rom24-quickmud-python 2.8.21__tar.gz → 2.9.3__tar.gz

{rom24_quickmud_python-2.8.21 → rom24_quickmud_python-2.9.3}/CHANGELOG.md

@@ -7,6 +7,567 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [2.9.3]
+
+### Added
+- **`INV-014` — OBJECT-REGISTRY-MEMBERSHIP locked in** (`docs/parity/CROSS_FILE_INVARIANTS_TRACKER.md`, `tests/integration/test_inv014_object_registry_membership.py`): ROM `src/db.c:create_object` appends every freshly built `OBJ_DATA` to the global `object_list` unconditionally — every world-scan consumer (`src/magic.c:3737 spell_locate_object`, decay sweep, save) walks that list. Previously only `mud/spawning/obj_spawner.py:spawn_object` appended to `mud.models.obj.object_registry`; six other production sites built `Object` instances without registering, leaving freshly-created corpses, gore objects, money piles, shop clones, and DB-restored inventory invisible to `locate object`. Added `mud.models.object.create_object(prototype, *, instance_id=None)` as the canonical factory (appends and returns); routed every production construction site through it. `mud/skills/handlers.py:_iterate_world_objects` now walks `object_registry` first, computing the holder per ROM `src/magic.c:3747` (outermost `in_obj` chain → `carried_by` → `in_room` → `None` rendered as "somewhere"); a legacy room/character walk remains as a compat backstop for unit tests. Eight enforcement tests cover every construction path plus the homeless-object locate symptom.
+
+### Fixed
+- **`spell_locate_object` could not find homeless objects** (`mud/skills/handlers.py:_iterate_world_objects`): an object spawned but not yet placed (`in_room=None, carried_by=None, in_obj=None`) was skipped entirely, while ROM `src/magic.c:3756-3762` reports it as "one is in somewhere" (the `in_room == NULL` branch). Iterator now walks `object_registry` so registered-but-unplaced objects surface at parity with ROM.
+- **Six production sites built `Object` instances without registering** (handler.py, combat/death.py ×2, rom_api.py, commands/shop.py fallback, models/conversion.py): `create_money`, `_fallback_gore`, `_fallback_corpse`, `recursive_clone`, `_clone_inventory_object` (when `spawn_object` returned `None`), and `load_objects_for_character` all bypassed `object_registry`. ROM `create_object` always appends. All six now route through `create_object` (or `spawn_object` where applicable).
+
+## [2.9.2]
+
+### Added
+- **`INV-013` — OBJECT-LOCATION-COHERENCE locked in** (`docs/parity/CROSS_FILE_INVARIANTS_TRACKER.md`, `tests/integration/test_inv013_object_location_coherence.py`): `Object.location` is no longer a stored field — it is now a property dispatching to the three canonical ROM container fields (`in_room`, `carried_by`, `in_obj`) per `src/handler.c:1626 obj_to_char` / `1953 obj_to_room` / `1968 obj_to_obj`. Setting `location` to a Room/Character/Object sets the matching ROM field and clears the other two; setting it to None clears all three; reads return whichever ROM field is non-None. Eliminates the latent divergence where callers had to remember to update both `obj.location` and `obj.in_room`. Seven enforcement tests.
+
+### Fixed
+- **`MobInstance.add_to_inventory` cleared `carried_by` right after setting it** (`mud/spawning/templates.py:445-446`): the method did `obj.carried_by = self; obj.location = None`. Pre-INV-013 the second line was a redundant legacy-field reset; under the new dispatch it cleared the carried_by we just set. Deleted the redundant `obj.location = None` line. Surfaced as a P-reset regression (`tests/test_spawning.py::test_reset_P_fills_mob_carried_container`) when the reset_handler could no longer find the mob-carried container.
+- **`make_corpse` left money objects with no container linkage** (`mud/combat/death.py:441`): money was appended to `corpse.contained_items` but `money_obj.location` was set to None, leaving `money_obj.in_obj` unset. Per ROM `src/handler.c:1968 obj_to_obj`, money inside a corpse must have `in_obj = corpse`. Changed to `money_obj.location = corpse`.
+- **Two `act_wiz` parity tests passed `location=-1` to the `Object` constructor** (`tests/integration/test_act_wiz_command_parity.py:398,582`): a Room or None was expected; `-1` was a stale placeholder from earlier refactors. Removed the kwarg — the tests already set `obj.in_room = room` on the next line.
+
+## [2.9.1]
+
+### Fixed
+- **`test_wait_and_daze_decrement_on_violence_pulse` — long-standing red test was asserting non-ROM behavior** (`tests/test_game_loop_wait_daze.py:27`, ROM `src/comm.c:616-621` + `src/fight.c:191-196`): the test created a `Character` without a descriptor and expected wait/daze to decrement by 1 per `game_tick`. ROM's descriptor input loop (`comm.c:616-621`) decrements only for characters with a `d->character` descriptor; descriptor-less actors decrement in `PULSE_VIOLENCE`-sized chunks via `violence_update` (`fight.c:191-196`). Fix: add `ch.desc = object()` to the test fixture so it exercises the descriptor path it was clearly written for. Renamed to `test_wait_and_daze_decrement_per_pulse_for_connected_character` to match its actual contract. Per AGENTS.md "a test asserting non-ROM behavior is a bug in the test, not the implementation."
+
+## [2.9.0]
+
+### Added
+- **`INV-012` — OBJECT-LIST-CANONICAL locked in** (`docs/parity/CROSS_FILE_INVARIANTS_TRACKER.md`, `tests/integration/test_inv012_object_list_canonical.py`): single canonical `Object` runtime class. `mud.models.obj.ObjectData` deleted. `object_registry` is now populated at spawn (`mud/spawning/obj_spawner.py:spawn_object`) and drained at extract (`mud/game_loop.py:_extract_obj`, including recursive contents per ROM `src/handler.c:2063-2067`). New ROM-named fields on `Object`: `in_room`, `in_obj`, `carried_by` (dataclass fields with `compare=False` to avoid `__eq__` graph recursion), `pIndexData` (read+write `@property` aliased to `prototype`), `contains` (read-only `@property` aliased to `contained_items`). Eight enforcement tests + smoke tests for `get_obj_world` and `obj_update`.
