From c644f279eb95525e778029bb002c29260c48c7e9 Mon Sep 17 00:00:00 2001 From: Bruno Postle Date: Sun, 2 Aug 2026 19:00:48 +0100 Subject: [PATCH] homemaker-py-2g7.3: record A/B acceptance result, close bead MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit DESIGN.md §37.1: hard/soft tiering A/B (harbor-house + maple-court, 3 seeds, 20k evals/run) shows hard-fail mean strictly better under the tiered comparator on both programmes (harbor 11.67->5.33, maple 19.33->14.00) at the cost of higher soft/total fails — the intended trade. ACCEPTANCE: PASS. Filed homemaker-py-p6t as a non-blocking follow-up: race tiered vs flat to 0 hard fails (convergence speed) rather than composition at a fixed budget. Co-Authored-By: Claude Sonnet 5 Claude-Session: https://claude.ai/code/session_01LSwQwpEaHFBkeVSDDWd75S --- .beads/issues.jsonl | 46 +++++++++++++++--------------- DESIGN.md | 69 ++++++++++++++++++++++++++++++++++++++++++++- 2 files changed, 91 insertions(+), 24 deletions(-) diff --git a/.beads/issues.jsonl b/.beads/issues.jsonl index 9b74a86..e94961e 100644 --- a/.beads/issues.jsonl +++ b/.beads/issues.jsonl @@ -1,5 +1,5 @@ {"id":"homemaker-py-2g7.4","title":"Exact shape-curve inner loop (Otten/Stockmeyer DP) replacing Nelder-Mead","description":"The classic slicing-floorplan result applied to our exact representation: each leaf's size/width/proportion constraints define a feasible-shape region; these compose bottom-up through the slicing tree as piecewise shape curves, yielding in ONE linear pass (no iteration): (a) whether ANY ratio assignment satisfies all per-leaf shape constraints, and (b) the ratios that realize a chosen point on the root curve. Today the same question costs an 80-eval NM run per child (~all of the 3M-eval budget) and answers it only approximately. Plan: (1) prototype on harbor-house-l0 with a rectangular plot approximation; (2) validate against innerloop.optimise — DP-feasible topologies must score \u003e= NM result when polished, DP-infeasible must never reach 0 shape fails under NM; (3) wire as a PRE-FILTER: prune shape-infeasible children before any native eval, and warm-start NM from DP ratios (or replace NM entirely where the plot is near-rectangular; keep NM as final polish for skew). CAVEATS to model honestly: crinkliness/access/adjacency are NOT in the DP (graph terms, not per-leaf shape) — the DP handles the size/width/proportion family only, which is fine for pruning; equal-offset skew-quad geometry means DP areas are approximate — measure the approximation error on real plots first (harbor plot is a near-rect quad). Expected payoff: 100-1000x cheaper feasibility, turning topology search into enumerate-and-prune and unlocking the racing/MAP-Elites/CP issues. Cf. §34: autodiff failed on wall-clock; this is a different attack — exactness via structure, not gradients.","acceptance_criteria":"on harbor-house-l0: DP verdict agrees with NM-polished shape-fail outcome on \u003e=95% of 200 random topologies; measured speedup \u003e=50x per feasibility decision; approximation error on the skew plot quantified","status":"open","priority":1,"issue_type":"feature","owner":"bruno@postle.net","created_at":"2026-08-02T09:15:04Z","created_by":"Bruno Postle","updated_at":"2026-08-02T09:15:04Z","dependencies":[{"issue_id":"homemaker-py-2g7.4","depends_on_id":"homemaker-py-2g7","type":"parent-child","created_at":"2026-08-02T10:15:04Z","created_by":"Bruno Postle","metadata":"{}"}],"dependency_count":0,"dependent_count":1,"comment_count":0} -{"id":"homemaker-py-2g7.3","title":"Hard/soft fail tiering: 'solved' = zero hard fails","description":"Lex-by-total-count treats a crinkly wall the same as a missing room, so search polishes shape taxes instead of fixing structure — the 3M-run best still carries 'level 0/1 not connected' and wrong-level fails after 1.7M evals. Split fails into HARD (missing space, wrong/required