Skip to content

canopusrecord engine body

field
frontiers
loop
body
docs

vela-science/vela

frontier

open-transformer-circuit-evidence

evidence

vea_2d9b693def772f4d

resultsread-only vieweressays

Search Canopus

Jump to a section, signal, campaign, document, primitive, work path, frontier, record index, atlas, constellation, agent, capability, or full-state search.

frontiers / frontier

Open-transformer circuit evidence

CC-BY-4.0vfr_c94f4f79b4e341ad

id: vfr_c94f4f79b4e341ad
license: CC-BY-4.0
findings: 42
accepted core: 14
contested: 0

42

findings

6

links

22

sources

42

evidence

0

contested

0.73

avg conf

recordoverview state sources proofenginereviewbodygraph

Evidence atom

back to sources

Why do circuit faithfulness metrics (KL divergence, logit difference) fail to detect cooperative inhibition heads that individually score near-zero on attribution but prove critical for behavior, and what principled attribution metric would catch such non-additive interactions?

id: vea_2d9b693def772f4d
frontier: Open-transformer circuit evidence
source: vs_4d88f4c63fd49a20
finding: vf_d3dd34cd06e3d5ce

evidence boundary

unknown

theoretical

An evidence atom is an inspectable support unit. It is not a finding by itself; it supports or challenges a finding through review.

finding binding

methodological

Why do circuit faithfulness metrics (KL divergence, logit difference) fail to detect cooperative inhibition heads that individually score near-zero on attribution but prove critical for behavior, and what principled attribution metric would catch such non-additive interactions?

inspect finding →

source binding

Discovering Transformer Circuits via a Hybrid Attribution and Pruning Framework (2024); Transformer Circuit Faithfulness Metrics are not Robust (2024)

vs_4d88f4c63fd49a20

inspect source →

review context

unverified

1 events

1 reviewable changes and 0 evaluation records target this atom or its bound objects.

Evidence statement

Why do circuit faithfulness metrics (KL divergence, logit difference) fail to detect cooperative inhibition heads that individually score near-zero on attribution but prove critical for behavior, and what principled attribution metric would catch such non-additive interactions?

extraction method

manual_curation

support relation

unknown

condition refs

vcnd_52be67291d5b4c56

Caveats

missing evidence locator

Review, event, and evaluation records

2

events

vev_a861131b43c2f3a2finding.asserted
Manual finding added to frontier state
reviewer:will-blair · 2026-05-29

reviewable changes

vpr_6a5c8894b2da8c78finding.add
Manual finding added to frontier state
applied · reviewer:will-blair · 2026-05-29

evaluations

No evaluation rows are attached.