bruno/homemaker-layout
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
homemaker-layout/experiments
History
Claude 225e673ae3
Staged harness re-scored under a different objective than it searched 
run_staged_search.py reported MISMATCH on its BASELINE arm -- the
LEAFSHARE=0/MULTIUSE=0 control every A/B compares against. Two facts
combined: driver.search_staged had no collapse_insearch parameter at all,
so every inner search() call inherited search()'s True default
unconditionally; and no example patterns.config sets the key, so the final
_native_score rescore got False from a bare load_config. Search optimised
one objective, the rescore graded another.

The 7ua fix pinned the key inside a fitness.load_config monkeypatch, but
that patch was installed only `if leaf_share or multi_use` -- so it fixed
every arm except the control.

Fixed in the right place: search_staged now HAS the parameter (default
True, byte-identical to the inherited default), threaded into all three
internal search() calls. The harness chooses the arm explicitly (COLLAPSE,
default 1), passes it to the search, and passes the SAME value to
_native_score, which overrides the key rather than hoping the config
carries it. The rescore mirrors the search by construction.

Verified on programme-house, budget 150:
  baseline    MISMATCH 1.56663e-08 vs 1.51708e-08  ->  OK
  COLLAPSE=0  (knob did not exist)                 ->  OK 1.66216e-08
  LEAFSHARE=1 / MULTIUSE=1                         ->  OK

COLLAPSE=0 scoring differently confirms the knob is not a no-op, and the
default arm's search result is unchanged, so no prior staged number moves.

Audited the other three search_staged callers: run_and_capture_91f.py
already pins collapse_insearch: True; run_island_ab.py never re-scores;
probe_harbor_floor.py did NOT pin it and had the same bug -- now fixed, and
that is the harness which produced every 13.x floor number.

The recorded mitigating factor -- only the continuous score moved, the fail
count matched, and the run_*_ab.sh greps read only the count -- is true and
is exactly what made it dangerous: a harness that reports MISMATCH on its
own control, invisibly to the metric of record, trains everyone to ignore
the warning.

Closes homemaker-py-4ok.