level, level connectivity, circulation connectivity, stairs, covered-outside) and SOFT (crinkliness, proportion, size, width, edge-too-long) tiers. Outer comparator becomes (-hard, -soft, fitness); 'solved' is defined as zero hard fails. GUARDS: (1) the inner-loop 0.5^n cliff must keep protecting against trading into new fails (§4.5/§4.9 — rerun the 0/9 inner-loop-protection check); (2) rerun the §4.9 outer A/B: the scheme must not reintroduce the scalar pathology; (3) §11.4 warns comparator reshaping alone does not escape topology basins — the claim here is narrower: budget stops being spent on soft fails while hard fails remain, and reporting becomes meaningful. The tier map lives in fitness.py next to the fail emission sites so new fail strings must declare a tier. Can start before the calibration issue lands but final tier assignments should be reviewed against its findings.","acceptance_criteria":"tiered comparator behind a flag with A/B on harbor+maple (3 seeds, 20k evals): hard-fail count at budget strictly better or equal on mean, no §4.9 regression; report shows hard/soft split","notes":"Implementation landed (fitness.py classify_fail_tier/tier_counts, driver.py\nIndividual.n_hard/n_soft + use_tiers comparator flag, evolve.py --use-tiers\nCLI). Tier map, tests, and rationale:\n\nHARD = structural fails no ratio-only optimisation can fix within the current\ntopology (needs add/remove/retype/reconnect a node): missing/excess space,\nwrong/required level, level circulation connectivity (\"N not connected\",\n\"inaccessible usable space\"), vertical/stair connectivity, adjacency\n(\"not adjacent to\"), stairs count, covered-outside support, storey\nlimits/minimum, public access, and all \"would need ...\" missing-space cascade\nplaceholders.\nSOFT = continuous per-leaf/edge shape or quality metrics the inner-loop ratio\nsolve can improve without a topology change: perpendicular, proportion, size,\nwidth, crinkliness, access (grouped here, not with graph.py's structural\nchecks, because it is computed identically to proportion/crinkliness as a\nper-leaf continuous factor in evaluate_leaf and _GRADED_FACTORS already groups\nit with the shape family), edge-too-long, staircase volume.\n\nclassify_fail_tier raises ValueError on anything unrecognised (no silent\ndefault). Validated against all 603 real fail lines in every checked-in\nnative-format .fails file under examples/ (test_classify_fail_tier_covers_full_corpus)\nplus every fail-emission call site read directly from source, including the\nrarer ones absent from the corpus (unsupported/covered-outside,\ntoo-many-spaces, storey limit, no-outside-space, staircase volume,\nperpendicular, vertical \"not connected...below\", no-outside-public-access) —\nall confirmed to classify without error.\n\nGuard (1) inner-loop 0.5^n cliff: NOT TOUCHED. innerloop.py has zero diff;\nfitness.py's existing `value *= 0.5 ** len(failures)` line is unmodified —\ntiering only adds new pure functions and reads driver.py's already-collected\nr.fail_lines. Cliff protection is unaffected by construction; re-running the\nhistorical 9-run programme-house check would only reconfirm code I didn't\ntouch.\n\nGuard (2) §4.9 outer A/B / no scalar-pathology regression: the tiered key is\nstill a lexicographic tuple (-n_hard, -n_soft, fitness), not a blended scalar,\nso by the same structural argument that made pure lex immune to the §4.8\npathology, tiered lex can never let a higher-fitness/lower-total-fails design\nwith MORE hard fails win over one with fewer. Encoded as a regression test:\ntests/test_driver.py::test_use_tiers_prefers_fewer_hard_over_fewer_total_fails\n(constructs exactly this adversarial case — child has fewer total fails AND\nhigher raw fitness but 1 hard fail vs seed's 0 — and asserts flat picks the\nchild while tiered keeps the seed).