Files
buzz/migrations
Riley Crane 839c4b637b refactor(accumulator): cut the relay-kinds + CLI path; daemon is the only door
Riley-approved parsimony pass now that the standalone daemon supersedes
the relay-event path:

- Remove kinds 30640/4640 end to end: kind.rs constants/allowlists/
  asserts, ingest.rs scope arms, migration 0041 + schema.sql FTS
  exclusion + migrator expectations (back to 40), and the entire
  'buzz folds' CLI group with its buzz-accumulator dependency. The
  branch now touches ONLY crates/buzz-accumulator plus the workspace
  manifests - nothing for the team relay to redeploy.
- Estimator keeps tokens, loses dollars: est_cost_usd and the curated
  PRICING table are gone (a 3-row USD table ignoring output tokens is
  false precision that goes silently stale); estimate_tokens + window
  fit stay so the UI can still gate a priced Run button.
- Strict citation grammar: a bracket is a citation iff its content is
  exactly one id; the multiple-of-64 hex-run splitting is deleted and
  a jammed multi-id bracket is now refused loudly.
- Collapse the one-entry schema registry: BUILTIN_SCHEMAS/builtin()
  and the duplicate unknown-schema checks in plan_run/complete_run are
  gone; FoldSpec::validate is the single seam and the schema string
  field stays as the versioning hook.
- Doc honesty for the SQLite reality: MAX_PRIOR_OUTPUT_BYTES is a
  model-input budget (value unchanged), artifacts persist locally
  instead of as NIP-44 relay events, and the daemon docs state the
  HTTP API is loopback-only by design - /run spends money, so any
  non-loopback bind must add auth first.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Signed-off-by: Riley Crane <rileycrane@squareup.com>
2026-09-04 10:04:48 -07:00
..