+
+### Fixed
+- **`object_registry` was never populated in production** (now-live behavior): every iteration over the global instance list was a no-op before this consolidation, silently disabling locate-object spells (`mud/world/obj_find.py:get_obj_world`, `mud/magic/effects.py`), mobprog oload triggers (`mud/mobprog.py`), global object scans (`mud/skills/handlers.py`), music decay (`mud/music/__init__.py`), and object decay tick (`mud/game_loop.py:obj_update`). Surfaced and closed under INV-012 — six smoke + correctness tests gate the behavior; per-system end-to-end coverage deferred to follow-up sessions.
+
+### Changed
+- ~12 `isinstance(target, ObjectData)` / `isinstance(target, (Object, ObjectData))` branches across `mud/skills/handlers.py` and `mud/game_loop.py` collapsed to `Object`-only.
+- 17 helper signatures in `mud/game_loop.py` and 3 in `mud/handler.py` re-typed from `ObjectData` to `Object`. 4 dual-shape `getattr(obj, "contained_items", None) or getattr(obj, "contains", [])` fallbacks in `mud/game_loop.py` collapsed.
+- `mud/mob_cmds.py:_extract_runtime_object` dead `isinstance(obj, ObjectData)` dispatch branch removed; local cleanup is the single canonical path.
+- 35 test fixtures across 9 test files migrated from `ObjectData(item_type=X, ...)` to `Object(instance_id=None, prototype=ObjIndex(...))`.
+- `tests/conftest.py` adds an autouse fixture that snapshots/clears/restores `object_registry` between tests, preventing leakage from the 10 test files that use `spawn_object`.
+
+### Removed
+- `mud.models.obj.ObjectData` (the dual-class runtime). Re-export dropped from `mud.models.__init__`.
+
+## [2.8.79]
+
+### Added
+- **`INV-011` — CARRY-WEIGHT-COHERENCE locked in** (`docs/parity/CROSS_FILE_INVARIANTS_TRACKER.md`, `tests/integration/test_inv011_carry_weight_coherence.py`): new cross-file invariant codifies that `ch.carry_weight` / `ch.carry_number` stay in lockstep with `ch.inventory + ch.equipment.values()` after every mutation path. ROM mirrors via `src/handler.c:1626 obj_to_char` / `1642 obj_from_char`. Five enforcement tests cover the canonical helpers (`Character.add_object` / `equip_object` / `remove_object`) and the runtime extract paths.
+
+### Fixed
+- **`_extract_obj` left stale carry counters on the carrier** (`mud/game_loop.py:784 _remove_from_character`, ROM `src/handler.c:2051 → 1642 obj_from_char`): the runtime extract path used by `_extract_obj`, corpse decay, and light-source decay removed the obj from `character.inventory` / `character.equipment` but never re-derived `carry_weight` or decremented `carry_number`. Every extract on a carried object skewed encumbrance upward; over a long-running session a player could become permanently over-encumbered with zero visible items. Fix: after the inventory/equipment removal, call `_recalculate_carry_weight()` and subtract the obj's `_object_carry_number` slot cost. Surfaced and closed under INV-011.
+
+### Watch list
+- New cross-file watch item: dual `Object` (`mud/models/object.py`) vs `ObjectData` (`mud/models/obj.py`) runtime classes — `object_registry: list[ObjectData]` is never populated because `spawn_object` constructs `Object`. Every iteration over `object_registry` (mobprog oload triggers, `get_obj_world`, locate-object spells, object decay) is a no-op in production. Parallel shape to INV-008. Tracked in `CROSS_FILE_INVARIANTS_TRACKER.md` watch list pending a multi-session consolidation strategy.
+
+## [2.8.78]
+
+### Added
+- **`INV-010` — ROOM-PEOPLE-COHERENCE locked in** (`docs/parity/CROSS_FILE_INVARIANTS_TRACKER.md`, `tests/integration/test_inv010_room_people_coherence.py`): new cross-file invariant codifies the bidirectional contract between `char.room` and `room.people`. Six enforcement tests exercise the canonical helpers (`Room.add_character` / `Room.remove_character`), the `char_to_room` NULL → temple fallback, the imm_commands `_char_from_room` / `_char_to_room` duplicates, `MobInstance.move_to_room`, the raw `room.people.remove/append` pattern from `do_recall`, and a global registry sweep.
+
+### Fixed
+- **Dual `room_registry` divergence** (`mud/models/room.py:204`, ROM `src/db.c:get_room_index`): `mud.models.room` declared a second `room_registry` dict that the world loader never populated. `char_to_room`'s temple fallback (`mud/models/room.py:67`) and `mud/game_loop.py:525`'s limbo lookup read from this empty dict and silently no-oped — a NULL-room redirect dropped the character on the floor with `ch.room = None` instead of routing to `ROOM_VNUM_TEMPLE`. Fix: re-export the canonical `mud.registry.room_registry` from `mud/models/room.py` so both readers see the world-loaded table. Surfaced and closed under INV-010.
+
+## [2.8.77]
+
+### Fixed
+- **`TRAIN-001` — `do_train` listing branch crashed on unrecognized args** (`mud/commands/advancement.py:315-324`, ROM `src/act_move.c:1713-1745`): any `train <typo>` (e.g. `train magic`, `train magic missile`) routed through the listing branch and raised `AttributeError: 'Character' object has no attribute 'perm_str'`. ROM reads `ch->perm_stat[STAT_*]`; QuickMUD stores the same data on `char.perm_stat: list[int]` — there are no `perm_str`/`perm_int`/`perm_wis`/`perm_dex`/`perm_con` attributes on `Character`. Fix: index `char.perm_stat` by STAT_STR..STAT_CON (0..4) for the max-stat comparison. Tests: `tests/test_advancement.py::test_train_lists_available_stats_without_crash`, `::test_train_lists_only_unmaxed_stats`.
