> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tldt.me/llms.txt
> Use this file to discover all available pages before exploring further.

# Generate Derived Memory

> Use AI to turn selected Raw evidence into a reviewable Derived draft.

Derived generation helps when your source material is accurate but difficult to use in conversation. It compresses selected Raw files into a focused, first-person Markdown candidate.

## How to generate

<Steps>
  <Step title="Select up to 20 Raw documents">
    Choose only the sources needed for the topic. Smaller, focused selections make review easier.
  </Step>

  <Step title="Choose a preset or custom instruction">
    Pick a server-owned preset, or describe the summary you need. For example, ask for a project context card or a recruiter-focused role summary.
  </Step>

  <Step title="Choose the exact model and target path">
    TLDT uses the model you select. The generated document receives a target path and keeps source provenance.
  </Step>

  <Step title="Review the Private candidate">
    Generation always creates a Private candidate. Save it through the normal document flow only after checking every claim.
  </Step>
</Steps>

## What generation does not do

Generation does not publish, mark a document Public, invent missing experience, strengthen ownership or causality, silently change models, or silently truncate the selected sources. If the request is too large, TLDT returns a context-overflow error.

## Review the result

Reject or edit any candidate that changes a date, metric, denominator, time range, actor, seniority, or outcome. Also check that it preserves negative context and does not turn a preference or plan into a fact.

<Note>Use generation for organization, not authorization. You remain the reviewer who decides whether a candidate becomes part of a public version.</Note>