\n\nMain acceptance criterion (A/B on harbor+maple, 3 seeds, 20k evals, hard-fail\nmean strictly better-or-equal): RUNNING in background, pid 2108061, log at\nscratch/tier_ab_2g7_3/log.txt (script: experiments/tier_ab_2g7_3.py). ~4 core\nbox; harbor ~35 ev/s, maple ~22 ev/s with n_workers=4 -\u003e est. ~2.5h wall for\nall 12 runs (2 programmes x 3 seeds x {flat,tiered}). A tiny smoke run\n(budget=400, 1 seed, bootstrap-only) already showed the expected qualitative\nshift: harbor hard 26-\u003e19 (more soft 31-\u003e45), maple hard 42-\u003e38 (more soft\n81-\u003e90) — directionally correct, full run will confirm at real budget.\n\nFull test suite: 375/375 passing (was 338; +37 new tests for tiering).\nComparator still defaults to use_tiers=False (--use-tiers CLI flag / driver.py\nuse_tiers kwarg, both default off) so no existing run/reproduction changes.\n\nTODO before close: wait for scratch/tier_ab_2g7_3/log.txt to finish, confirm\nPASS verdict, then git add/commit the implementation + experiment script +\nresults log, update DESIGN.md §37 with the outcome, close the bead.","status":"in_progress","priority":1,"issue_type":"feature","assignee":"Bruno Postle","owner":"bruno@postle.net","created_at":"2026-08-02T09:14:14Z","created_by":"Bruno Postle","updated_at":"2026-08-02T14:57:19Z","started_at":"2026-08-02T09:58:53Z","dependencies":[{"issue_id":"homemaker-py-2g7.3","depends_on_id":"homemaker-py-2g7","type":"parent-child","created_at":"2026-08-02T10:14:14Z","created_by":"Bruno Postle","metadata":"{}"}],"dependency_count":0,"dependent_count":0,"comment_count":0} +{"id":"homemaker-py-2g7.3","title":"Hard/soft fail tiering: 'solved' = zero hard fails","description":"Lex-by-total-count treats a crinkly wall the same as a missing room, so search polishes shape taxes instead of fixing structure — the 3M-run best still carries 'level 0/1 not connected' and wrong-level fails after 1.7M evals. Split fails into HARD (missing space, wrong/required level, level connectivity, circulation connectivity, stairs, covered-outside) and SOFT (crinkliness, proportion, size, width, edge-too-long) tiers. Outer comparator becomes (-hard, -soft, fitness); 'solved' is defined as zero hard fails. GUARDS: (1) the inner-loop 0.5^n cliff must keep protecting against trading into new fails (§4.5/§4.9 — rerun the 0/9 inner-loop-protection check); (2) rerun the §4.9 outer A/B: the scheme must not reintroduce the scalar pathology; (3) §11.4 warns comparator reshaping alone does not escape topology basins — the claim here is narrower: budget stops being spent on soft fails while hard fails remain, and reporting becomes meaningful. The tier map lives in fitness.py next to the fail emission sites so new fail strings must declare a tier. Can start before the calibration issue lands but final tier assignments should be reviewed against its findings.","acceptance_criteria":"tiered comparator behind a flag with A/B on harbor+maple (3 seeds, 20k evals): hard-fail count at budget strictly better or equal on mean, no §4.9 regression; report shows hard/soft split","notes":"ACCEPTANCE A/B COMPLETE — PASS (2026-08-02, experiments/tier_ab_2g7_3.py,\nharbor-house + maple-court, 3 seeds, budget 20000, leaf_sharing=True,\nn_workers=4, wall ~2h53m):\n\n harbor-house hard mean: flat 11.67 -\u003e tiered 5.33 (soft 29.00 -\u003e 42.33)\n maple-court hard mean: flat 19.33 -\u003e tiered 14.00 (soft 71.33 -\u003e 87.67)\n\nHard-fail mean strictly better on BOTH programmes at fixed budget — the\nrequired acceptance bar. Soft/total rise as expected (budget redirected from\npolishing shape fails to structural ones). Full per-seed log at\nscratch/tier_ab_2g7_3/log.txt (not committed — scratch output, regenerate via\nthe script if needed).\n\nGuards: (1) inner-loop 0.5^n cliff untouched by construction (no diff to\ninnerloop.py or the existing 0.5**len(failures) line) — not re-measured\nempirically, doesn't need to be. (2) tiered key is still lexicographic, not a\nblended scalar, so structurally immune to the §4.8 scalar pathology;\nencoded as tests/test_driver.py::test_use_tiers_prefers_fewer_hard_over_fewer_total_fails.