+
+## [2.8.76]
+
+### Fixed
+- **`CAST-001` — `do_cast` target resolution honors ROM `TAR_*` dispatch** (`mud/commands/combat.py:704`, ROM `src/magic.c:301-536`): `do_cast` previously defaulted `target = char` whenever no target arg was given, regardless of the spell's target type. Casting `'magic missile'` mid-combat (offensive spell, `ch->fighting != NULL`, no explicit victim) hit the caster instead of the fighting victim (`Your magic missile scratches you.`). Fix: dispatch on `skill.target` against the ROM `TAR_*` matrix — `"victim"` / `"character_or_object"` (TAR_CHAR_OFFENSIVE / TAR_OBJ_CHAR_OFF) default to `char.fighting` and error `"Cast the spell on whom?"` if not fighting; `"friendly"` (TAR_CHAR_DEFENSIVE) defaults to self; `"self"` / `"ignore"` (TAR_CHAR_SELF / TAR_IGNORE) bind to the caster. Object-only and PK-gate paths are noted as scope-deferred in `docs/parity/MAGIC_C_AUDIT.md`. Tests: `tests/test_skills_spells_cast_listing.py::test_do_cast_offensive_no_target_defaults_to_fighting_victim`, `::test_do_cast_offensive_no_target_no_fight_errors`.
+
+### Docs
+- Added `docs/parity/MAGIC_C_AUDIT.md` as a stub for the eventual full `magic.c` audit; currently tracks only `CAST-001 ✅ FIXED` plus scope notes for object-targeted spells and PK gates.
+
+## [2.8.75]
+
+### Fixed
+- **`do_cast` failed to dispatch handlers and resolve targets** (`mud/commands/combat.py:704-805`, ROM `src/magic.c:299-360`): three latent bugs the `2.8.74` `find_spell` fix exposed —
+  - Handler lookup used `getattr(skill, "handler", None)`, which is always `None` because `SkillRegistry` stores handlers in `skill_registry.handlers[name]` rather than on the `Skill` object. Every cast therefore returned `"The spell '<name>' is not fully implemented yet."` Fixed to read from `skill_registry.handlers.get(skill.name)`.
+  - Handler call passed three arguments (`spell_func(char, target, spell_level)`) but every entry in `mud/skills/handlers.py` is `(caster, target=None)`. The `TypeError` was swallowed by the broad `except` and returned as `"Spell cast failed: …"`. Fixed by matching the canonical `(caster, target)` signature used by `skill_registry.use()`.
+  - Target lookup iterated `getattr(room, "characters", [])`, but the `Room` model stores occupants on `room.people`. Result: `cast magic missile fido` silently self-targeted. Fixed by iterating `room.people`.
+  - `do_cast` also passed the handler's raw return value through to the dispatcher, which expects a `str`. Damage-spell handlers return an `int`; the dispatcher choked on `if "…" not in result`. Spells emit their flavour text via `char.messages` (the `apply_damage` / `_send_to_char` path), so `do_cast` now always returns a `"You cast <name>."` acknowledgment string.
+
+### Added
+- `tests/test_skills_spells_cast_listing.py::test_do_cast_magic_missile_dispatches_handler_and_damages_target` — end-to-end regression that places a caster + target in a `Room`, casts `'magic missile' Fido` through `do_cast`, and asserts the handler ran (HP dropped, mana consumed, no `"is not fully implemented yet"` / `"Spell cast failed"` string). This locks down all three latent paths above.
+
+## [2.8.74]
+
+### Fixed
+- **`do_skills` / `do_spells` listing always returned "No skills found." / "No spells found."** (`mud/commands/misc_info.py:93-260`, ROM `src/skills.c:256-485`): both handlers iterated `getattr(mud.registry, "skill_table", {})`, a non-existent attribute that silently fell back to an empty dict. They also read the wrong skill fields (`spell_fun`, `skill_level`) and the wrong learned-percent source (`char.pcdata.learned` rather than `char.skills`). Rewritten to iterate `mud.skills.skill_registry.skills`, discriminate spells via `skill.type == "spell"`, look up class-level requirements via `skill.levels[ch_class]`, and read learned percentages from `char.skills` — matching the data source `do_practice` already uses successfully. Added ROM-faithful `[all] [max [min]]` argument parsing.
+- **`do_cast <name>` reported "You don't know any spells of that name."** (`mud/commands/combat.py:704-791`, ROM `src/magic.c:299-360`): exact-match lookup against `char.skills` rejected ROM prefix abbreviations, and single-token argument parsing broke multi-word spells like `magic missile`. Rewritten to use `mud.utils.string_editor.first_arg` (ROM `one_argument` parity, single-quote aware) and `skill_registry.find_spell(...)` for prefix matching. Mana cost now follows ROM's `UMAX(skill.min_mana, 100 / (2 + level - skill.skill_level[class]))` formula, with the level+2 == required → 50 mana edge case.
+
+### Added
+- `tests/test_skills_spells_cast_listing.py` — 10 regression tests covering the three listing/cast paths against the canonical `data/skills.json` table, including NPC short-circuit, ROM "Arguments must be numerical or all." error, `cast 'magic missile' …` quoted multi-word parsing, and `cast magic …` prefix matching.
+
+## [2.8.73]
+
+### Added
+- **`magic.c + magic2.c` subsystem closed at 100%**: `spell_pass_door()` parity (`mud/skills/handlers.py:5884` mirroring `src/magic.c:3864`) verified by new runtime-path integration coverage in `tests/integration/test_spell_affects_persistence.py::TestSpellPassDoorIntegration` — the affect persists through `game_tick()`, drops `AffectFlag.PASS_DOOR` on expiry, emits the ROM "You feel solid again." wear-off message exactly once, and a duplicate cast on an already-affected target is rejected with the ROM "already out of phase" message.
+
+### Changed
+- `docs/parity/ROM_C_SUBSYSTEM_AUDIT_TRACKER.md` P0-3 row updated from 98% → 100% (no remaining missing functions).
