You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[Rewrite P2] NIP currency: implement audit-identified gaps in Rust #870
Implement the NIP gaps Phase 0's audit finds, in the Rust core/relay — protocol currency is judged at the capture date, and the audit (#861) produces the authoritative list.
Anything deferred carries the reason in docs/rewrite/nip-parity.md
Acceptance:supportedNips in the Rust relay matches #861's audited must-have list; every advertised NIP has implementation + vectors.
Estimate basis (2026-10-24): $2.50 - the NIP gaps list is unknown until #861's audit output is read in full; bracketed by job 008's per-module cost at one to three modules of work. Model and per-unit costs: .factory/estimates.md.
Shipped in the squash-merge-free merge 4adaa42392526d750f0cd7614d317ed192ffedb1 (PR #1094, merged into rewrite/main at 2026-11-09T16:41:37Z by Cameri; tree byte-identical to the reviewed head c2e3db8, git diff c2e3db8 4adaa42 empty).
What landed
The advertised set is the Phase 0 audit's 33 must-have rows plus NIP-67 — 34 ids, ascending, pinned by a test that restates the list literally (an expectation read from the constant stays green under every wrong value).
Six NIP vectors (14, 22, 28, 50, 65, 89), NIP-44's 6-byte prefix and clause ceilings derived at compile time, NIP-09 a-tag deletion with a 100-distinct-address cap and resolve+delete sharing one transaction, and a nip-generic-events harness session (93 steps over 17 sessions, 1 declared tolerance; census untouched at 658).
NIP-67 hints["EOSE", <subid>, [<hint>…]] per the registry — feat(nip67): support EOSE completeness hints #718's {"completeness": …} object is a different message and is rejected in the code comment.
Completeness is provable, not asserted.finish needs six conditions: clean stream end, no backend page cap, no spent page slot, the backend's matcher agreeing on refused rows, its predicate agreeing on search, and its tag index able to answer every criterion. The last two are declared capabilities beside page_bound() (search_rule, empty_tag_rule) rather than rules hardcoded in the relay.
Ten review rounds. Six shipped d4bbe4a; Greptile then failed the merge gate on a P1 the series never looked for — LMDB collected a page in one read transaction and read its values in another, so a key removed between them vanished without a trace and a short page read as complete, sending a false finish. Rounds 7–10 closed that (StreamedEvent::Vanished behind find_by_filters, counted as a spent slot), the second door the static review then found ({"#d":[""]} selects nothing from either index while Filter::matches accepts it), and the text those fixes outran.
Cost $6.635 against $2.50 scoped, 34 sessions, 14.93 h — the estimate was wrong in kind, and the regret line in the job's runs.jsonl says so.
Filed rather than fixed: #1090 (uncapped e-id arm), #1093 (finish can be false under nip56.hideActionableReports, default off), #1096–#1101 (pre-existing defects the round-7 static review and the approver found in touched files). The two Postgres arms — delete_by_addresses, empty_tag_rule — are read, not measured; one tests/parity.rs case against CI's postgres:16-alpine closes both.
Activity
chappie-daemon commentedon Nov 9, 2026
Shipped in the squash-merge-free merge
4adaa42392526d750f0cd7614d317ed192ffedb1(PR #1094, merged intorewrite/mainat 2026-11-09T16:41:37Z by Cameri; tree byte-identical to the reviewed headc2e3db8,git diff c2e3db8 4adaa42empty).What landed
a-tag deletion with a 100-distinct-address cap and resolve+delete sharing one transaction, and anip-generic-eventsharness session (93 steps over 17 sessions, 1 declared tolerance; census untouched at 658).["EOSE", <subid>, [<hint>…]]per the registry — feat(nip67): support EOSE completeness hints #718's{"completeness": …}object is a different message and is rejected in the code comment.finishneeds six conditions: clean stream end, no backend page cap, no spent page slot, the backend's matcher agreeing on refused rows, its predicate agreeing onsearch, and its tag index able to answer every criterion. The last two are declared capabilities besidepage_bound()(search_rule,empty_tag_rule) rather than rules hardcoded in the relay.Ten review rounds. Six shipped
d4bbe4a; Greptile then failed the merge gate on a P1 the series never looked for — LMDB collected a page in one read transaction and read its values in another, so a key removed between them vanished without a trace and a short page read as complete, sending a falsefinish. Rounds 7–10 closed that (StreamedEvent::Vanishedbehindfind_by_filters, counted as a spent slot), the second door the static review then found ({"#d":[""]}selects nothing from either index whileFilter::matchesaccepts it), and the text those fixes outran.Cost $6.635 against $2.50 scoped, 34 sessions, 14.93 h — the estimate was wrong in kind, and the
regretline in the job'sruns.jsonlsays so.Filed rather than fixed: #1090 (uncapped
e-id arm), #1093 (finishcan be false undernip56.hideActionableReports, default off), #1096–#1101 (pre-existing defects the round-7 static review and the approver found in touched files). The two Postgres arms —delete_by_addresses,empty_tag_rule— are read, not measured; onetests/parity.rscase against CI'spostgres:16-alpinecloses both.