process · distilled-experience-memory TEC-0236

Distill lessons from past attempts into memory

After each task, have the agent judge whether it succeeded, distill a short reusable strategy or warning from the attempt, store it, and retrieve only the relevant ones at the start of later tasks.

The agent keeps a growing bank of lessons rather than a log: each item is a generalisable strategy from a success or a preventative lesson from a failure, written in its own words and indexed for retrieval. At test time it retrieves the few items that match the new task, acts, judges the outcome, and writes back. It is Hashimoto's ratchet run inside the agent with no person in the loop, and it differs from a guide file in two ways that matter: lessons come from the agent's own judged outcomes, and only the relevant ones are read, rather than the whole file on every task. The judgment step is the weak point, since the agent grades its own work.

Requires: A task stream long enough to accumulate lessons, a store with retrieval, and a success judge, ideally with access to ground truth.

Addresses: Improving its own scaffolding from failures

Contexts: Autonomous agent, Coding agent

Does it work?

nothing measured0 supporting · 0 contesting sources

Efficacy claims — what this technique actually moves, under which conditions, and whether that has been contested.

2 more conditions from claims reviewed by AI, listed with the claims below.

No efficacy claim filed yet. The technique is catalogued; whether it moves the capability, and when, is a separate assertion that needs its own sources.

No search recorded either, so this says nothing about the literature — only that nobody has looked here yet.

Code

Sources

Either way, bring evidence: a link, or what you saw.

Whether a technique works is a claim. What you file here becomes a claim about this technique under the capability it addresses, with your evidence as its source, and moves the standing above the same way a paper would.