+
+## [2.8.72]
+
+### Fixed
+- **`magic.c + magic2.c` affect persistence runtime parity** (`src/update.c:765-768`): `mud/affects/engine.py:tick_spell_effects` now mirrors ROM `char_update()` by applying the 20% per-point-pulse spell-strength fade (`level--` on `number_range(0, 4) == 0`) while durations decay through the real `game_tick()` path.
+
+### Added
+- **`magic.c + magic2.c` affect-persistence integration coverage**: `tests/integration/test_spell_affects_persistence.py` now proves three ROM runtime contracts through `game_tick()` — affects do not decay before `PULSE_TICK`, affect `level` can fade on the point pulse, and multi-entry spell affects emit exactly one wear-off message when they expire.
+
+## [2.8.71]
+
+### Fixed
+- **`skills.c` game-loop integration parity** (`src/fight.c:192-196`, `src/fight.c:2952`, `src/update.c:update_handler`): descriptor-less actors no longer burn `wait`/`daze` every Python tick. `mud/game_loop.py:violence_tick` now mirrors ROM's split semantics — connected characters recover one pulse at a time, while descriptor-less actors recover in `PULSE_VIOLENCE` chunks on the combat cadence. The stale hardcoded timer burn was removed from `mud/combat/engine.py:multi_hit`.
+
+### Added
+- **`skills.c` runtime-path integration coverage**: `tests/integration/test_skills_integration.py` now proves a skill command (`bash`) enters combat and advances through `game_tick()` on the violence boundary, and locks wait-state recovery to ROM cadence. Companion unit expectations were updated to match the restored timer model.
+
+## [2.8.70]
+
+### Fixed
+- **`PMOTE-003` — `do_pmote` and `do_smote` skip NPC viewers with `desc == NULL`** (`src/act_comm.c:1130`, `src/act_wiz.c:392-393`): ROM's pmote/smote viewer loops immediately `continue` when `vch->desc == NULL || vch == ch`, which excludes all NPC observers. Python's `mud/commands/imm_emote.py` had weakened that guard to `desc is None and not is_npc`, so NPCs incorrectly received both personalized pmote output and smote broadcasts. Fix: restore the ROM rule in both loops by skipping any non-self viewer whose `desc` is `None`. Locked in by `tests/integration/test_act_comm_gaps.py::TestPmoteGaps::test_pmote_003_npc_viewers_do_not_receive_pmote_or_smote`.
+
+## [2.8.69]
+
+### Fixed
+- **`PMOTE-002` — `do_pmote` viewer broadcast routes actor `$N` through PERS** (`src/act_comm.c:1136,1188`): ROM sends pmote output to each eligible viewer via `act("$N $t", vch, ..., ch, TO_CHAR)`, which evaluates `PERS(ch, vch)` per recipient so an invisible actor renders as `"someone"` to viewers without `DETECT_INVIS`. Python's `mud/commands/imm_emote.py:do_pmote` previously hardcoded `f"{char_name} {substituted}"`, leaking the invisible actor's identity on both the matched-name and no-match branches. Fix: route the actor prefix through `mud.world.vision.pers(char, viewer)` per viewer. Locked in by `tests/integration/test_act_comm_gaps.py::TestPmoteGaps::test_pmote_002_invisible_actor_renders_as_someone_to_unaided_viewer`.
+
+## [2.8.68]
+
+### Changed
+- Untrack `log/orphaned_helps.txt` (test artifact that drifted on every run) and add it to `log/.gitignore`. No code change.
+
+### Docs
+- **`PMOTE-001` — closed as ✅ FIXED in `docs/parity/ACT_COMM_C_AUDIT.md`** (`src/act_comm.c:1098-1192`): audit row was stale. `do_pmote` is fully implemented at `mud/commands/imm_emote.py:170` with NOEMOTE block, Moron guard, the letter-by-letter substitution loop mirroring ROM C `src/act_comm.c:1131-1175`, apostrophe/possessive handling (`"Bob's"` → `"your"`), and trailing-`s` absorption (`"Bobs"` → `"yous"`). Covered by 5 tests in `tests/integration/test_act_comm_gaps.py::TestPmoteGaps`. Two new sub-gaps filed during the re-check for follow-up: `PMOTE-002` (TO_CHAR `$N` prefix should route through PERS — invisible-actor leak parallel to EMOTE-001) and `PMOTE-003` (NPC viewers should be skipped via `desc is None`, not gated on `not is_npc` — same divergence in `do_smote`).
+
+## [2.8.67]
+
+### Fixed
+- **`DAMMSG-001/002/003` — `dam_message` per-hit broadcasts route `$n` / `$N` through PERS per recipient** (`src/fight.c:2218-2228`): ROM's `dam_message` ends with three `act()` calls — `TO_NOTVICT`, `TO_CHAR`, `TO_VICT` — each evaluating `PERS(ch, looker)` and `PERS(victim, looker)` independently for every observer. Python's `dam_message` in `mud/combat/messages.py` previously returned `DamageMessages` containing three pre-rendered strings baked with `attacker.name` / `victim.name`, then the engine shipped them via `_broadcast_room` / `_push_message` without per-recipient PERS substitution — leaking both identities to every observer regardless of `can_see`. This is the highest-volume PERS leak in combat (fires on every melee swing and spell hit). Fix: (a) `dam_message` now returns templates with `{attacker}` / `{victim}` placeholders (ROM colour codes `{3...{x` doubled to survive `str.format()`); (b) new `render_for(template, attacker, victim, observer)` helper in `mud/combat/messages.py` substitutes both names through `pers()` per recipient; (c) `_dispatch_damage_messages` renamed to `_broadcast_damage_messages` (back-compat alias kept) and now iterates `room.people` for TO_NOTVICT, calling `render_for` per occupant — TO_CHAR uses `observer=attacker`, TO_VICT uses `observer=victim`. The `damage()` return value (consumed by `multi_hit`'s `attack_message` results) is also rendered for the attacker's view so scripted/test callers receive PERS-correct strings. Locked in by three new failing-first tests in `tests/integration/test_dam_message_pers.py` (one per direction); `tests/test_combat_messages.py` updated to call `render_for` on the returned templates.
