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

Re-verify 1ph: the iio bug could never have touched it

Browse source 
d86 asked for the rigorous version of 35's spot-check -- check out the
codebase at the 1ph commit, backport the iio stale-leaf-share fix, re-run
the historical seed sets.

One constraint had to be worked around: this repository's history begins
2026-07-30, six days AFTER the 1ph commit (2026-07-24), so that checkout
does not exist here. Closest reachable stand-in is 391f510 (2026-07-30), a
genuine ancestor of the iio fix and therefore pre-iio, carrying that era's
examples/ and objective. Measured there, not at the true 1ph commit, and
that is a real limitation of the reproduction.

Protocol as published: programme-house, budget 3000, 4 workers, seeds 1-20,
ON vs OFF, both arms finished with --collapse. Run twice over one worktree,
as-is then with the 22-line iio hunk applied.

                            OFF    ON     W/L/T    mean diff  t(df=19)
  published 1ph             7.95   7.10   11/6/3   +0.85      2.38
  pre-iio 391f510           8.05   7.10   11/6/3   +0.95      2.59
  same + iio fix            8.05   7.10   11/6/3   +0.95      2.59

The published verdict reproduces, and the iio fix changes nothing: 0 of 40
(seed, arm) cells differ, per-seed counts equal cell by cell.

It could not have been otherwise. The bug needs a leaf carrying a STALE
share/share_type, and leaf-sharing only stamps a share where a code has
count > 1 -- programme-house declares count: 1 for all six codes. Over 8
constructed seeds at that commit, programme-house had share_type set on 0
of 56 leaves; harbor-house on 24 of 128. _collapse_value reads
leaf.share_type, so the bug is structurally unreachable on the 1ph
protocol. That is why 35 saw 2/3 harbor seeds diverge by 5-8 fails while
programme-house at N=20 moves not one cell: harbor has codes at counts 10,
6 and 5.

20's retroactive caveat is discharged for the 1ph section and stays live
for harbor-house/qpk, where shares exist and divergence was measured.

Kept experiments/rerun_1ph_protocol.sh and both result TSVs.

Closes homemaker-py-d86.

Lint at parity (46); tests 384 passed, 0 failed.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MJ84Feep79Hhm3E4zZJmnB
This commit is contained in:
Claude 2026-08-29 13:21:29 +00:00
parent f6eeee7191
commit 50fb293fa5
No known key found for this signature in database
5 changed files with 167 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
.beads/issues.jsonl
View file

