---

pgmcp_experiment: language-product-cost-indexed-frontier title: LanguageProduct cost-indexed frontier versus legacy cost-distinct frontier date: 2026-07-31 project: liblevenshtein-rust kind: feature_refactor status: decided verdict: accepted p_value: 0.000000 plan: /home/dylon/.claude/plans/liblevenshtein-currently-supports-levens-generic-sunrise.md git_ref: 3f362b4

LanguageProduct cost-indexed frontier versus legacy cost-distinct frontier

Kind: feature_refactor | Status: decided | Correction: benjamini_hochberg

Method

Question: Does the generic cost-indexed LanguageProduct frontier reduce mean per-input transition latency across the pre-registered literal, alternation, and branching cases while preserving exact accepting distance?

Criterion was locked in the external implementation plan dated 2026-07-25 before measurement. This database registration precedes sample ingestion; correctness is a separate property/differential gate.

Hypotheses

H1. Across the fixed Phase-0 pattern/input corpus, the canonical minimum-cost LanguageProduct frontier decreases mean per-input transition latency relative to the legacy cost-distinct frontier while returning the same accepting distance. — ✅ accepted

• metric: transition_latency_ns (ns) · predicted: decrease · planned n/arm: 51
• pre-registered criterion (locked 2026-07-31 20:08:36Z): {"type": "welch_t", "params": {"tail": "less", "alpha": 0.05, "min_effect": {"kind": "cohens_d", "threshold": 0.5}}}

Measurements & Decisions

Metric	Test	Statistic	p	Effect	95% CI	Verdict
transition_latency_ns	welch_t	-11.866418	0.000000	-0.968889	[-497970.2678, -356308.5597]	accepted

Decision on transition_latency_ns:

ACCEPTED (criterion: welch_t, correction: BenjaminiHochberg) [0] WelchT: statistic=-11.8664, p=0.000000, effect=-0.9689, 95% CI=[-497970.2678, -356308.5597]

Operator note: ACCEPT for the implementation plan's runtime hypothesis: exact accepting-distance equality was asserted for every case before timing, 600-case differential/property tests independently pass, and treatment mean latency is materially lower. Limitation: this run used the performance governor but was not CPU-pinned; the effect is substantially larger than ambient variance. State expansions are not uniformly lower (158 vs 167 literal, 118 vs 57 alternation, 824 vs 371 branching), so no structural-pruning claim is made.

What did NOT work

Nothing rejected (or no decisions yet).

Reproducibility

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

Timeline

• 2026-07-31 20:08:36Z — opened: LanguageProduct cost-indexed frontier versus legacy cost-distinct frontier
• 2026-07-31 20:08:36Z — criterion_locked: Across the fixed Phase-0 pattern/input corpus, the canonical minimum-cost LanguageProduct frontier decreases mean per-input transition latency relative to the legacy cost-distinct frontier while returning the same accepting distance.
• 2026-07-31 20:08:52Z — run: treatment (treatment)
• 2026-07-31 20:08:52Z — run: control (control)
• 2026-07-31 20:09:02Z — decided: accepted on transition_latency_ns (welch_t)

---

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