metrics

Exp 0 - metrics report

Layer 2　·　PCFG toy 4L d_model=448 / pcfg　·　matryoshka_hook_resid_post_L2　·　1,792 latents in 8 blocks　·　1,016,600 tokens over 3400 docs　·　edge: reverse coverage ≥ 0.5, both endpoints fire ≥ 20

Block pair 0->1 - 111 candidate edges

• Out-degree: 24 parents, 74 children, 19 multi-parented (PolyFrac 25.7%); top-1 parent holds 32.4% of edges, Gini 0.966, max out-degree 36.
• Superparents (out-degree flag): 0 (0 also pass the old fire-rate AND-gate)
• Independence null: mean edge PMI 1.22; 0 edges (0.0%) at chance level (PMI < 0.5). 42 edges dropped by the joint-support guard (n_joint < 30).
• Recon-ablation contribution filter (Tree-SAE-inspired baseline): 102/111 edges pass (91.9%).
• Frequency control: mean survival 1.024 over 111 testable edges; 0 (0.0%) are frequency-driven (survival < 0.5).
• Sibling redundancy (global Jaccard — confounded proxy, not the splitting verdict; the parent-conditioned version is in the stage-03 second pass): mean 0.132 over 9 parents; 0 over the 0.5 global threshold.
• Joint-child (exact union, parents with edges): R_supp mean 0.785, R_mass mean 0.794; 0 parents with one child holding >=90% of their energy (rename candidates).
• Joint-child coverage (min(1, ΣF) upper bound — saturates when children co-fire, kept only for contrast with the exact union): 0.180.

parent -> child	R	F	PMI	recon P/C gain	recon?	surv	sib	parent label	child label
219 -> 379	1.00	0.00	1.39	0.57/0.00	n	1.00	0.02	feature 219	feature 379
8 -> 395	1.00	0.11	4.22	9.06/0.73	Y	1.00	0.22	feature 8	feature 395
98 -> 395	1.00	0.01	1.62	6.81/0.73	Y	1.00	0.22	feature 98	feature 395
34 -> 395	1.00	0.01	1.37	1.31/0.73	Y	1.00	0.22	feature 34	feature 395
2 -> 395	1.00	0.03	3.00	12.13/0.73	Y	1.00	0.22	feature 2	feature 395
109 -> 395	1.00	0.19	4.74	0.44/0.73	Y	1.00	-	feature 109	feature 395
119 -> 395	1.00	0.49	5.67	2.02/0.73	Y	1.00	-	feature 119	feature 395
12 -> 395	0.99	0.01	2.02	0.49/0.73	Y	1.00	0.22	feature 12	feature 395

Block pair 1->2 - 1 candidate edges

• Out-degree: 1 parents, 1 children, 0 multi-parented (PolyFrac 0.0%); top-1 parent holds 100.0% of edges, Gini 0.996, max out-degree 1.
• Superparents (out-degree flag): 0 (0 also pass the old fire-rate AND-gate)
• Independence null: mean edge PMI 2.11; 0 edges (0.0%) at chance level (PMI < 0.5). 0 edges dropped by the joint-support guard (n_joint < 30).
• Recon-ablation contribution filter (Tree-SAE-inspired baseline): 1/1 edges pass (100.0%).
• Frequency control: mean survival 1.013 over 1 testable edges; 0 (0.0%) are frequency-driven (survival < 0.5).
• Sibling redundancy: n/a (child block not in SIBLING_BLOCKS).
• Joint-child (exact union, parents with edges): R_supp mean 0.928, R_mass mean 0.932; 0 parents with one child holding >=90% of their energy (rename candidates).
• Joint-child coverage (min(1, ΣF) upper bound — saturates when children co-fire, kept only for contrast with the exact union): 0.035.

parent -> child	R	F	PMI	recon P/C gain	recon?	surv	sib	parent label	child label
271 -> 544	0.91	0.03	2.11	0.13/0.05	Y	1.01	-	feature 271	feature 544

Block pair 2->3 - 0 candidate edges

• Out-degree: 0 parents, 0 children, 0 multi-parented (PolyFrac 0.0%); top-1 parent holds 0.0% of edges, Gini 0.000, max out-degree 0.
• Superparents (out-degree flag): 0 (0 also pass the old fire-rate AND-gate)
• Independence null: mean edge PMI nan; 0 edges (0.0%) at chance level (PMI < 0.5). 1 edges dropped by the joint-support guard (n_joint < 30).
• Recon-ablation contribution filter (Tree-SAE-inspired baseline): 0/0 edges pass (0.0%).
• Frequency control: mean survival nan over 0 testable edges; 0 (0.0%) are frequency-driven (survival < 0.5).
• Sibling redundancy: n/a (child block not in SIBLING_BLOCKS).
• Joint-child (exact union, parents with edges): R_supp mean nan, R_mass mean nan; 0 parents with one child holding >=90% of their energy (rename candidates).
• Joint-child coverage (min(1, ΣF) upper bound — saturates when children co-fire, kept only for contrast with the exact union): n/a.