+
+## [2.8.66]
+
+### Fixed
+- **`FIGHT-014` — `_auto_sacrifice` TO_ROOM broadcast routes `$n` through PERS** (`src/act_obj.c:1856`, dispatched from `src/fight.c:961-970`): ROM's `act("$n sacrifices $p to Mota.", ch, obj, NULL, TO_ROOM)` substitutes `$n` per-listener through PERS so an invisible attacker renders as `"someone"` to room observers without `DETECT_INVIS`. Python's `_auto_sacrifice` in `mud/combat/engine.py` previously pre-rendered the broadcast via `expand_placeholders(...)` baked into a single `_broadcast_room` string keyed on `attacker.name`, leaking the attacker's identity. Fix: route through `_broadcast_pos_change(attacker, "{name} sacrifices {corpse} to Mota.", corpse=corpse_name)`. Orphan `SimpleNamespace` import removed as a side effect (the only callsite was the broadcast we just rewrote). Locked in by `tests/integration/test_auto_sacrifice_pers.py::test_fight_014_auto_sacrifice_broadcast_uses_pers_for_invisible_attacker`.
+
+### Notes
+
+- Python's `_auto_sacrifice` re-implements sacrifice logic inline rather than dispatching to `do_sacrifice` like ROM does at `src/fight.c:970`. That structural divergence (parallel implementation vs ROM dispatch) is tracked as FIGHT-015 (reserved) for a future session.
+
+## [2.8.65]
+
+### Fixed
+- **`FIGHT-013` — WEAPON_SHOCKING TO_ROOM broadcast routes `$n` through PERS** (`src/fight.c:673-674`): ROM's `act("$n is struck by lightning from $p.", victim, wield, NULL, TO_ROOM)` substitutes `$n` per-listener through PERS. Python's shocking branch in `process_weapon_special_attacks` previously baked `victim.name` into a fixed `_broadcast_room` string. Fix: route through `_broadcast_pos_change`. Closes the FIGHT-009..013 weapon-proc PERS sweep — all five weapon special-attack TO_ROOM broadcasts now ROM-faithful. Locked in by `tests/integration/test_weapon_proc_pers.py::TestWeaponProcBroadcastPers::test_fight_013_shocking_broadcast_uses_pers_for_invisible_victim`.
+
+### Notes
+
+- With FIGHT-009..013 closed the `mud/combat/engine.py` PERS surface is fully normalized for both position-change broadcasts (FIGHT-004..008) and weapon-proc broadcasts (FIGHT-009..013). Remaining PERS surfaces in combat: `dam_message` (ROM `src/fight.c:2035-2233`, the per-hit damage-tier broadcast surface — hundreds of `act()` lines), and `do_sacrifice` in `mud/combat/engine.py:956` (single `_broadcast_room` call with `expand_placeholders`-rendered fixed string).
+
+## [2.8.64]
+
+### Fixed
+- **`FIGHT-012` — WEAPON_FROST TO_ROOM broadcast PERS + ROM-true wording** (`src/fight.c:663`): Two divergences. (a) PERS gap on `$n` — invisible victim leaks identity. (b) Wording — ROM `act("$p freezes $n.", ...)` puts the weapon first (e.g. `"the sword freezes Alice."`) but Python previously emitted `f"{victim.name} is frozen by {weapon_name}."` (subject/object inverted). Fix: `_broadcast_pos_change(victim, "{weapon} freezes {name}.", weapon=weapon_name)` — single change closes both sub-gaps. Locked in by `tests/integration/test_weapon_proc_pers.py::TestWeaponProcBroadcastPers::test_fight_012_frost_broadcast_uses_pers_and_rom_wording`.
+
+## [2.8.63]
+
+### Fixed
+- **`FIGHT-011` — WEAPON_FLAMING TO_ROOM broadcast routes `$n` through PERS** (`src/fight.c:654`): ROM's `act("$n is burned by $p.", victim, wield, NULL, TO_ROOM)` substitutes `$n` per-listener through PERS. Python's flaming branch in `process_weapon_special_attacks` previously baked `victim.name` into a fixed `_broadcast_room` string. Fix: route through `_broadcast_pos_change`. Locked in by `tests/integration/test_weapon_proc_pers.py::TestWeaponProcBroadcastPers::test_fight_011_flaming_broadcast_uses_pers_for_invisible_victim`.
+
+## [2.8.62]
+
+### Fixed
+- **`FIGHT-010` — WEAPON_VAMPIRIC TO_ROOM broadcast routes `$n` through PERS** (`src/fight.c:643`): ROM's `act("$p draws life from $n.", victim, wield, NULL, TO_ROOM)` substitutes `$n` per-listener through PERS. Python's vampiric branch in `process_weapon_special_attacks` previously baked `victim.name` into a fixed `_broadcast_room` string. Fix: route through `_broadcast_pos_change`. Locked in by `tests/integration/test_weapon_proc_pers.py::TestWeaponProcBroadcastPers::test_fight_010_vampiric_broadcast_uses_pers_for_invisible_victim`.
+
+## [2.8.61]
+
+### Fixed
+- **`FIGHT-009` — WEAPON_POISON TO_ROOM broadcast routes `$n` through PERS** (`src/fight.c:614-615`): ROM's `act("$n is poisoned by the venom on $p.", victim, wield, NULL, TO_ROOM)` substitutes `$n` per-listener through PERS. Python previously baked `victim.name` into a fixed `_broadcast_room` string. Fix: route through `_broadcast_pos_change`, now extended to accept `**extra` template kwargs (`{weapon}` for ROM `$p`). First commit in the FIGHT-009..013 weapon-proc PERS sweep. Locked in by `tests/integration/test_weapon_proc_pers.py::TestWeaponProcBroadcastPers::test_fight_009_poison_broadcast_uses_pers_for_invisible_victim`.