Lint at parity (46); tests 381 passed, 0 failed.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MJ84Feep79Hhm3E4zZJmnB
2026-08-29 12:11:38 +00:00
..
results Checkpoint long searches; the cold-start runs were lost to a reclaimed box 2026-08-29 05:43:58 +00:00
ab_cpsat_assign.py homemaker-py-2g7.5: CP-SAT exact room-code assignment (seeder + reassign op) 2026-08-04 09:19:36 +01:00
ab_crinkliness_mode_ssz.py ssz: crinkliness is declared per space; there is no daylight attribute 2026-08-28 22:39:09 +00:00
ab_shapecurve_prune.py homemaker-py-wkh: DP-exact hard pre-filter for driver.py's shape-feasibility prune 2026-08-03 21:10:28 +01:00
ab_shapecurve_warmstart.py homemaker-py-6xh: wire shape-curve DP into driver.py as an NM warm-start 2026-08-03 18:43:28 +01:00
ab_ssz_search.py ssz: crinkliness is declared per space; there is no daylight attribute 2026-08-28 22:39:09 +00:00
accept_innerloop.py Rename package: homemaker → homemaker-layout 2026-06-14 08:18:06 +01:00
audit_crinkliness_truth.py Owner's daylight ruling: corridors need it; retract the 61% headline 2026-08-28 23:14:21 +00:00
audit_programme_config.py §39.4 completion + §39.5 retraction + §39.6: the usage namespace is NOT clean 2026-08-26 10:09:14 +00:00
autodiff_spike.py Spike: torch-autodiff inner-loop ratio optimisation (homemaker-py-2ax) — negative 2026-08-01 10:26:43 +01:00
bakeoff_harbor.json Swap inner loop default from CMA-ES to Nelder-Mead (homemaker-py-d6d) 2026-06-14 08:51:22 +01:00
bakeoff_harbor.py Swap inner loop default from CMA-ES to Nelder-Mead (homemaker-py-d6d) 2026-06-14 08:51:22 +01:00
bakeoff_innerloop.json Bake-off: CMA-ES confirmed as inner-loop optimiser (homemaker-py-d0s) 2026-06-13 09:47:15 +01:00
bakeoff_innerloop.py Rename package: homemaker → homemaker-layout 2026-06-14 08:18:06 +01:00
bakeoff_native.json Swap inner loop default from CMA-ES to Nelder-Mead (homemaker-py-d6d) 2026-06-14 08:51:22 +01:00
bakeoff_native.py Swap inner loop default from CMA-ES to Nelder-Mead (homemaker-py-d6d) 2026-06-14 08:51:22 +01:00
bench_batch_oracle.py Rename package: homemaker → homemaker-layout 2026-06-14 08:18:06 +01:00
benchmark_vs_urbevolve.py Rename package: homemaker → homemaker-layout 2026-06-14 08:18:06 +01:00
diag_2f45907.py Rename package: homemaker → homemaker-layout 2026-06-14 08:18:06 +01:00
diag_connectivity_yql.py §39.9: level-not-connected is destroyed by the resize, not by the search 2026-08-26 14:39:07 +00:00
diag_depth_balance.py x3b/§13.10: productionise leaf-sharing — per-code share grain + CLI wiring 2026-06-28 22:04:35 +01:00
diag_edge_too_long.py x3b/§13.10: productionise leaf-sharing — per-code share grain + CLI wiring 2026-06-28 22:04:35 +01:00
diag_exposure_frontage.py Frontage feasibility as a pre-flight check; retract the "3x short" claim 2026-08-26 16:20:52 +00:00
diag_leaf_shapefail.py erc.1: Diagnostic A — per-leaf shape-fail vs density (§13.1) 2026-06-23 22:06:04 +01:00
diag_leaf_sharing.py x3b/§13.10: productionise leaf-sharing — per-code share grain + CLI wiring 2026-06-28 22:04:35 +01:00
diag_residual_91f.py homemaker-py-91f: residual diagnostic on the current full default stack 2026-08-01 19:38:46 +01:00
diag_slack_localization.py erc.2: Diagnostic B — undersize-despite-slack localization (§13.2) 2026-06-23 22:47:34 +01:00
dump_areas.pl Scaffold homemaker-py with validated geometry port 2026-06-10 20:50:20 +01:00
dump_areas.py Rename package: homemaker → homemaker-layout 2026-06-14 08:18:06 +01:00
dump_leaf_quality.pl Native fitness: leaf quality terms + cost model (homemaker-py-gnw) 2026-06-13 07:59:21 +01:00
genome_parity.py Rename package: homemaker → homemaker-layout 2026-06-14 08:18:06 +01:00
leaf_parity.py Rename package: homemaker → homemaker-layout 2026-06-14 08:18:06 +01:00
measure_seed_geometry.py Phase 7 §12.2: proportion-aware constructive seeding + storey_minimum fix (leu.2, cq1) 2026-06-20 14:04:42 +01:00
migrate_crinkliness_key.py ssz: crinkliness is declared per space; there is no daylight attribute 2026-08-28 22:39:09 +00:00
migrate_usage_key.py §39.7: access requirements become a declared usage: attribute (homemaker-py-sel) 2026-08-26 13:39:41 +00:00
operator_locality.py Rename package: homemaker → homemaker-layout 2026-06-14 08:18:06 +01:00
optimize_fullfitness.py Rename package: homemaker → homemaker-layout 2026-06-14 08:18:06 +01:00
penalty_reshape.json Lexicographic outer-search comparison, preserve inner-loop cliff (homemaker-py-yg5) 2026-06-14 09:20:03 +01:00
penalty_reshape.py Lexicographic outer-search comparison, preserve inner-loop cliff (homemaker-py-yg5) 2026-06-14 09:20:03 +01:00