Block pair 3->4 - 0 candidate edges

• Out-degree: 0 parents, 0 children, 0 multi-parented (PolyFrac 0.0%); top-1 parent holds 0.0% of edges, Gini 0.000, max out-degree 0.
• Superparents (out-degree flag): 0 (0 also pass the old fire-rate AND-gate)
• Independence null: mean edge PMI nan; 0 edges (0.0%) at chance level (PMI < 0.5). 0 edges dropped by the joint-support guard (n_joint < 30).
• Recon-ablation contribution filter (Tree-SAE-inspired baseline): 0/0 edges pass (0.0%).
• Frequency control: mean survival nan over 0 testable edges; 0 (0.0%) are frequency-driven (survival < 0.5).
• Sibling redundancy: n/a (child block not in SIBLING_BLOCKS).
• Joint-child (exact union, parents with edges): R_supp mean nan, R_mass mean nan; 0 parents with one child holding >=90% of their energy (rename candidates).
• Joint-child coverage (min(1, ΣF) upper bound — saturates when children co-fire, kept only for contrast with the exact union): n/a.

Block pair 4->5 - 0 candidate edges

• Out-degree: 0 parents, 0 children, 0 multi-parented (PolyFrac 0.0%); top-1 parent holds 0.0% of edges, Gini 0.000, max out-degree 0.
• Superparents (out-degree flag): 0 (0 also pass the old fire-rate AND-gate)
• Independence null: mean edge PMI nan; 0 edges (0.0%) at chance level (PMI < 0.5). 0 edges dropped by the joint-support guard (n_joint < 30).
• Recon-ablation contribution filter (Tree-SAE-inspired baseline): 0/0 edges pass (0.0%).
• Frequency control: mean survival nan over 0 testable edges; 0 (0.0%) are frequency-driven (survival < 0.5).
• Sibling redundancy: n/a (child block not in SIBLING_BLOCKS).
• Joint-child (exact union, parents with edges): R_supp mean nan, R_mass mean nan; 0 parents with one child holding >=90% of their energy (rename candidates).
• Joint-child coverage (min(1, ΣF) upper bound — saturates when children co-fire, kept only for contrast with the exact union): n/a.

Block pair 5->6 - 0 candidate edges

• Out-degree: 0 parents, 0 children, 0 multi-parented (PolyFrac 0.0%); top-1 parent holds 0.0% of edges, Gini 0.000, max out-degree 0.
• Superparents (out-degree flag): 0 (0 also pass the old fire-rate AND-gate)
• Independence null: mean edge PMI nan; 0 edges (0.0%) at chance level (PMI < 0.5). 3 edges dropped by the joint-support guard (n_joint < 30).
• Recon-ablation contribution filter (Tree-SAE-inspired baseline): 0/0 edges pass (0.0%).
• Frequency control: mean survival nan over 0 testable edges; 0 (0.0%) are frequency-driven (survival < 0.5).
• Sibling redundancy: n/a (child block not in SIBLING_BLOCKS).
• Joint-child (exact union, parents with edges): R_supp mean nan, R_mass mean nan; 0 parents with one child holding >=90% of their energy (rename candidates).
• Joint-child coverage (min(1, ΣF) upper bound — saturates when children co-fire, kept only for contrast with the exact union): n/a.

Block pair 6->7 - 3 candidate edges

• Out-degree: 3 parents, 2 children, 1 multi-parented (PolyFrac 50.0%); top-1 parent holds 33.3% of edges, Gini 0.987, max out-degree 1.
• Superparents (out-degree flag): 0 (0 also pass the old fire-rate AND-gate)
• Independence null: mean edge PMI 4.78; 0 edges (0.0%) at chance level (PMI < 0.5). 4 edges dropped by the joint-support guard (n_joint < 30).
• Recon-ablation contribution filter (Tree-SAE-inspired baseline): 3/3 edges pass (100.0%).
• Frequency control: mean survival 1.069 over 1 testable edges; 0 (0.0%) are frequency-driven (survival < 0.5).
• Sibling redundancy: n/a (child block not in SIBLING_BLOCKS).
• Joint-child (exact union, parents with edges): R_supp mean 0.707, R_mass mean 0.693; 0 parents with one child holding >=90% of their energy (rename candidates).
• Joint-child coverage (min(1, ΣF) upper bound — saturates when children co-fire, kept only for contrast with the exact union): 0.042.

parent -> child	R	F	PMI	recon P/C gain	recon?	surv	sib	parent label	child label
1504 -> 1698	1.00	0.00	4.70	0.01/0.02	Y	-	-	feature 1504	feature 1698
1463 -> 1698	0.94	0.00	4.76	0.03/0.02	Y	-	-	feature 1463	feature 1698
1362 -> 1663	0.87	0.12	4.87	0.20/0.08	Y	1.07	-	feature 1362	feature 1663