Run 2026-09-14_x7_rejudge_votes5

Preregistered decision rule, v0.3 (PREREGISTRATION-v0.3.md section 5)

• Validity precondition (B4-strict answered on 12/24 filler notes, rate 50.0% [31, 69], must be below 10%): FAIL
• H1 (planted recall 10/22 vs decoy false positives 1/12, Fisher one-sided): p = 0.0296 -> PASS
• H2a (B7 planted card pairs 3 in 300 draws, expected 1.396, hypergeometric): p = 0.1651; arm-label permutation vs B1: gap = +0.0067, p = 0.3203 -> FAIL
• H2b (non-NONE rate B7 36/300 vs B1 37/300, Fisher one-sided): p = 0.5985 -> FAIL
• H3-strict (S0 two-note recall 10/22 vs B4-strict one-note recall 2/22 on bridge notes, Fisher one-sided): p = 0.0078 -> PASS (not interpretable: the validity precondition failed)
• H4 (partner-domain filler, estimation only, not a recombination test): S0 recall 10/22 = 45.5% [27, 65] vs S1 recall 0/22 = 0.0% [0, 15]
• H5 (recall by writer family, estimation only, no pass/fail):
• A: 8/10 = 80.0% [49, 94]
• B: 2/12 = 16.7% [5, 45]

Decision: NULL (signal requires the validity precondition, H1 and H3-strict all passing; H2, H4, H5 are reported and do not enter the rule).

Run summary

Total measured cost: $88.273001

Models: - cards: claude-haiku-4-5-20251001 (alias haiku) - critic: claude-haiku-4-5-20251001 (alias haiku) - critic_haiku: claude-haiku-4-5-20251001 (alias haiku) - critic_sonnet: claude-sonnet-5 (alias sonnet) - generator: claude-sonnet-5 (alias sonnet) - match_gold: claude-haiku-4-5-20251001 (alias haiku)

Exploratory finds (non-planted survivors, manual review only): 47

Oracle recall on planted: 45.5% [27, 65]; permutation p = 0.0001

T1. Corpus

notes	cards	cards_per_note	bridges	decoys	leakage_failures	corpus_sha256
96	521	5.43	22	12	0	d044765e9dff

T2. Sampler enrichment; base rate 0.47% of 134289 cross-note card pairs are planted

arm	draws	planted_hits	distinct_bridges	expected_hits	p_hypergeom
B1	300	1	1	1.396	0.754
B7	300	3	3	1.396	0.165

T3. Oracle set: generator and critic without the sampler; mean cosine to gold on planted = 0.566

S0 group	units	NONE rate	non-NONE rate	survivor rate (post critic + dupgate)
planted	22	22.7% [10, 43]	77.3% [57, 90]	72.7% [52, 87]
decoy	12	75.0% [47, 91]	25.0% [9, 53]	8.3% [1, 35]
random	36	94.4% [82, 98]	5.6% [2, 18]	2.8% [0, 14]
recall on planted (gold match)	22			45.5% [27, 65]

T4. Per arm: NONE rate, critic kill rate, survivors, recovered planted bridges, measured cost

arm	units	NONE	critic kill	dup	survivors	bridges reachable	bridges recovered	recall	cost usd	usd per recovered
B1	300	87.7% [83, 91]	48.6% [33, 64]	0	19	1	0	0.0% [0, 79]	27.0952	n/a
B4	68	63.2% [51, 74]	12.0% [4, 30]	0	22	22	2	9.1% [3, 28]	7.2638	3.6319
B7	300	88.0% [84, 91]	36.1% [22, 52]	0	23	3	0	0.0% [0, 56]	27.0758	n/a
S0	70	68.6% [57, 78]	18.2% [7, 39]	0	18	22	10	45.5% [27, 65]	6.998	0.6998
S1	44	88.6% [76, 95]	20.0% [4, 62]	0	4	22	0	0.0% [0, 15]	3.9967	n/a

T5. Label-permutation null over the S0 oracle set

statistic	observed	expected under null	p	shuffles
survivors among planted S0 units	16	5.684	0.0001	10000

T6. Recombination vs single-note reflection on the same bridge notes

arm	bridges reachable	recovered	recall	fisher p (S0 > B4)
S0 two notes	22	10	45.5% [27, 65]	0.00785
B4 one note	22	2	9.1% [3, 28]

T7. Exploratory control (not preregistered): does the gold mechanism appear when the partner note is replaced by a mechanism-free filler from the same domain?

arm	bridges reachable	recovered	recall	fisher p (S0 > S1)
S0 bridge note + true partner	22	10	45.5% [27, 65]	0.000261
S1 bridge note + partner-domain filler	22	0	0.0% [0, 15]
B4 bridge note alone	22	2	9.1% [3, 28]

T8. Recall after critic, per critic (same generations, before the duplicate gate)

critic	model id	correct recoveries	surviving	killed	decoy answers	decoy surviving	kill rate on random
haiku	claude-haiku-4-5-20251001	10	10	0	3	1	50.0% [9, 91]
sonnet	claude-sonnet-5	10	9	1	3	1	50.0% [9, 91]

T9. Filler abstention under the strict single-note prompt: the single-note arm split by note kind. The filler row is the v0.3 validity precondition; bridge and decoy rows for comparison

note kind	units	answered	NONE	errors	answer rate
bridge	44	13	31	0	29.5% [18, 44]
filler	24	12	12	0	50.0% [31, 69]