Circuit duplicate (cross-model)

circuit:duplicate — this is the circuit (a composition of operators: a writer-op feeding a reader-op’s K/Q/V port). Not the duplicate operator, which is the head class this circuit is named after (the circuit is keyed by its reader operator). circuit:duplicate here vs op:duplicate there.

same-token reader (duplicate-token detection; IOI initiator)

Defining edge: (reader-side; writer often layer-0)

Cross-model edge liveness (path-patch: remove the writer from the reader’s key → attention collapse)

model	reader	writer	key collapse	writer is	value mover	value ΔV-out
gpt2	0.5	—	(skipped)	—	—	—
gpt2-medium	1.11	0.13	+0%	sink	0.14	0.02
gpt2-large	6.17	1.1	+4%	sink	3.1	0.40
gemma-2-2b	1.4	0.2	+3%	sink	0.0	0.19
Llama-3.2-1B	1.22	0.17	+9%	sink	0.17	0.28
Qwen2.5-1.5B	6.6	0.6	+13%	sink	3.8	0.12

Cross-model causal dossier (necessity / sufficiency / redundancy — via the ResidualVM)

The operator-dossier battery, lifted to this circuit and run on the unified ResidualVM (find_heads locates the heads, ablate_heads + nll measure the rest). Two next-token metrics: induction-NLL (in-context copy) and generic-NLL (general LM).

model	reader	necessity Δind-NLL	necessity Δgen-NLL	sufficiency (keep-only, ind)	reader redundancy
gpt2	0.5	+0.46	+0.22	+1%	distributed
gpt2-medium	7.11	+0.16	-0.01	+1%	distributed
gpt2-large	5.8	+0.41	+0.00	+0%	distributed
gpt2-xl	4.12	+0.35	+0.00	+1%	distributed
gemma-2-2b	1.4	-0.39	-0.83	-11%	distributed
Llama-3.2-1B	0.9	+0.97	+0.03	-1%	distributed
Qwen2.5-1.5B	8.3	+1.32	-0.01	-3%	distributed

• Necessity — Δ NLL when the circuit’s heads are mean-ablated (higher = more load-bearing for that behaviour). Generic-NLL necessity is small everywhere — these circuits are task-specific, not general-LM.
• Sufficiency — reconstruction coverage keeping only the circuit’s heads (MLPs intact); a small head-set that reconstructs the behaviour is an executable decompilation. (Generic-NLL coverage is omitted as a headline — with MLPs intact a tiny head-set scores high for reasons unrelated to the circuit; induction-NLL is the meaningful attention-circuit metric. Negative = keeping so few heads is worse than the all-ablated floor, the known keep-1-is-net-negative effect.)
• Redundancy — reader-head solo-vs-cumulative on induction-NLL: bottleneck = one head carries it, distributed = the population shares it.

Dossier data: runs/disassembly/circuits/dossier_summary.json (circuit_dossier_xmodel.py, built on the ResidualVM).

Data: runs/disassembly/circuits/atlas_summary.json. Regenerate: circuit_catalog_doc.py.