File diff suppressed because one or more lines are too long

60
DESIGN.md
View file

@ -2784,6 +2784,13 @@ is real per-seed noise (not a directional bias) that does not appear to overturn
verdict above, but the exact historical per-seed numbers quoted in this section were not re-measured verdict above, but the exact historical per-seed numbers quoted in this section were not re-measured
under the fix. See §35 for the mechanism and what was (and wasn't) re-confirmed. under the fix. See §35 for the mechanism and what was (and wasn't) re-confirmed.
**Caveat DISCHARGED for this section (`homemaker-py-d86`, §38.18).** The 1ph protocol has now been
re-run at N=20 on a pre-iio commit with and without the fix backported: all 40 (seed, arm) cells are
identical, and the reason is structural rather than lucky — programme-house declares `count: 1` for
every code, so no leaf ever acquires a share and the stale-share bug cannot fire on this programme
at all. The numbers above stand. The caveat remains live for harbor-house/qpk, where shares do exist
and §35 measured real divergence.
## 21. Insert/relocate-circulation repair operator (`homemaker-py-8sh`) — DONE (mixed, kept off) ## 21. Insert/relocate-circulation repair operator (`homemaker-py-8sh`) — DONE (mixed, kept off)
**Motivation.** qi6's remaining candidate (§18): mechanism (a), an explicit search-time **Motivation.** qi6's remaining candidate (§18): mechanism (a), an explicit search-time
@ -5778,6 +5785,59 @@ over 2/3/4 workers, asserting each is internally stable and deliberately not
asserting they agree with each other) and asserting they agree with each other) and
`test_scoring_a_frozen_design_is_deterministic`. `test_scoring_a_frozen_design_is_deterministic`.
### 38.18 The 1ph verdict re-verified: the iio bug could never have touched it (`homemaker-py-d86`)
`d86` asked for the rigorous version of §35's spot-check: take the codebase at the
`1ph` commit, backport the `iio` stale-leaf-share fix, and re-run the actual
historical seed sets to see whether the published verdict would have changed.
**One constraint had to be worked around.** This repository's history begins
**2026-07-30**, six days *after* the 1ph commit (2026-07-24) — the historical
checkout the issue asks for does not exist here. The closest reachable stand-in
is `391f510` (2026-07-30), which is a genuine ancestor of the iio fix
(`929be5b`, 2026-08-01) and therefore pre-iio, and which carries that era's
`examples/` and objective. Everything below is measured there, not at the true
1ph commit, and that is a real limitation of the reproduction.
**Protocol as published:** programme-house `init.dom`, budget 3000, 4 workers,
seeds 1–20, ON vs OFF, both arms finished with `--collapse`. Run twice over the
same worktree — once as-is, once with the 22-line `iio` `fitness.py` hunk applied.
| codebase | OFF | ON | W/L/T | mean diff | t (df=19) |
|---|---|---|---|---|---|
| published 1ph (2026-07-24) | 7.95 | 7.10 | 11/6/3 | +0.85 | 2.38 |
| pre-iio `391f510` | **8.05** | **7.10** | **11/6/3** | **+0.95** | **2.59** |
| the same, + iio fix backported | 8.05 | 7.10 | 11/6/3 | +0.95 | 2.59 |
Two results:
1. **The published verdict reproduces.** ON beats OFF, significant at N=20, with
a win/loss/tie split identical to the published 11/6/3.
2. **The iio fix changes nothing — 0 of 40 (seed, arm) cells differ.** Not a
coincidence of means: the per-seed fail counts are equal cell by cell.
**And it could not have been otherwise.** The iio bug needs a leaf carrying a
*stale* `share`/`share_type` — leftover multiplicity from a code the leaf has
since been retyped away from. Leaf-sharing only ever stamps a share when a code
has `count > 1`, and **programme-house declares `count: 1` for all six of its
codes**. Measured directly over 8 constructed seeds at that commit:
| programme | leaves | `share > 1` | `share_type` set |
|---|---|---|---|
| programme-house | 56 | **0** | **0** |
| harbor-house | 128 | 24 | 24 |
`_collapse_value` reads `leaf.share_type`; on programme-house it is never set,
so the bug is **structurally unreachable on the 1ph protocol**. That is why §35
saw 2 of 3 harbor seeds diverge by 5–8 fails while programme-house at N=20 moves
not one cell — harbor has codes at counts 10, 6 and 5.
§20's retroactive caveat is therefore **discharged for the 1ph section** and
stays live for harbor-house/qpk, where shares exist and the divergence was
measured. Worth noting for future archaeology: "was this measurement affected by
bug X" is often answerable from the programme's structure without re-running
anything.
## 39. Config audit: requirements that actively fight the engine (`homemaker-py-ju3`) — measured 2026-08-25 ## 39. Config audit: requirements that actively fight the engine (`homemaker-py-ju3`) — measured 2026-08-25
The corpus `patterns.config` targets and `costs.config` values were estimated The corpus `patterns.config` targets and `costs.config` values were estimated

26
experiments/rerun_1ph_protocol.sh Executable file
View file

@ -0,0 +1,26 @@
#!/usr/bin/env bash
# Re-run the 1ph protocol (DESIGN.md §20): programme-house init.dom, budget 3000,
# 4 workers, seeds 1-20, ON vs OFF, BOTH arms finished with --collapse.
set -u
W="$1"; TAG="$2"; OUT="$3"
cd "$W/examples/programme-house"
: > "$OUT"
for seed in $(seq 1 20); do
for arm in on off; do
flag=""; [ "$arm" = "off" ] && flag="--no-collapse-insearch"
t0=$(date +%s)
PYTHONPATH="$W/src" timeout 600 python -m homemaker_layout.evolve init.dom \
--budget 3000 --seed "$seed" --workers 4 --collapse $flag \
--output "$OUT.$arm.dom" > "$OUT.$arm.log" 2>&1
t1=$(date +%s)
n=$(PYTHONPATH="$W/src" python - "$OUT.$arm.dom" <<'PY'
import copy, sys
from homemaker_layout import dom, fitness
conf, cost = fitness.load_config(".")
_, f = fitness.Fitness(conf, cost).score_with_fails(copy.deepcopy(dom.load(sys.argv[1])))
print(len(f))
PY
)
echo -e "$TAG\t$seed\t$arm\t$n\t$((t1-t0))" >> "$OUT"
done
done