+
+## [2.8.60]
+
+### Fixed
+- **`PROMPT-CMD-005` legacy-test hygiene**: two pre-existing legacy assertions missed in 2.8.54's PROMPT-CMD-005 rollout were asserting the pre-fix stored value (no trailing space). Updated `tests/integration/test_config_commands.py::test_prompt_custom` and `tests/test_player_auto_settings.py::TestPrompt::test_prompt_set_custom` to ROM-true stored values with trailing space, per AGENTS.md "ROM is the source of truth" rule. No production code change.
+
+## [2.8.59]
+
+### Fixed
+- **`FIGHT-008` — POS_DEAD TO_CHAR self-message red colour + blank-line spacing** (`src/fight.c:861`): ROM's `send_to_char("{RYou have been KILLED!!{x\n\r\n\r", victim)` wraps the death notice in red colour codes and appends two `\n\r` pairs so a blank line follows the message. Python's `mud/combat/engine.py:_position_change_message` previously returned the bare `"You have been KILLED!!"` — no colour, no extra spacing. Fix: return `"{RYou have been KILLED!!{x\n"`. The protocol layer in `mud/net/protocol.py:send_to_char` auto-appends one `\r\n` to every message, so the embedded trailing `\n` plus the auto-append produces the same visual blank-line spacing ROM gets from its two `\n\r` pairs. Closes the FIGHT-004..008 position-change-broadcast sweep. Locked in by `tests/integration/test_invisibility_combat.py::TestPositionChangeBroadcastPers::test_fight_008_pos_dead_self_message_wraps_red_and_appends_blank_line`.
+
+### Notes
+
+- Closes the 2026-05-23 `fight.c` PERS sweep on the position-change broadcast surface (FIGHT-004..008 all ✅ FIXED in `docs/parity/FIGHT_C_AUDIT.md`).
+- Weapon-proc broadcasts at `mud/combat/engine.py:1496/1510/1531/1541/1551` have analogous PERS gaps but are deferred to a follow-up cluster (FIGHT-009..013 reserved).
+
+## [2.8.58]
+
+### Fixed
+- **`FIGHT-007` — POS_DEAD TO_ROOM broadcast (PERS + red colour + two-bang wording)** (`src/fight.c:860`): ROM's `act("{R$n is DEAD!!{x", victim, 0, 0, TO_ROOM)` had three divergences vs Python's previous `f"{victim.name} is DEAD!!!"` baked broadcast: (a) `$n` must route through PERS per-listener so invisible victims render as `"someone"` for observers without `DETECT_INVIS`; (b) missing ROM red colour codes `{R...{x` that the ANSI translation layer consumes on websocket send; (c) wording typo — three exclamation marks instead of ROM's two. Fix: `mud/combat/engine.py:_position_change_message` DEAD branch now uses `_broadcast_pos_change(victim, "{{R{name} is DEAD!!{{x")`. Legacy assertion in `tests/test_combat.py` (×1) updated to ROM-exact form. Locked in by `tests/integration/test_invisibility_combat.py::TestPositionChangeBroadcastPers::test_fight_007_pos_dead_broadcast_uses_pers_and_red_colour_and_two_bangs`.
+
+## [2.8.57]
+
+### Fixed
+- **`FIGHT-006` — POS_STUNNED TO_ROOM broadcast routes `$n` through PERS** (`src/fight.c:853-854`): ROM's `act("$n is stunned, but will probably recover.", victim, NULL, NULL, TO_ROOM)` substitutes `$n` per-listener through PERS. Python's STUNNED branch in `mud/combat/engine.py:_position_change_message` previously baked `victim.name` into a fixed `_broadcast_room` string. Fix: route through the `_broadcast_pos_change` helper. Locked in by `tests/integration/test_invisibility_combat.py::TestPositionChangeBroadcastPers::test_fight_006_pos_stunned_broadcast_uses_pers_for_invisible_victim`.
+
+## [2.8.56]
+
+### Fixed
+- **`FIGHT-005` — POS_INCAP TO_ROOM broadcast routes `$n` through PERS** (`src/fight.c:845-846`): ROM's `act("$n is incapacitated and will slowly die, if not aided.", victim, NULL, NULL, TO_ROOM)` substitutes `$n` per-listener through PERS. Python's INCAP branch in `mud/combat/engine.py:_position_change_message` previously baked `victim.name` into a fixed `_broadcast_room` string. Fix: route through the `_broadcast_pos_change` helper introduced in 2.8.55. Locked in by `tests/integration/test_invisibility_combat.py::TestPositionChangeBroadcastPers::test_fight_005_pos_incap_broadcast_uses_pers_for_invisible_victim`.
+
+## [2.8.55]
+
+### Fixed
+- **`FIGHT-004` — POS_MORTAL TO_ROOM broadcast routes `$n` through PERS** (`src/fight.c:837-838`, `src/handler.c:2618`): ROM's `act("$n is mortally wounded, and will die soon, if not aided.", victim, NULL, NULL, TO_ROOM)` renders `$n` per-listener through `PERS(victim, looker)`, so an invisible victim shows as `"someone"` to room observers without `DETECT_INVIS`. Python previously baked `victim.name` into a single fixed broadcast string via `_broadcast_room`, leaking the victim's identity to every recipient. Fix: new `_broadcast_pos_change` helper in `mud/combat/engine.py` iterates `room.people`, calls `mud.world.vision.pers(victim, listener)` per recipient, and dispatches through the same fire-and-forget websocket path as `broadcast_room`. Same channel-arc PERS fix pattern (SAY-002/EMOTE-001/TELL-003/SHOUT-003/YELL-001) applied to combat death messaging. Locked in by `tests/integration/test_invisibility_combat.py::TestPositionChangeBroadcastPers::test_fight_004_pos_mortal_broadcast_uses_pers_for_invisible_victim`.
+
+### Notes
+
+- First gap closed in the 2026-05-23 `fight.c` PERS sweep. FIGHT-005..008 (POS_INCAP / POS_STUNNED / POS_DEAD TO_ROOM and POS_DEAD TO_CHAR) stable-IDed in `docs/parity/FIGHT_C_AUDIT.md` and will land in follow-up commits using the new `_broadcast_pos_change` helper.
