Sprint 13 closed: KV completeness and PUC with control values

Sprint 13 is closed. ADR-0014 (Product Metrics — KV Completeness and PUC) is Accepted.

Vision asked for idempotent project understanding. Architecture named Knowledge Vitality and Project Understanding Consistency. Sprint 13 finishes the deferred KV inputs and gives PUC both a substrate probe and a control experiment with baselines.

Knowledge Vitality completeness

Archived-recovery and evidence-reuse rates now sit on memory_stats and the Handshake vitality line. Sync counts recoveries when upsert clears archived_at. Search records linked evidence access so reuse is measurable.

PUC — two layers

Fingerprint probe. Same store ⇒ same CEM id fingerprint, handshake top-N, and fixed search hit lists (scripts/puc-probe.mjs).

Control experiment. Same question suite scored for:

  1. Control — no Meronq (baseline / control values)
  2. meronq-mcp — product strict MCP (apps/mcp-server)
  3. meronq-local-mcp — optional dogfood MCP

Runner: pnpm puc:control (Cursor SDK; needs CURSOR_API_KEY). Protocol: docs/metrics/puc-control.md.

First dogfood table

Arm Score
control 0.04
mcp 1.00
liftMcp 0.96

Source: docs/metrics/runs/2026-07-28T21-54-36-226Z/summary.md.

Principle

Sprint 11: forget relevance, not facts.
Sprint 12: real work leaves a signal.
Sprint 13: measure memory health fully — and measure understanding with control values, not only Meronq’s own fingerprints.

← Back to blog