkliness fails in the constructed bal+share seed are UNDER-exposed (crink\u003c0.62, landlocked rooms with no facade + no uncovered-O neighbour) — zero over-exposed sliver fails. So the erc crinkliness residual is genuine under-daylighting, validating the interior light-well premise. Default outside_divisor=6 was too sparse (null: harbor 147-\u003e142, crinkliness even rose). odiv=3 is the seed-optimal joint setting: harbor seed fails 147-\u003e129 (-18), maple 219-\u003e206 (-14), landlocked fails drop, at cost of more leaves (harbor +4, maple +8). Because it ADDS leaves it carries the §13.4 wash-out risk; A/B to convergence pending."}
{"_type":"memory","key":"warm-x0-initialization-bug-pattern-when-a-topology","value":"warm_x0 initialization bug pattern: when a topology operator explicitly sets division ratios on a newly-created node (e.g. compound_fix sets node.division=[0.25,0.25] for t3), parent.ratios has no entry for that node (it was a leaf). warm_x0 defaults it to 0.5, corrupting the inner loop's starting point and making the operator invisible to lex comparison. Fix: only propagate child ratios for nodes where the parent node was NOT already divided; stale hidden nodes revealed by structural mutations (swap flipping b.below) must NOT contribute their pre-writeback values. See driver.py lines 259-267 (fixed 2026-06-14)."}
{"_type":"memory","key":"cli-tool-style-prefer-python-m-homemaker-module","value":"CLI tool style: prefer python -m homemaker.module --parameters pattern, installable via pip install -e . with pyproject.toml entry_points. Not standalone bin/ scripts."}
{"_type":"memory","key":"homemaker-py-3l6-fix-leaf-sharing-evolve-runs","value":"homemaker-py-3l6 fix: leaf-sharing evolve runs now auto-finish before write via driver.polish_finish — unfold_shared_leaves() then a warm-started leaf_sharing=False polish search (--polish-budget, default budget//2). Makes the written .dom honest under canonical homemaker-fitness (internal==canonical when leaf_sharing off). Interrupt path forces polish_budget=0 (unfold+rescore only). This is yaa's unfold-then-polish, made automatic; Schedule B annealing is still kpu."}
{"_type":"memory","key":"homemaker-py-pythonpath-set-pythonpath-home-bruno-src","value":"homemaker-layout PYTHONPATH: package installed as 'homemaker-layout' via pip install -e . so 'import homemaker_layout' works from anywhere without PYTHONPATH. For running tests use 'python -m pytest' from project root /home/bruno/src/homemaker-layout (pyproject.toml adds src/ automatically). Never try pip show homemaker — that's the old homemaker-addon conflict."}
{"_type":"memory","key":"unfold-strategy-for-shared-leaves-homemaker-py-8iv","value":"Unfold strategy for shared leaves (homemaker-py-8iv, resolved 2026-07-16): use the BALANCED GRID (operators._grow_balanced/_size_subtree_equal), NOT circulation-aware slicing. Slicing a shared leaf perpendicular to its access edge so every child touches the corridor was implemented + A/B-tested and LOST decisively (150k-eval warm-start polish from evolved-3M: slice 41 fails/3.5e-14 vs grid 25 fails/2.4e-09, grid ahead at every milestone). Reason: k rooms all touching one wall are intrinsically thin slices; that geometric debt (proportion/long/width) is unfixable without topology change, while the grid's squarer children let local search re-route access cheaply via level_retype/place_missing/level_fix. Lesson: at the sharing-\u003eno-sharing transition, prioritise squarer children and leave access to local search; do not reintroduce slicing in Schedule B (kpu)."}
{"_type":"memory","key":"correction-to-urb-fitness-bug-memory-bruno-2026","value":"CORRECTION to urb-fitness-bug memory (Bruno, 2026-06-12): 'C' is NOT a 'covered' type — Is_Covered is a geometric predicate (indoor space above). Urb's generic types are canonically UPPERCASE: C=circulation, O=outside, S=sahn (get_space_types qw/C O S/; corpus is 100% uppercase, never 'c'/'o' leaves). The mixed-case designs that fired the latent ratio_type first-match bug were created by homemaker's own operator type pool emitting lowercase 'c'/'o' — fixed: driver/operators now emit uppercase generics only, and class checks use t[0].lower() in 'cos'. The Urb class-sum patch stays as defensive hardening (zero impact on canonical designs). Native port (3y7/gnw): treat type classes case-insensitively, generics canonically uppercase."}