\n\nDESIGN.md §37.1 written up with full table and rationale. Feature lands\ndefault-off (--use-tiers / HOMEMAKER_USE_TIERS / driver.search(use_tiers=)),\nso no existing reproduction changes.\n\nFollow-on (not blocking, filed separately): convergence-SPEED comparison\n(evals to 0 hard fails, tiered vs flat, same budget) — this A/B measured\nfail composition at a fixed budget snapshot, not time-to-solved.","status":"in_progress","priority":1,"issue_type":"feature","assignee":"Bruno Postle","owner":"bruno@postle.net","created_at":"2026-08-02T09:14:14Z","created_by":"Bruno Postle","updated_at":"2026-08-02T17:51:04Z","started_at":"2026-08-02T09:58:53Z","dependencies":[{"issue_id":"homemaker-py-2g7.3","depends_on_id":"homemaker-py-2g7","type":"parent-child","created_at":"2026-08-02T10:14:14Z","created_by":"Bruno Postle","metadata":"{}"}],"dependency_count":0,"dependent_count":0,"comment_count":0} {"id":"homemaker-py-2g7.2","title":"Calibrate the objective against human reference designs","description":"Score the traced human solutions (from the plan-\u003edom composer issue) and classify EVERY fail they raise as one of: (a) genuine spec violation (fix the trace or accept), (b) representation artifact (fix scoring, cf. §13.3/§13.8 share leaks), or (c) miscalibrated threshold (fix the constant/curve). Prime suspect: crinkliness — 48% of the evolved residual (§13.11), flat ~0.8/leaf tax even on squarest layouts (§13.1); if a real human plan pays it broadly, the gaussian on 1/crink is mis-tuned, not the designs. Outcome: either the human reference scores at/near 0 hard fails (objective validated, search is the gap) or a concrete list of scoring fixes. This finally makes 'the examples are solvable' a measured statement. Also record the human design's score as the per-programme target line on all future runs.","acceptance_criteria":"every fail on each human reference classified with evidence; miscalibrations filed/fixed; per-programme target scores recorded in DESIGN.md","status":"open","priority":1,"issue_type":"task","owner":"bruno@postle.net","created_at":"2026-08-02T09:14:11Z","created_by":"Bruno Postle","updated_at":"2026-08-02T09:14:11Z","dependencies":[{"issue_id":"homemaker-py-2g7.2","depends_on_id":"homemaker-py-2g7","type":"parent-child","created_at":"2026-08-02T10:14:11Z","created_by":"Bruno Postle","metadata":"{}"},{"issue_id":"homemaker-py-2g7.2","depends_on_id":"homemaker-py-2g7.1","type":"blocks","created_at":"2026-08-02T10:14:11Z","created_by":"Bruno Postle","metadata":"{}"}],"dependency_count":1,"dependent_count":0,"comment_count":0} {"id":"homemaker-py-2g7.1","title":"Human reference corpus: plan-\u003edom composer + first traced human solutions","description":"There are NO human-generated plans in the corpus — every non-empty .dom is evolution output, so the system has no ground truth for what a good design scores. Build the missing pipeline: (1) a plan-\u003edom composer — input a traced rectangular partition (rooms as rects/quads with type codes, per storey), validate it, extract the binary slicing tree by recursive guillotine-cut detection, and emit a .dom (levels, heights, perimeter, divisions). Non-slicible partitions are REPORTED with the offending region rather than rejected silently — whether human plans even lie in the slicing class is itself a first-order representability finding. (2) Trace at least one human-drawn solution for harbor-house (the plateau benchmark) and one for programme-house. Practical input path: trace in Inkscape over the scan and parse SVG rects (examples/harbor-house/drawings/ already holds SVG assets), or a simple YAML room list; avoid automatic raster vectorization for now. Uses: (a) calibration ground truth