record state
frontier-ownedReview status
This finding is part of accepted frontier state. Review events, reviewable changes, and proof state explain how it can change.
unreviewed
frontiers / frontier
Finding bundle
back to staterecord state
frontier-ownedThis finding is part of accepted frontier state. Review events, reviewable changes, and proof state explain how it can change.
finding statement
finding typeNo entity list is declared.
evidence
source-boundtheoretical · manual state transition
proof impact
packet context1 reviewable changes and 0 evaluation records are attached to this finding id.
Evidence and conditions
method
manual state transition
evidence type
theoretical
conditions
Provenance
source title
Quantifying LLM Attention-Head Stability (2026)
authors
reviewer:will-blair
Residual stream activations show 30-50% higher seed consistency than individual attention head activations across model refits, suggesting residual-level features rather than head-specific circuits provide more robust ground truth for circuit discovery and universality claims.
vs_bd3d16868b182c1d · manual_curation
outgoing
vf_de8d1e8b8646db6aResidual stream stability (index 5) confirms the induction head scaling pattern (index 0) at a higher level of abstraction; suggests that phase transitions in in-context learning rely on stable residual-level representations rather than individual head circuits.
incoming
events
vev_b3ffb68e18e1115ffinding.assertedManual finding added to frontier state
reviewer:will-blair · 2026-05-29
reviewable changes
vpr_9dbf934c907e59d5finding.addManual finding added to frontier state
applied · reviewer:will-blair · 2026-05-29
evaluations
No evaluation record targets this finding id.