Spike. A composable, explainable risk score per symbol — a reason list, not a 0.83 float. Combines PageRank + churn + fan-in + coverage + effect tags + contract/public-API surface + nearby anchored invariants.
Rules:
- Derived at query time, never stored (so it can never go stale).
- Transparent, tunable weights — never learned (a learned risk model is unexplainable, the antithesis of rag-rat's provenance value).
Why it helps an agent: turns "this symbol is central" (PageRank alone) into "central + fragile + under-tested + externally visible — here's why." The cheapest unifier of the proof+consequence layer.
Part of #115 epic. Ref: docs/plans/2026-06-14-agent-value-strategy.md §4c.
Spike. A composable, explainable risk score per symbol — a reason list, not a 0.83 float. Combines PageRank + churn + fan-in + coverage + effect tags + contract/public-API surface + nearby anchored invariants.
Rules:
Why it helps an agent: turns "this symbol is central" (PageRank alone) into "central + fragile + under-tested + externally visible — here's why." The cheapest unifier of the proof+consequence layer.
Part of #115 epic. Ref:
docs/plans/2026-06-14-agent-value-strategy.md§4c.