{"_type":"memory","key":"9o5-multi-use-leaves-is-path-a-superposition","value":"9o5 multi-use leaves is path (a) — superposition as SEARCH RELAXATION that COLLAPSES to specific usage at the end, NOT path (b) loose-fit/no-collapse. Bruno's intent: codes with SIMILAR leaf requirements form an interchangeable equivalence class; during evolution the solver doesn't commit which leaf serves which specific usage (smoother landscape, no fighting over exact leaf usage); at the end the layout is CONDENSED to specific usages by brute-forcing the in-class assignment (3 interchangeable usages over 3 leaves = 3! = 6 combinations to check, pick best). 'Derive automatically' compatibility = requirement-similarity grouping. This reverses the issue's stated 'path b preferred' note."}
{"_type":"memory","key":"experiment-seeding-pitfall-run-search-scaled-py-s","value":"Experiment seeding pitfall: run_search_scaled.py's default PH_SEED (c964…dom) is a FINISHED programme-house design — passing it warm-starts and floors at ~3 fails, NOT a blank-slate topology search. For blank-slate runs comparable to §11.5/§11.6 baselines, seed from examples/programme-house/init.dom (a bare undivided plot; driver bootstrap auto-triggers only on bare plots). Bit the 6zy sweep — first pass used c964 and falsely showed 3-fail floor across the whole grid."}
{"_type":"memory","key":"island-model-psk-14-is-a-null-priming","value":"Island model (psk, §14) is a NULL: priming a population from N converged independent elites + crossover-heavy migration does not beat best-of-N at equal total budget (maple island 124 vs control 116). The child_probe instrument shows WHY: area-matched crossover across independently-converged elites almost never synthesizes (1-3 of ~64 children beat the better parent, max drop 2-5) because the slicing encoding is non-canonical (9gp), so splices are disruptive not combinatorial. Search-machinery null #3 after graded-objective and niching/restarts; residual stays geometry/shape-bound."}
{"_type":"memory","key":"urb-fitness-bug-found-fixed-2026-06-12","value":"Urb fitness bug found+fixed 2026-06-12 (patch in /home/bruno/src/urb, uncommitted): ProgrammeDriven.pm ratio_o/ratio_type grepped case-insensitively over the ratios hash and took the FIRST key — nondeterministic (x4.5 score swings) for designs with mixed-case type classes (both 'c' circulation and 'C' covered). Fixed to SUM the class (matches Is_Circulation//Is_Outside semantics); 35/35 corpus scores unchanged. CRITICAL for homemaker-py-3y7/gnw: the native port must implement class-SUM ratios. Building.pm has the same unpatched pattern (site-driven path, not used by our oracle). Also: the memetic search reward-hacked this bug before the fix — search results predating it are noise artifacts."}
{"_type":"memory","key":"collapse-global-94g-and-any-label-usage-optimisation","value":"collapse_global (94g) and any label/usage optimisation CANNOT fix geometry-intrinsic fails. The harbor-house 15-fail best layout contains long-thin cells that are useless whatever room usage is assigned — their width/proportion/crinkliness fails are shape-bound, not label slack. Two consequences: (1) do not over-claim collapse gains — only ~2-3 of that layout's fails are reclaimable relabel slack, the rest are geometry- or building-level bound; (2) the threshold objective must not be tuned to 'pass' a degenerate cell via a permissive room type — a metric-pass on a physically useless space is gaming, not a fix. Real remedies for these are geometry/topology search (cell shape) and circulation placement, filed separately, not the collapse."}
