concept
active
concept:anchoring-strength-sanchoring strength S
Composite score S = ρd − dr − log k predicting anchoring success.
Neighborhood — ranked by edge-count
Frameworks (2)
framework
- Unified Contextual Control Theory (UCCT)implementsA theory that pretrained latent patterns are bound to task targets via external semantic anchors; formalized by anchoring strength S.
- AIC/BIC Model Selection Criteriaanalogous_toUsed as theoretical motivation for UCCT's log k budget term as complexity penalty mirroring model selection
Claims (3)
claim
- A central claim about the operational value of S.
- S = ρd − dr − log k is a predictive correlate of anchoring success across few-shot, SFT, and CoT.associated_withUCCT's practical utility claim.
- Applied contribution claim: S enables 'add 2 more examples to cross threshold' decisions
Methods (3)
method
- whitening and z-scoring procedureimplementsCalibration protocol: whiten embeddings on dev pool, z-score ρd and dr per layer.
- per-dev z-scalingassociated_withStandardizing ρd and dr using dev-set means and stds to form dimensionless components of S.
- Compute per-layer S(ℓ) = ρ̃d(ℓ) - d̃r(ℓ) - log k after whitening and standardization.
Concepts (8)
concept
- Anchoring strength S = ρd - dr - log krelated_toThe calibrated score measuring how effectively anchors bind target patterns; a predictive correlate of success.
- Logistic success surrogateassociated_withPhenomenological fit P(success)=σ(αS+β) used to summarize sharpness and midpoints.
- cohesion ρdassociated_withWithin-cluster tightness of target pattern representations.
- Mismatch drassociated_withDistance between prior knowledge centroid and target pattern centroid, e.g., 1 - cos(eprior, eT).
- representational mismatch drassociated_withDistance between prior and target representations.
- Prior-Target Mismatch (dr)associated_withMeasures how far the target PT is from the prior P_prior; increases anchoring difficulty
- anchor budget kassociated_withNumber of few-shot exemplars provided.
- Target Pattern Cohesion (ρd)associated_withMeasures how tightly the target pattern PT clusters in representation space; one of three components of S
Questions (2)
question
- How much anchor budget is needed to flip behavior?associated_withPractical question addressed by S and k50.
- The specific gap UCCT addresses that prior phase/representation work left open
Artifacts (1)
artifact
- Main paper presenting UCCT and semantic anchoring framework.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- External structure (in-context examples, retrieval, tuning) that biases latent pattern activation.
- Spearman ρ measuring rank-order agreement between logit-based self-report and probe score; the paper's primary monotonic association metric
- Fixed set of representative task-solving agents (Opus 4.6, Sonnet 4.6, Qwen3-235B) used to compute harness-updating capability metrics
- Maximum layer-wise anchoring score across layers.
- Parameters controlling the influence of conditioning signals in the generative process.
- Classical RL algorithm adapted by the paper with modifications including clipped-surrogate losses and length-normalized advantages for agentic training.
- Scalar parameter modulating how strongly a steering vector shifts model activations; set to 15 for Exp1 and ±16 for Exp2