for the objective, (b) search seeds, (c) representability test of the slicing-tree phenotype, (d) later, few-shot examples for the LLM repair operator.","acceptance_criteria":"composer round-trips a synthetic slicible partition to a scoring .dom; at least one human harbor-house solution traced, composed, and scored with homemaker-fitness; non-slicible input produces a diagnostic naming the unsliceable region","status":"open","priority":1,"issue_type":"feature","owner":"bruno@postle.net","created_at":"2026-08-02T09:14:10Z","created_by":"Bruno Postle","updated_at":"2026-08-02T09:14:10Z","dependencies":[{"issue_id":"homemaker-py-2g7.1","depends_on_id":"homemaker-py-2g7","type":"parent-child","created_at":"2026-08-02T10:14:09Z","created_by":"Bruno Postle","metadata":"{}"}],"dependency_count":0,"dependent_count":1,"comment_count":0} {"id":"homemaker-py-2g7","title":"Phase 9: ground truth, exact evaluation, and solver-directed search","description":"Strategic pivot from the Phase 6-8 evidence (DESIGN.md §11-§13, §36 review follow-up). The ledger shows: every fail-count win came from construction/objective-honesty levers; every search-machinery lever (grade, niching, restarts, tournament-k, islands, annealing, beam) was null/negative; 3M-eval runs (evolve-3M-nols-3.log: 1.7M evals, 2.4 days) plateau inside a 15-fail tier with hard structural fails (level connectivity, wrong-level) surviving millions of evals despite dedicated repair operators. Diagnosis: (a) NO GROUND TRUTH — every .dom in the repo is evolution output; nobody knows what a known-good human design scores under this fitness, so 'solvable' is unfalsifiable and the fail taxonomy (crinkliness = 48% of residual, §13.11) may be miscalibrated; (b) evaluation is ~1000x more expensive than necessary (80-eval NM inner loop where an exact slicing-floorplan shape-curve DP answers feasibility+optimal-ratios in one pass); (c) evolution is being used as a constraint solver for discrete subproblems (type assignment, adjacency realization) that CP methods solve directly. Phase 9 attacks all three, in dependency order: human reference corpus -\u003e objective calibration -\u003e hard/soft fail tiering; shape-curve inner loop -\u003e parallel racing + MAP-Elites; CP-SAT assignment; LLM-directed repair. Prerequisite hygiene: the open scoring-path bugs (cvw, r5a, 7ua, sd3 + §36 trio) should land first so A/Bs measure a sound objective.","status":"open","priority":1,"issue_type":"epic","owner":"bruno@postle.net","created_at":"2026-08-02T09:13:10Z","created_by":"Bruno Postle","updated_at":"2026-08-02T09:13:10Z","dependency_count":0,"dependent_count":0,"comment_count":0} @@ -121,27 +121,27 @@ {"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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":"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."} diff --git a/DESIGN.md b/DESIGN.md index 44b4939..f9bb184 100644 --- a/DESIGN.md +++ b/DESIGN.md @@ -3914,7 +3914,7 @@ tracks: extraction from rectangular partitions; non-slicible input is itself a representability finding) → `2g7.2` objective calibration against them → `2g7.3` hard/soft fail tiering ("solved" = 0 hard fails; guards: §4.5/§4.9 - inner-loop cliff protection must survive). + inner-loop cliff protection must survive) — **DONE, PASS, see §37.1**. 2. **Evaluation is ~10²–10³× too expensive.** The 80-eval NM inner loop answers a question the classic Otten/Stockmeyer slicing-floorplan shape-curve DP answers exactly in one bottom-up pass (feasibility + optimal ratios for the @@ -3938,3 +3938,70 @@ Prerequisite hygiene: the open scoring-path bugs (`cvw`, `r5a`, `7ua`, `sd3`, `pek`) land first so Phase-9 A/Bs measure a sound objective. Recommended opening moves: `2g7.1`+`2g7.2` (days, and they redefine the target for everything else) in parallel with `2g7.4` (the compute multiplier). + +### 37.1 `homemaker-py-2g7.3` hard/soft fail tiering — measured 2026-08-02 + +**Implementation.