+
+## [2.8.54]
+
+### Fixed
+- **`PROMPT-CMD-004` — `do_prompt` truncates custom template to 50 chars** (`src/act_info.c:943-944`): ROM caps the raw `argument` at 50 chars (`if (strlen(argument) > 50) argument[50] = '\0';`) before `strcpy`/`smash_tilde`/suffix-append. Python previously stored the full untruncated string. Fix: `mud/commands/auto_settings.py:do_prompt` slices `arg = arg[:50]` before smash_tilde. Locked in by `tests/integration/test_prompt_cmd_parity.py::test_prompt_cmd_004_truncates_template_to_50_chars`.
+- **`PROMPT-CMD-005` — `do_prompt` appends trailing space unless template ends in `%c`** (`src/act_info.c:946-947`): ROM appends a literal space to every custom prompt template unless the (already smash_tilded) buf already ends with `%c` — `if (str_suffix("%c", buf)) strcat(buf, " ")`, and `str_suffix` (src/db.c:3784) returns TRUE when `"%c"` is *not* a suffix of buf. So prompts that don't end in a colour-code escape gain a trailing space; prompts that do (e.g. ANSI colour close) are left alone. Python previously skipped this normalization, so `prompt TAG>` stored `"TAG>"` instead of ROM's `"TAG> "`. Fix: `mud/commands/auto_settings.py:do_prompt` now applies `arg = arg + " "` when not `arg.endswith("%c")`, after smash_tilde and the 50-char truncation. Legacy assertions in `tests/test_player_prompt.py` (×4) and `tests/integration/test_prompt_rom_parity.py` (×1) updated to ROM-exact stored values. Locked in by `tests/integration/test_prompt_cmd_parity.py::test_prompt_cmd_005_appends_trailing_space_unless_pct_c_suffix`.
+
+### Notes
+
+- Closes the `do_prompt` warm-up shelf — PROMPT-CMD-001..005 now all ✅ FIXED in `docs/parity/ACT_INFO_C_AUDIT.md`.
+- PROMPT-CMD-004 and PROMPT-CMD-005 committed jointly because the legacy test assertions in `tests/test_player_prompt.py` couple both fixes; separate commits would assert transient half-states.
+
+## [2.8.53]
+
+### Changed
+- **`TELL-006` — closed as ✅ ANALYZED-INERT** (`src/act_comm.c:893,926,937`): ROM's `buf[0] = UPPER(buf[0])` on the buffered tell strings is provably a no-op in ROM C itself — the formatted string always begins with `'{'` (colour code `{k`), and `UPPER('{') == '{'` since `{` is not lowercase. Unlike TELL-004 (behaviorally masked by a separate code path), this transformation has no reachable behavior to mirror and no failing test can be written. Doc-only close; no code change. `docs/parity/ACT_COMM_C_AUDIT.md` row flipped from 🔄 OPEN to ✅ ANALYZED. With this, the entire `do_tell` gap row (TELL-001..006) is closed.
+
+## [2.8.52]
+
+### Fixed
+- **`YELL-001` — `do_yell` TO_VICT `$n` routes through PERS for invisible-yeller protection** (`src/act_comm.c:1059`, `src/handler.c:2618`): ROM's `act("$n yells '$t'", ..., TO_VICT)` substitutes `$n` per-listener through `PERS(ch, victim)`. So an invisible yeller renders as `"someone"` to listeners without `DETECT_INVIS`. Python's `do_yell` already iterates per-listener (area-wide loop over `character_registry` with area filter) but hardcoded `char.name` into the rendered string. Fix: `mud/commands/communication.py:do_yell` now substitutes `mud.world.vision.pers(char, victim)` for the yeller's name inside its existing per-listener loop. Closes the channel-message arc (say / tell / emote / shout / yell all now route through PERS). Locked in by `tests/integration/test_shout_yell_parity.py::test_yell_001_invisible_yeller_renders_as_someone_to_listener`.
+
+### Notes
+
+- Completes the 2026-05-22/23 act_comm.c channel re-audit arc: `do_say` (SAY-001..004), `do_emote` (EMOTE-001/002), `do_tell` (TELL-001..005), `do_shout` (SHOUT-001..003), `do_yell` (YELL-001). All five channel commands now route `$n` through `pers()` and emit ROM-exact wording / colour codes. PMOTE-001 remains as the only open ROM `do_pmote` greenfield port; TELL-006 deferred (MINOR cosmetic).
+- Full suite at `4629 passed, 4 skipped` (+1 vs 2.8.51; zero regressions).
+
+## [2.8.51]
+
+### Fixed
+- **`SHOUT-003` — `do_shout` TO_VICT `$n` routes through PERS for invisible-shouter protection** (`src/act_comm.c:836`, `src/handler.c:2618`): ROM's `act("$n shouts '$t'", ...)` substitutes `$n` per-listener through `PERS(ch, victim)`. So an invisible shouter renders as `"someone"` to listeners without `DETECT_INVIS`. Python previously broadcast one fixed message string via `broadcast_global(message, channel="shout", exclude=char, should_send=_should_receive)`, leaking the invisible shouter's identity to every recipient. Fix: replaced `broadcast_global` in `mud/commands/communication.py:do_shout` with a per-listener loop over `character_registry` (mirroring ROM's `descriptor_list` iteration at `src/act_comm.c:825-838`) that filters by `SHOUTSOFF` / `QUIET` / muted_channels and renders `mud.world.vision.pers(char, victim)` per recipient. Locked in by `tests/integration/test_shout_yell_parity.py::test_shout_003_invisible_shouter_renders_as_someone_to_listener`.
+
+### Notes
+
+- Third of three `do_shout` gaps from the 2026-05-23 re-audit. `do_shout` is now complete. YELL-001 (same PERS pattern in `do_yell`) remains open and is the last item in the channel-message arc.
