Phase 13 plan — journal and lifecycle commands

[mmacy]

Implementation plan for phase 13 of the spec: the engine substrate for authored triggers and quests, with no interpreter — the optional source field on the base command model, session-state blocks for trigger fired-marks and the journal, the MarkTriggerFired/AddJournalEntry/RecordNote referee commands with their events, the journal in the player view, and additive persistence of the new blocks. Phase 13 is order-independent of phase 12 (landed) and must precede phase 14, whose interpreter issues these commands and reads this state.

Notable decisions

• source is one field on the base Command (str | None, min_length=1), inherited everywhere, read by nothing in execution — pinned by a twin-session state-equality test, not prose. Serialization consequence planned: the four goldens that store serialized commands (phase4_delve, phase5_milestone, phase11_gates including its refusals block, phase12_wipe) regenerate with "source": null keys as the entire diff.
• Re-marking a fired trigger is idempotent-accept: state records the first firing, the event records each one. Load-bearing because replay_game raises on any rejected replayed command — a rejecting re-mark would poison every replay of a repeatable trigger's log under phase 14's semantics.
• Fired-marks are a list[str] in first-fired order — ordered and deterministic where sets are banned; nothing consumes a count.
• JournalEntry carries text plus a rounds clock stamp, pinned because append-time is capturable only at append: the view API promises front ends never replay the event log, and a compacted save sheds the log entirely. The plan argues the naming divergence from DeathRecord.round explicitly. The player view ships the entries verbatim (the spec's word) — no mirror view model.
• Visibility follows the flag precedent: journal-entry events player-visible (authored text as content data, the spec's carve-out); trigger-fired and note events referee-visibility. Fired-marks never reach the player view.
• Everything is additive at schema_version 3: two new payload keys with empty defaults, no migration, no new rejection codes, no Ruleset flag, no spec edit, and no adaptations entry — the phase touches no SRD text (B/X has no journal or trigger machinery; verified by search).

Review provenance

Rubber-ducked by a fresh skeptical reviewer with the ordered reading list (spec, phase 11/12 plans, AGENTS.md, the touched code, the goldens). First-pass verdict: SOLID, with four non-blocking wording/citation findings — the journal load expression shown as a raw-dict get instead of the per-entry model_validate pattern, the golden-regeneration fact citing session_state instead of the generators' own dumps (and omitting phase11_gates's refusals block), a docstring-framing justification attributed to a module docstring that never carried it, and the unremarked rounds/round naming divergence. All four were folded in and re-verified by the same reviewer, context intact: each fix CONFIRMED, final verdict SOLID. The reviewer's verified-good list covered, among ~23 checks: the four-golden census, the absence of any SRD surface, spec fidelity of the three commands' semantics and visibilities, the census/docs-gate mechanics, the cycle-free views.py import of JournalEntry, and the fuzzer impact of the new required string fields.

https://claude.ai/code/session_013aS1Fbhon3y9aipXJ8s24a