{"_type":"memory","key":"experiment-harness-gotcha-the-leaf-sharing-relaxed-objective","value":"Experiment harness gotcha: the leaf-sharing RELAXED objective (§13.3) is injected ONLY by monkeypatching fitness.load_config in the parent process (run_staged_search.py / probe scripts). This is parent-process-only and does NOT propagate into ProcessPoolExecutor workers (n_workers\u003e1), which re-import fitness fresh and score under the STRICT on-disk patterns.config -\u003e r.n_fails MISMATCH (worker strict vs parent relaxed re-score). ALL §13.x floor runs were therefore SERIAL. Any future PARALLEL leaf-sharing experiment will silently mis-score until leaf_sharing lives on disk/CLI (tracked: homemaker-py-x3b). The parallel driver itself is correct; both paths score via load_config(programme_dir)."}
{"_type":"memory","key":"programme-house-optimisation-result-2026-06-14-15","value":"Programme-house optimisation result (2026-06-14/15): best achievable is 1 fail (l1 wrong level, score ~0.005). 0 fails is geometrically impossible: l1 (min 27m²) must occupy ll (~23m²) at level 0, which eliminates the t3-adj-C provider; dividing ll into lll(l1)+llr(C) gives llr proportion ~6:1 (fails). Python memetic optimizer achieves 1 fail in 50k evals vs Perl optimiser's 2-3 fails. Winning topology: TWO C nodes at level 0 — ll(C) for t3-adj-C via geometric contact, rl(C) for staircase via tree-sibling adjacency to rrr(O). Best .dom: scratch/from-warmstart-fixed.dom and scratch/from-compound3-fixed.dom."}
{"_type":"memory","key":"never-use-corpus-filenames-candidate-001-dom-candidate","value":"Never use corpus filenames (candidate-001.dom, candidate-002.dom, generated.dom, init.dom, etc.) as --output targets when running experiments. These are test fixtures. Always write experimental outputs to scratch/ or a timestamped path. Lesson from 2026-06-14: warm-start runs overwrote candidate-001/002.dom and broke graph tests."}
{"_type":"memory","key":"proportion-aware-constructive-seeding-leu-2-12-2","value":"Proportion-aware constructive seeding (leu.2/§12.2): sizing seed cuts from target AREAS only regresses (thin slivers wreck aspect); you must ALSO pick each cut's rotation for child squareness. It is a convergence ACCELERATOR via a deeper local optimum around the constructed topology: wins where that topology is roughly right and budget is scarce (harbor -13%, maple -10% at 20k evals) but DELAYS small programmes where the seed must be restructured by undivide (programme-house regresses at fixed budget, yet reaches the floor given budget - speed, not asymptote). Default-on. Also: n_storeys must honour storey_minimum, not just level: keys (programme-house storey_minimum:2, all rooms level:0 - was seeded 1 storey short; cq1)."}
{"_type":"memory","key":"multi-storey-staircase-consistency-when-dividing-or-retyping","value":"Multi-storey staircase consistency: when dividing or retyping a circulation (C) leaf at one level, the same structural change should be propagated to the matching leaf on ALL other storeys so the stair core path is maintained. The optimizer cannot fix staircase disruptions through trial-and-error geometry alone — it requires a synchronized multi-level operator that applies the same topology change to every storey simultaneously."}
{"_type":"memory","key":"warm-x0-initialization-bug-pattern-when-a-topology","value":"warm_x0 initialization bug pattern: when a topology operator explicitly sets division ratios on a newly-created node (e.g. compound_fix sets node.division=[0.25,0.25] for t3), parent.ratios has no entry for that node (it was a leaf). warm_x0 defaults it to 0.5, corrupting the inner loop's starting point and making the operator invisible to lex comparison. Fix: only propagate child ratios for nodes where the parent node was NOT already divided; stale hidden nodes revealed by structural mutations (swap flipping b.below) must NOT contribute their pre-writeback values. See driver.py lines 259-267 (fixed 2026-06-14)."}
