Skip to content

Memory and evidence

North Star memory is a compact claim supported by evidence, not a pile of chat transcripts.

Each project keeps a short statement of what is true now. It includes the verified source and deployed version when that matters, plus the remaining boundary or next useful entry point.

The claim changes only after the affected surface is verified.

Evidence can include:

  • tests, type checks, and build results
  • browser checks on desktop and mobile
  • live health responses
  • source and deployment commit receipts
  • screenshots or recorded observations
  • approval and rejection decisions
  • recovery files and rollback instructions

Each item is stored once and referenced. North Star does not duplicate the same proof into every summary.

Memory answers, “What should the next run believe?”

Evidence answers, “Why should it believe that?”

Keeping those questions separate makes it possible to stay concise without losing the audit trail.

A conversation can reveal a preference, a correction, or a new goal. North Star treats that as input, not automatically as permanent truth. Load-bearing facts should become an explicit project update with enough context to understand later.

Next: Briefing