---

pgmcp_experiment: degenerate-metric-walkers title: Degenerate Hamming and indel dictionary walkers date: 2026-07-31 project: liblevenshtein-rust kind: optimization status: decided verdict: rejected p_value: n/a plan: /home/dylon/.claude/plans/liblevenshtein-currently-supports-levens-generic-sunrise.md git_ref: 3f362b4

Degenerate Hamming and indel dictionary walkers

Kind: optimization | Status: decided | Correction: benjamini_hochberg

Method

Question: Do the candidate Hamming and indel walkers satisfy the Phase-8 compound shipping gate of at least two-times faster at k in {1,2} on at least two dictionary sizes and at least four-times fewer enumerated edges?

The compound ship rule was locked in the external plan dated 2026-07-25. The first benchmark run was invalidated after it exposed sibling DoubleArrayTrieChar base-allocation corruption; D14 fixed that defect and this experiment ingests only the complete post-fix rerun.

Hypotheses

H1. Across the pre-registered 3 dictionary sizes × 3 query lengths × k in {1,2} matrix for both metrics, the candidate walker achieves the runtime clause and an aggregate edge-enumeration reduction of at least 4x relative to the honest baseline. — ❌ rejected

• metric: edge_reduction_factor (ratio) · predicted: increase · planned n/arm: —
• pre-registered criterion (locked 2026-07-31 20:10:56Z): {"type": "absolute_threshold", "params": {"op": "ge", "arm": "treatment", "stat": "mean", "value": 4.0}}

Measurements & Decisions

Metric	Test	Statistic	p	Effect	95% CI	Verdict
edge_reduction_factor	absolute_threshold	1.228961	n/a	n/a	—	rejected

Decision on edge_reduction_factor:

REJECTED (criterion: absolute_threshold, correction: BenjaminiHochberg) [0] AbsoluteThreshold: statistic=1.2290, p=n/a

Operator note: REJECT shipping the specialized walkers. The runtime clause passed comfortably: both Hamming and indel candidates were more than 2x faster at k=1 and k=2 for all three dictionary sizes and all three query lengths. The required structural clause failed: aggregate edge reduction was only 322,794/262,656 = 1.229x, and no measured configuration approached 4x. Per the locked compound rule, the walkers remain benchmark-only and Phase 8 must use the repaired GeneralizedAutomaton/reference path.

What did NOT work

• edge_reduction_factor: rejected (test=absolute_threshold, p=n/a)

Reproducibility

• git ref: 3f362b4
• See each hypothesis's pre-registered criterion above; raw samples are retained in experiment_samples.

Timeline

• 2026-07-31 20:10:56Z — opened: Degenerate Hamming and indel dictionary walkers
• 2026-07-31 20:10:56Z — criterion_locked: Across the pre-registered 3 dictionary sizes × 3 query lengths × k in {1,2} matrix for both metrics, the candidate walker achieves the runtime clause and an aggregate edge-enumeration reduction of at least 4x relative to the honest baseline.
• 2026-07-31 20:11:10Z — run: treatment (treatment)
• 2026-07-31 20:11:19Z — decided: rejected on edge_reduction_factor (absolute_threshold)

---

Rendered from the pgmcp experiment record (the structured source of truth). Edit the experiment, not this file.