feat(history): the event history is a SQLite file, history.db

[V3RON]

Closes #629

Status

Implement: done Review: passed (round 2) Hardware: n/a Gate: waiting for the stack

Assumptions

• Assumption: the name of a set-aside -wal or -shm file — each file keeps its own name plus .damaged-<now> (history.db-wal.damaged-<now>), the reading of "each to <name>.damaged-<now>" that needs no new naming rule.
• Assumption: how the port tells "node:sqlite could not be loaded" from "the file is damaged" — NodeSqliteOpener.open throws SqliteUnavailableError for a failed load; any other throw from open is damage.
• Assumption: the status.get history field stays optional in the schema — making it required breaks about a hundred existing status fixtures in contract, gateway and CLI tests that the spec does not touch; every daemon of this protocol sends it, and the CLI prints no History line when it is absent.
• Assumption: the protocol is bumped from 42 to 43 (the spec says by one), with the four tests and manuals that name 42 moved to 43.
• Assumption: after a set-aside the fresh file takes only the events seen since the history was created, not the older ring events, and a ring event emitted before the history existed is written only when nothing was set aside — the red test a file whose quick_check answers something but ok expects only the later event in the fresh file, and the daemon test needs daemon.started, emitted while the set-aside awaits.
• Assumption: a failed insert or commit sends replay back to the ring, but a projection that throws (a clean rollback) leaves the file readable, so replay still reads it — the red tests a projection that throws leaves neither… and the conformance suite expect the file's rows after a projection failure, replays the ring with sinceTs once writing has stopped expects the ring after an insert failure.
• Assumption: HISTORY_PROJECTIONS is also exported from src/history/index.ts — the boundary rule lets the daemon import only a component's index, and the host has to hand the registry in.
• Assumption: the starting form of simlock status prints no History: line — before the socket claim the history is not yet started, so writing: false would read as a fault; --json carries history in that answer.
• Assumption: simlock config set writes eventLog as {} when rotateBytes was its only key, rather than deleting the block.

Review

Spec review: 2 blocking, 1 fixed, 1 notes. Code review: 3 blocking, 3 fixed, 0 notes. Claims review: 2 blocking, 2 fixed, 3 notes.
Rounds: 2. Tokens: 30.3M.
Assessment: complete

Fixed unreviewed (text, rule 14): src/contract/AGENTS.md:8 Imported by (0c1a557).

Rejected:

• spec: Assumption 6 (after a projection throws, replay still reads the file) contradicts the spec's "ring once writing has stopped" — [intentional] the assumption is declared, and the spec's own Tests line and the red tests require the file's rows.

Written by an agent.

[V3RON]

Review notes

Not blocking, not verified. Each is one reviewer's claim.

• spec: src/history/index.ts:261 — Assumption 6 differs from the spec's "ring once writing has stopped". Someone may want to settle which rule wins.
• claims: src/cli/AGENTS.md:5, src/daemon/AGENTS.md:5, src/gateway/AGENTS.md:6 — the Uses lines were fixed in ce9b4dc; recorded here for the ledger.

Written by an agent.