Module refresh_ledger 

Phase-exact stale-refresh evidence ledger (bead ibuuh.25).

Defines the canonical stale-refresh phase model and captures machine-readable timings, counters, and correctness artifacts for each phase. Downstream performance beads use this ledger as their proof framework: “what changed, how much, and was correctness preserved?”

Phase model

┌─────────┐   ┌─────────┐   ┌──────────┐   ┌─────────┐   ┌──────────┐   ┌──────────┐
│  Scan   │──▶│ Persist │──▶│ Lexical  │──▶│ Publish │──▶│ Analytics│──▶│ Semantic │
│ (disc.) │   │ (DB)    │   │ (rebuild)│   │ (commit)│   │ (stats)  │   │ (vectors)│
└─────────┘   └─────────┘   └──────────┘   └─────────┘   └──────────┘   └──────────┘
                                                              │
                                                              ▼
                                                         ┌──────────┐
                                                         │ Recovery │
                                                         │ (error)  │
                                                         └──────────┘

Modules

corpus_families

Standard benchmark corpus family identifiers.

Structs

BenchmarkCorpusConfig

Configuration for generating a benchmark corpus.

EquivalenceArtifacts

Correctness artifacts captured after a refresh for equivalence checking.

LedgerBuilder

Builder for incrementally recording phase data during a refresh cycle.

PhaseRecord

Timing and counter data for a single phase.

RefreshLedger

Complete evidence ledger for a single stale-refresh cycle.

RefreshLedgerEvidence

Single-shot derived evidence summary suitable for benchmark comparison and operator dashboards. Computed from a RefreshLedger in O(phases) time with zero allocations beyond the output structs.

RefreshLedgerEvidenceComparison

Cross-run comparison summary suitable for benchmark CI gates and regression dashboards. Computed by RefreshLedgerEvidence::compare_to.

RefreshPhaseDelta

One phase’s regression signal between baseline and current.

RefreshPhaseShare

Share of total wall-clock time spent in a single phase.

RefreshReadinessMilestones

User-facing readiness timing summary derived from a refresh ledger.

RefreshThroughputProfile

Per-phase throughput summary derived from a PhaseRecord.

RegressionVerdictThresholds

CI-bench-gate threshold configuration. Project-specific values let bench harnesses tune their tolerance: a noisy benchmark runner picks looser thresholds than a deterministic CI worker.

Enums

RefreshPhase

Canonical phases of a stale-refresh cycle.

RefreshSearchReadinessState

Why ordinary search can or cannot see the refreshed lexical asset yet.

RegressionVerdict

Hard-gate verdict for CI bench runners. Failure is the only signal that should cause a non-zero exit; Warning is for PR-comment / dashboard surfaces; Clean is the steady-state.