{"_type":"memory","key":"correction-to-urb-fitness-bug-memory-bruno-2026","value":"CORRECTION to urb-fitness-bug memory (Bruno, 2026-06-12): 'C' is NOT a 'covered' type — Is_Covered is a geometric predicate (indoor space above). Urb's generic types are canonically UPPERCASE: C=circulation, O=outside, S=sahn (get_space_types qw/C O S/; corpus is 100% uppercase, never 'c'/'o' leaves). The mixed-case designs that fired the latent ratio_type first-match bug were created by homemaker's own operator type pool emitting lowercase 'c'/'o' — fixed: driver/operators now emit uppercase generics only, and class checks use t[0].lower() in 'cos'. The Urb class-sum patch stays as defensive hardening (zero impact on canonical designs). Native port (3y7/gnw): treat type classes case-insensitively, generics canonically uppercase."}

27
src/homemaker_layout/driver.py
View file

@ -842,6 +842,8 @@ def collapse_best(
leaf_sharing: bool = False,
superpose: bool = False,
multi_use: bool = False,
max_share: int | None = None,
conn_grade: bool = False,
log=None,
**collapse_kw,
) -> SearchResult:
@ -855,11 +857,23 @@ def collapse_best(
geometry, so it cannot touch shape-intrinsic fails (long-thin cells, etc.).
Updates ``result.best`` in place with the canonically re-scored relabelling
when it helps; otherwise leaves the result untouched."""
when it helps; otherwise leaves the result untouched.
homemaker-py-sd3: the evaluator built here is deliberately CANONICAL
(``collapse_insearch=False``) regardless of whether the run being finished
used in-search collapse — both the keep-better guard and the reported
post-collapse fail count must match what ``homemaker-fitness`` reports for
the written ``.dom`` (no in-search override on disk), not this run's
in-search objective. ``max_share``/``conn_grade`` are still threaded through
so the evaluator's config otherwise matches the run (matters when
``leaf_sharing`` is on, e.g. a kpu/anneal grain that hasn't been unfolded
yet, or qi6's graded scalar in the reported grade).
"""
if result.best is None:
return result
fit = _fitness_for(str(programme_dir), leaf_sharing, superpose, multi_use=multi_use)
fit = _fitness_for(str(programme_dir), leaf_sharing, superpose, max_share,
conn_grade, collapse_insearch=False, multi_use=multi_use)
tree, base_fails, coll_fails, applied = fit.collapse_finish(
result.best.root, **collapse_kw
)
@ -1019,9 +1033,16 @@ def search_annealed(
created = operators.unfold_shared_leaves(best_root, above=1)
_log(f"[anneal] finish: de-share (grain off), rescore only — unfolded "
f"{created} leaf-{'copy' if created == 1 else 'copies'}")
# homemaker-py-sd3: forward collapse_insearch/multi_use from the phase
# kwargs (same family as the collapse_best bug) — omitting them left
# this rescore silently defaulting to _evaluate's collapse_insearch=True
# even on a --no-collapse-insearch run, contradicting the run's own
# objective on interrupt/no-polish exits.
ind, used = _evaluate(
best_root, programme_dir, None, x0=None, budget=seed_budget,
inner_kw={}, lineage="unfold", leaf_sharing=False, superpose=superpose)
inner_kw={}, lineage="unfold", leaf_sharing=False, superpose=superpose,
collapse_insearch=search_kw.get("collapse_insearch", True),
multi_use=search_kw.get("multi_use", False))
r = SearchResult(best=ind, population=[ind], n_evals=used, n_topologies=1)
r.n_distinct_signatures = 1
r.history = [(0, ind.fitness, ind.lineage)]

1
src/homemaker_layout/evolve.py
View file

@ -367,6 +367,7 @@ def main(argv=None) -> int:
r, programme_dir,
superpose=args.superpose,
multi_use=args.multi_use,
conn_grade=args.conn_grade,
local_search=args.collapse_local_search,
log=lambda m: print(m, file=sys.stderr, flush=True),
)

26
src/homemaker_layout/fitness.py
View file

@ -881,13 +881,29 @@ class Fitness:
Returns ``(tree, base_fails, collapsed_fails, applied)``. Both the input
and returned trees are UNMERGED — scoring is done on throwaway deepcopies
because ``score_with_fails`` merges the tree in place."""
