What your coding agents have learned and whether it helps

Native lifecycle clients capture the work they can observe; every supported MCP client can deliberately read and write durable findings. This page keeps the knowledge, its provenance, and the causal evidence about whether delivering it improved later work.

Hook and MCP capture health

Is remembering worth it?

Knowledge scope Search what it knows Kind

Drag to orbit, scroll to zoom, and select a node to inspect its evidence and history.

File
Symbol
Discovered
Task

Deterministic wiring checks, passive lifecycle traces, and randomized model evaluations are kept separate. Only matched evaluation pairs produce a causal effect interval.

Client Model Task Arm

Universal Cognitive Runtime

Event integrity, typed projection, cross-client efficiency, mistake recurrence, native guard enforcement, and the correctness-first release verdict. Smoke results remain separate from powered effectiveness claims.

Missing evidence producers

A blank release metric is never treated as zero. This table names the evidence class that must be produced before the metric can be calculated.

Evidence tiers

Executable artifacts

Causal cohorts

Cross-client mistake transfer

Natural producer harvests are kept separate from oracle controls. A claim passes only when the pre-registered capture, prevention, correctness, timing, and negative-control gates all pass.

Concurrent writer integrity

Accepted-write loss, JSONL corruption, orphaned findings, later-consumer delivery, and behavior remain separate facts.

Recent injection traces

Client capability contract

Rules/MCP-only clients are not labelled as native capture clients.