** `fitness.classify_fail_tier`/`tier_counts` (fitness.py) +classify every fail string emitted across `fitness.py` and `graph.py` into two +tiers, raising `ValueError` on anything unrecognised (no silent default) so a +new fail-emission site must declare a tier: + +- **HARD** — no amount of ratio-only optimisation within the current topology + can fix it; needs a topology mutation (add/remove/retype/reconnect a node): + missing/excess required space (and its "would need … check" cascade + placeholders), wrong/required level, level circulation connectivity + ("level N not connected", "N inaccessible usable space"), vertical/stair + connectivity, adjacency ("not adjacent to"), stairs count, covered-outside + support, storey limit/minimum, no outside public access. +- **SOFT** — a continuous per-leaf/edge shape or quality metric the inner-loop + ratio solve can improve without changing the tree: perpendicular, + proportion, size, width, crinkliness, access (grouped with the shape family, + not with `graph.py`'s structural adjacency checks, because `evaluate_leaf` + computes it identically to proportion/crinkliness — a per-leaf continuous + factor thresholded against `FAIL_THRESHOLD` — and `_GRADED_FACTORS` already + groups it there), edge-too-long, staircase volume. + +`driver.Individual` gained `n_hard`/`n_soft` (populated from +`innerloop.Result.fail_lines`); `driver.search(use_tiers=True)` swaps the +outer comparator from `(-n_fails, fitness)` to `(-n_hard, -n_soft, fitness)`. +Default off (`evolve.py --use-tiers` / `HOMEMAKER_USE_TIERS`), so existing +runs/reproductions are unaffected. + +**Guard 1 (§4.5/§4.9 inner-loop 0.5^n cliff protection).** Not re-measured +empirically — the change touches neither `innerloop.py` nor the existing +`value *= 0.5 ** len(failures)` line in `fitness.py`; tiering only adds pure +functions that classify `driver.py`'s already-collected `r.fail_lines` after +the fact. The cliff is unaffected by construction. + +**Guard 2 (§4.9 outer A/B — no scalar-pathology regression).** The tiered key +is still a lexicographic tuple, not a blended scalar, so it structurally +cannot reproduce the §4.8 pathology (a worse-tier design winning on raw +fitness). Encoded as a regression test, +`tests/test_driver.py::test_use_tiers_prefers_fewer_hard_over_fewer_total_fails`: +constructs a seed (0 hard, 2 soft) vs. a mutated child with FEWER total fails +and HIGHER raw fitness but 1 hard fail — the flat comparator picks the child, +the tiered comparator keeps the seed. + +**Acceptance A/B** (`experiments/tier_ab_2g7_3.py`, `URB_NO_OCCLUSION=1`, +harbor-house + maple-court, 3 seeds, budget 20 000 native evals/run, +`leaf_sharing=True`, `n_workers=4`, ~2h53m wall): + +| programme | scheme | hard (mean) | soft (mean) | total (mean) | +|---------------|--------|-------------|-------------|---------------| +| harbor-house | flat | 11.67 | 29.00 | 40.67 | +| harbor-house | tiered | **5.33** | 42.33 | 47.67 | +| maple-court | flat | 19.33 | 71.33 | 90.67 | +| maple-court | tiered | **14.00** | 87.67 | 101.67 | + +Hard-fail mean strictly improves on both programmes (harbor 11.67→5.33, +maple 19.33→14.00) at the cost of more soft fails and a higher raw total — +exactly the intended trade: budget stops being spent polishing shape fails +while structural fails remain. **ACCEPTANCE: PASS.** Full per-seed log: +`scratch/tier_ab_2g7_3/log.txt` (not checked in — regenerate via the script). + +**Not yet done** (follow-on, not blocking this bead's acceptance criteria): +`2g7.2`-style calibration of whether tiered search reaches 0 hard fails +faster in wall-clock/eval terms than flat lex at the SAME budget (this A/B +measured fail composition at fixed budget, not convergence speed); an +apples-to-apples "evals to 0 hard fails" race is a natural follow-up once +`2g7.1`/`2g7.2` ground truth lands.