probe_harbor_floor.py Staged harness re-scored under a different objective than it searched 2026-08-29 12:11:38 +00:00
rebaseline_no_occlusion.py Rename package: homemaker → homemaker-layout 2026-06-14 08:18:06 +01:00
refine_sweep.py Rename package: homemaker → homemaker-layout 2026-06-14 08:18:06 +01:00
resolve_ratios.py Rename package: homemaker → homemaker-layout 2026-06-14 08:18:06 +01:00
run_1s3_confirm.sh homemaker-py-1s3: multi-use leaves N=15 confirmation -- does not replicate 2026-08-01 08:52:24 +01:00
run_1s3_confirm_staged.sh homemaker-py-1s3: multi-use leaves N=15 confirmation -- does not replicate 2026-08-01 08:52:24 +01:00
run_6zy_ab.sh 6zy/§11.8: co-tune topology diversity × tournament pressure — null robust 2026-06-29 22:56:23 +01:00
run_8sh_ab.sh 8sh: insert/relocate-circulation repair operator (qi6 mechanism (a)) 2026-07-24 19:48:13 +01:00
run_9gp_ab.sh Phase 7 §12.3: 9gp A/B measured — NEGATIVE; close 9gp + epic leu 2026-06-21 07:21:51 +01:00
run_9yx_sweep.sh homemaker-py-9yx: health-centre — non-synthetic third example programme 2026-07-30 00:09:37 +01:00
run_and_capture_91f.py homemaker-py-91f: residual diagnostic on the current full default stack 2026-08-01 19:38:46 +01:00
run_c3g_ab.sh c3g: circ-per-room granularity knob (circ_divisor) + A/B harness 2026-06-21 21:10:18 +01:00
run_coldstart_baseline.py coldstart programme-house seed 0 @ 1000: 7 fails (3h/4s) 2026-08-28 23:29:01 +00:00
run_depthbal_ab.sh erc.4: depth-balanced construction mechanism + floor probe (§13.4) 2026-06-25 22:36:24 +01:00
run_e01_sweep.py e01: harbor-house beam-width larger-N confirmation (null) 2026-07-29 00:22:55 +01:00
run_f1d_ab.sh f1d: ruin-and-recreate LNS operator, validated positive on programme-house 2026-07-26 09:31:42 +01:00
run_f1d_larger_n.sh f1d: ruin-and-recreate LNS operator, validated positive on programme-house 2026-07-26 09:31:42 +01:00
run_f1d_w3_ab.sh f1d: ruin-and-recreate LNS operator, validated positive on programme-house 2026-07-26 09:31:42 +01:00
run_interioro_ab.sh erc/ld2: interior-O light-well seeding — §13.6 positive on dense floors 2026-06-28 07:20:20 +01:00
run_island_ab.py psk/§14: island model — null (best-of-N at equal budget wins) 2026-06-29 06:20:29 +01:00
run_leafshare_ab.sh erc.3: explicit per-leaf multiplicity closes the leak; driver A/B wired (§13.3) 2026-06-24 18:16:17 +01:00
run_leu2_ab.sh Phase 7 §12.2: proportion-aware constructive seeding + storey_minimum fix (leu.2, cq1) 2026-06-20 14:04:42 +01:00
run_lj3_qjg_ab.sh lj3/qjg: larger-N + weight sweep finds bridge_circulation effect is noise 2026-07-25 12:20:22 +01:00
run_multiuse_ab.sh homemaker-py-1s3: multi-use leaves as permanent design goal (§26 path b) 2026-07-31 00:16:12 +01:00
run_qi6_ab.sh qi6: full-budget A/B for graded circulation-connectivity signal, negative result 2026-07-23 18:29:45 +01:00
run_search.py Rename package: homemaker → homemaker-layout 2026-06-14 08:18:06 +01:00
run_search_scaled.py 6zy/§11.8: co-tune topology diversity × tournament pressure — null robust 2026-06-29 22:56:23 +01:00
run_shareedge_ab.sh hph/§13.8: share-aware edge-too-long cap — shared leaves no longer penalised for aggregate wall length 2026-06-28 21:24:51 +01:00
run_sharefactor_sweep.sh erc.7: leaf-sharing × depth-balancing synergy CONFIRMED end-to-end (§13.5) 2026-06-26 16:47:14 +01:00
run_staged_search.py Staged harness re-scored under a different objective than it searched 2026-08-29 12:11:38 +00:00
run_synergy_ab.sh erc.7: leaf-sharing × depth-balancing synergy CONFIRMED end-to-end (§13.5) 2026-06-26 16:47:14 +01:00
run_xyu_sweep.sh xyu: n=18 ruin_recreate N=15 confirmation — weakened, not resolved 2026-07-29 10:10:23 +01:00
run_y51_sweep.sh y51: synthetic room-count sweep finds no clean ruin_recreate size threshold 2026-07-26 15:35:03 +01:00
run_y51_sweep_ln.sh y51: synthetic room-count sweep finds no clean ruin_recreate size threshold 2026-07-26 15:35:03 +01:00
shapecurve_spike.py homemaker-py-2g7.4: fix shape-curve DP to be rotation-invariant 2026-08-03 07:06:48 +01:00
sweep_failtypes.py Rename package: homemaker → homemaker-layout 2026-06-14 08:18:06 +01:00
tier_ab_2g7_3.py homemaker-py-2g7.3: hard/soft fail tiering behind --use-tiers flag 2026-08-02 16:00:39 +01:00
usage_map_proposal.yaml homemaker-py-sel: usage is a plain space attribute — drop the mapping table 2026-08-26 12:56:12 +00:00
validate_shapecurve.py homemaker-py-wkh: DP-exact hard pre-filter for driver.py's shape-feasibility prune 2026-08-03 21:10:28 +01:00
validate_shapecurve_multistorey.py homemaker-py-koo: multi-storey (below-link) support for the shape-curve DP 2026-08-03 23:30:18 +01:00
warm_vs_cold.py Rename package: homemaker → homemaker-layout 2026-06-14 08:18:06 +01:00