because ``score_with_fails`` merges the tree in place.
homemaker-py-sd3: ``base_fails``/``cand_fails`` are measured with
``collapse_insearch`` forced off, regardless of how ``self`` was
configured. The guard's job is to protect the CANONICAL fail count of
the written ``.dom`` — what ``homemaker-fitness`` reports on disk with
no in-search override — not this run's in-search objective. Scoring
with ``collapse_insearch`` on made the guard vacuous: ``score_with_fails``
re-applies its own ``collapse_global`` pass before counting fails, so
``base_fails`` already reflected an auto-collapsed tree and came out
equal to ``cand_fails`` regardless of what this method's own explicit
collapse actually did."""
import copy
base_fails = len(self.score_with_fails(copy.deepcopy(root))[1])
cand = copy.deepcopy(root)
self.collapse_global(cand, **kw)
cand_fails = len(self.score_with_fails(copy.deepcopy(cand))[1])
saved_insearch = self._collapse_insearch
self._collapse_insearch = False
try:
base_fails = len(self.score_with_fails(copy.deepcopy(root))[1])
cand = copy.deepcopy(root)
self.collapse_global(cand, **kw)
cand_fails = len(self.score_with_fails(copy.deepcopy(cand))[1])
finally:
self._collapse_insearch = saved_insearch
if cand_fails <= base_fails:
return cand, base_fails, cand_fails, True
return copy.deepcopy(root), base_fails, cand_fails, False

30
tests/test_collapse_global.py
View file

@ -309,3 +309,33 @@ def test_collapse_finish_is_keep_better_and_unmerged():
assert coll_f <= base_f
assert applied == (coll_f < base_f) or coll_f == base_f
assert len(tree.leaves()) == 2 # unmerged: both room leaves intact
def test_collapse_finish_guard_is_canonical_even_with_insearch_collapse_on():
# homemaker-py-sd3 regression: score_with_fails auto-collapses BEFORE
# counting fails whenever the Fitness instance itself is configured with
# collapse_insearch=True (as driver.collapse_best used to build its
# evaluator). That silently made base_fails equal the ALREADY-collapsed
# count, so the 94g keep-better guard compared a collapsed tree against a
# collapsed tree and could never see collapse_global's true effect.
# collapse_finish must force canonical (collapse_insearch=False) scoring
# for its own base/collapsed measurement regardless of self's conf.
conf = _conf({
"b1": {"size": [16.0, 4.0], "width": [4.0, 1.0], "proportion": [1.5, 0.5]},
"b2": {"size": [12.0, 3.0], "width": [3.5, 0.8], "proportion": [1.5, 0.5]},
}, collapse_insearch=True)
fit = Fitness(conf=conf)
assert fit._collapse_insearch is True
root = _two_leaf_root("b1", "b1") # both b1 -> missing b2 is a real fail
tree, base_f, coll_f, applied = fit.collapse_finish(root)
# canonical base: the pre-collapse tree really is missing b2 -- if the
# guard were still vacuous, base_f would already equal coll_f (both
# silently pre-collapsed) instead of reporting the true starting fail.
assert base_f >= 1
assert coll_f < base_f
assert applied
assert sorted(lf.type for lf in tree.leaves()) == ["b1", "b2"]
# collapse_finish must not leak its temporary override back onto self.
assert fit._collapse_insearch is True