40
experiments/results/d86_1ph_iiofix.tsv Normal file
View file

@ -0,0 +1,40 @@
iiofix 1 on 5 8
iiofix 1 off 8 6
iiofix 2 on 7 8
iiofix 2 off 8 6
iiofix 3 on 9 8
iiofix 3 off 8 6
iiofix 4 on 9 7
iiofix 4 off 11 5
iiofix 5 on 9 8
iiofix 5 off 8 7
iiofix 6 on 8 7
iiofix 6 off 11 5
iiofix 7 on 9 8
iiofix 7 off 11 6
iiofix 8 on 8 8
iiofix 8 off 7 7
iiofix 9 on 6 7
iiofix 9 off 9 6
iiofix 10 on 4 8
iiofix 10 off 7 7
iiofix 11 on 5 7
iiofix 11 off 5 6
iiofix 12 on 7 8
iiofix 12 off 6 7
iiofix 13 on 6 7
iiofix 13 off 9 6
iiofix 14 on 8 8
iiofix 14 off 7 6
iiofix 15 on 6 8
iiofix 15 off 8 6
iiofix 16 on 8 7
iiofix 16 off 8 6
iiofix 17 on 6 7
iiofix 17 off 8 6
iiofix 18 on 7 7
iiofix 18 off 8 6
iiofix 19 on 8 7
iiofix 19 off 7 6
iiofix 20 on 7 7
iiofix 20 off 7 6
1 iiofix 1 on 5 8
2 iiofix 1 off 8 6
3 iiofix 2 on 7 8
4 iiofix 2 off 8 6
5 iiofix 3 on 9 8
6 iiofix 3 off 8 6
7 iiofix 4 on 9 7
8 iiofix 4 off 11 5
9 iiofix 5 on 9 8
10 iiofix 5 off 8 7
11 iiofix 6 on 8 7
12 iiofix 6 off 11 5
13 iiofix 7 on 9 8
14 iiofix 7 off 11 6
15 iiofix 8 on 8 8
16 iiofix 8 off 7 7
17 iiofix 9 on 6 7
18 iiofix 9 off 9 6
19 iiofix 10 on 4 8
20 iiofix 10 off 7 7
21 iiofix 11 on 5 7
22 iiofix 11 off 5 6
23 iiofix 12 on 7 8
24 iiofix 12 off 6 7
25 iiofix 13 on 6 7
26 iiofix 13 off 9 6
27 iiofix 14 on 8 8
28 iiofix 14 off 7 6
29 iiofix 15 on 6 8
30 iiofix 15 off 8 6
31 iiofix 16 on 8 7
32 iiofix 16 off 8 6
33 iiofix 17 on 6 7
34 iiofix 17 off 8 6
35 iiofix 18 on 7 7
36 iiofix 18 off 8 6
37 iiofix 19 on 8 7
38 iiofix 19 off 7 6
39 iiofix 20 on 7 7
40 iiofix 20 off 7 6

40
experiments/results/d86_1ph_preiio.tsv Normal file
View file

@ -0,0 +1,40 @@
preiio 1 on 5 8
preiio 1 off 8 6
preiio 2 on 7 7
preiio 2 off 8 6
preiio 3 on 9 7
preiio 3 off 8 7
preiio 4 on 9 8
preiio 4 off 11 6
preiio 5 on 9 8
preiio 5 off 8 7
preiio 6 on 8 8
preiio 6 off 11 6
preiio 7 on 9 7
preiio 7 off 11 6
preiio 8 on 8 7
preiio 8 off 7 6
preiio 9 on 6 8
preiio 9 off 9 5
preiio 10 on 4 7
preiio 10 off 7 7
preiio 11 on 5 7
preiio 11 off 5 6
preiio 12 on 7 8
preiio 12 off 6 7
preiio 13 on 6 8
preiio 13 off 9 6
preiio 14 on 8 8
preiio 14 off 7 6
preiio 15 on 6 8
preiio 15 off 8 6
preiio 16 on 8 7
preiio 16 off 8 5
preiio 17 on 6 7
preiio 17 off 8 6
preiio 18 on 7 7
preiio 18 off 8 5
preiio 19 on 8 7
preiio 19 off 7 5
preiio 20 on 7 7
preiio 20 off 7 6
1 preiio 1 on 5 8
2 preiio 1 off 8 6
3 preiio 2 on 7 7
4 preiio 2 off 8 6
5 preiio 3 on 9 7
6 preiio 3 off 8 7
7 preiio 4 on 9 8
8 preiio 4 off 11 6
9 preiio 5 on 9 8
10 preiio 5 off 8 7
11 preiio 6 on 8 8
12 preiio 6 off 11 6
13 preiio 7 on 9 7
14 preiio 7 off 11 6
15 preiio 8 on 8 7
16 preiio 8 off 7 6
17 preiio 9 on 6 8
18 preiio 9 off 9 5
19 preiio 10 on 4 7
20 preiio 10 off 7 7
21 preiio 11 on 5 7
22 preiio 11 off 5 6
23 preiio 12 on 7 8
24 preiio 12 off 6 7
25 preiio 13 on 6 8
26 preiio 13 off 9 6
27 preiio 14 on 8 8
28 preiio 14 off 7 6
29 preiio 15 on 6 8
30 preiio 15 off 8 6
31 preiio 16 on 8 7
32 preiio 16 off 8 5
33 preiio 17 on 6 7
34 preiio 17 off 8 6
35 preiio 18 on 7 7
36 preiio 18 off 8 5
37 preiio 19 on 8 7
38 preiio 19 off 7 5
39 preiio 20 on 7 7
40 preiio 20 off 7 6