Publish a do-not-build register with evidence-based review triggers #93

Closed AtlantisPleb opened this 23h ago 1 comment

Outcome

Publish a durable register of retired product paths, the evidence that retired them, and the new evidence required before reconsideration.

Current behavior

docs/episode-triage.md records retired ideas in prose, but backlog generators and contributors cannot query those decisions. A new agent can rediscover an attractive old mechanism without seeing the failure that made it unsafe or obsolete.

Initial entries

  • Spark or other hosted agent custody.
  • NIP-90 data-vending-machine markets.
  • Bun in the production trust path.
  • GetAfter as a second forge.
  • Pay-for-online or presence-based mining.
  • Copilot as an executor target.
  • A monolithic all-in-one business operating system.
  • A Rust Sarah service.
  • Claude Code or another rival-owned runtime as a hard dependency.

Contract

  • Give each entry a stable ID, retired scope, decision date, evidence, violated principle, replacement path, owner, and reconsideration trigger.
  • Distinguish retired from deferred, rejected, and superseded.
  • Expose the register to humans, agents, issue templates, and FastFollow backlog generation.
  • Require new evidence and an explicit decision record before reopening a retired path.
  • Preserve history when a decision changes.

Acceptance criteria

  • An agent proposing a matching issue receives the register entry and replacement path.
  • The initial entries link current evidence and name a concrete reconsideration threshold.
  • A register update is reviewable, actor-attributed, and available through a machine-readable endpoint.
  • The register never blocks unrelated work through keyword matching alone.
  • FastFollow records when it suppresses a proposal because of a register entry.

Verification

Test discovery, matching, false-positive avoidance, state changes, API authorization, and FastFollow integration. Review the initial entries against current docs before publication.

Dependencies

Coordinate the FastFollow issue in project 7 and the promises registry issue in project 11.

  1. AtlantisPleb opened this issue 23h ago
  2. A AtlantisPleb Author 19h ago

    Implemented in e46af8833ffc42f5f65f26af1d8219e79c02e189: a versioned public do-not-build contract now records all nine initial decisions, exact match phrases, required decision metadata, and append-only history. Human docs and the issue-creation checklist link the machine-readable endpoint.

    FastFollow screening suppresses precise matches, records only bounded metadata plus a proposal fingerprint, avoids broad keyword blocking, and requires both new evidence and an explicit decision record before returning manual review. Contract, matching, false-positive, history, analytics, reconsideration, route-authority, and public endpoint coverage passes in the 185-test focused suite.

    Written by Devin

  3. closed this as completed 19h ago
Sign in with GitHub to comment on this issue.