+- Full suite at `4628 passed, 4 skipped, 1 deselected` (+1 vs 2.8.50; zero regressions despite the per-listener broadcast refactor).
+
+## [2.8.50]
+
+### Fixed
+- **`SHOUT-002` — `do_shout` TO_VICT wording drops the comma** (`src/act_comm.c:836`): ROM emits `act("$n shouts '$t'", ...)` — no comma between `shouts` and the open quote. Python broadcast `"{char.name} shouts, '{cleaned}'"` with an extra comma. Fix: drop the comma in the broadcast message string. Legacy assertions in `tests/test_communication.py` (×2) updated. Locked in by `tests/integration/test_shout_yell_parity.py::test_shout_002_to_vict_wording_drops_comma`.
+
+### Notes
+
+- Second of three `do_shout` gaps. SHOUT-003 (PERS for shouter) and YELL-001 (PERS for yeller) remain open.
+- Full suite at `4627 passed, 4 skipped, 2 deselected` (+1 vs 2.8.49; zero regressions).
+
+## [2.8.49]
+
+### Fixed
+- **`SHOUT-001` — `do_shout` TO_CHAR wording drops the comma** (`src/act_comm.c:824`): ROM emits `act("You shout '$T'", ...)` — no comma between `shout` and the open quote. Python previously returned `"You shout, '{cleaned}'"` with an extra comma. Fix: drop the comma in `mud/commands/communication.py:do_shout` return. Legacy assertions in `tests/test_communication.py` (×2) were baked to the buggy comma form and have been updated to the ROM-exact wording. Surfaced by a 2026-05-23 `do_shout` re-audit that found the prior "100% VERIFIED" claim incorrect (same audit-doc inflation that hit `do_say` / `do_emote` / `do_tell` this run). Locked in by `tests/integration/test_shout_yell_parity.py::test_shout_001_to_char_wording_drops_comma`.
+
+### Notes
+
+- First of three `do_shout` gaps from the 2026-05-23 re-audit. SHOUT-002 (TO_VICT wording), SHOUT-003 (PERS for shouter), plus YELL-001 (PERS for yeller — `do_yell` already gets wording right) all stable-IDed and queued.
+- Full suite at `4626 passed, 4 skipped, 3 deselected` (+1 vs 2.8.48; zero regressions).
+
+## [2.8.48]
+
+### Fixed
+- **`TELL-005` — `do_tell` wraps both lines with ROM `{k...{K...{k...{x` charcoal colour codes** (`src/act_comm.c:941-942`): ROM frames the tell channel in charcoal/black (`{k`), switches to bright/highlighted-charcoal (`{K`) for the message body, returns to `{k` for the closing quote, and resets with `{x`. The ANSI translation layer at `mud/net/ansi.py` consumes those tokens on websocket send. Python previously emitted no codes, so the tell channel rendered in the player's default terminal colour and the framing-vs-body contrast ROM relies on was lost. Fix: `mud/commands/communication.py:do_tell` return and `_handle_buffered_tell` formatted string now wrap with the ROM colour sequence. Legacy assertions in `tests/test_communication.py` (×5) and `tests/integration/test_communication_enhancement.py` were baked to the codeless plain-text form (per AGENTS.md — tests asserting Python behavior contradicting ROM are bugs in the **test**, not the implementation) and have been updated to the ROM-exact wording. Locked in by `tests/integration/test_tell_parity.py::test_tell_005_to_char_wraps_rom_color_codes` and `::test_tell_005_to_vict_wraps_rom_color_codes`.
+
+### Notes
+
+- Fifth of six TELL-NNN gaps from the 2026-05-22 `do_tell` re-audit. The audit's main cluster (TELL-001/002/003/004/005) is now closed. TELL-006 (uppercase first char of buffered tells — MINOR cosmetic) remains intentionally deferred and stable-IDed in `docs/parity/ACT_COMM_C_AUDIT.md`.
+- Full suite at `4625 passed, 4 skipped` (+2 vs 2.8.47 baseline 4623/4; zero regressions).
+
+## [2.8.47]
+
+### Fixed
+- **`TELL-004` — `do_tell` TO_CHAR `$N` routes through PERS for code-structure parity** (`src/act_comm.c:941`, `src/handler.c:2618`): ROM's `act("You tell $N '$t'", ...)` substitutes `$N` (capital) through `PERS(victim, ch)`. In practice this is masked by `get_char_world`'s `can_see` filtering during name lookup — an invisible target returns `"They aren't here."` before PERS would ever evaluate (same in ROM and Python). Fix is defensive code-parity: `mud/commands/communication.py:do_tell` return now substitutes `mud.world.vision.pers(target, char)` for the target name so the macro structure matches ROM regardless of future lookup-path changes. Test: `tests/integration/test_tell_parity.py::test_tell_004_to_char_uses_pers_for_target_name` pins the visible-target code path; the unreachable `"someone"` branch is exercised by sibling PERS tests (SAY-002, EMOTE-001, TELL-003).
+
+### Notes
+
+- Fourth of six TELL-NNN gaps from the 2026-05-22 `do_tell` re-audit. TELL-005 (charcoal colour codes) remains open. TELL-006 deferred (MINOR).
+- Full suite at `4623 passed, 4 skipped, 2 deselected` (+1 vs 2.8.46; zero regressions).
+
+## [2.8.46]
+
+### Fixed
+- **`TELL-003` — `do_tell` TO_VICT `$n` routes through PERS for invisible-sender protection** (`src/act_comm.c:942`, `src/handler.c:2618`): ROM's `act_new("$n tells you '$t'", ...)` substitutes `$n` per-target through `PERS(ch, victim)`, so an invisible sender renders as `"someone"` to a target without `DETECT_INVIS`. Python's `_handle_buffered_tell` hardcoded `sender.name`, leaking the invisible sender's identity (same pattern as pre-fix `do_say` / `do_emote`). Fix: substitute `mud.world.vision.pers(sender, target)` for the sender's name in the formatted string. Also applies to the buffered tell paths (linkdead / AFK / note-writing) since ROM's `sprintf(..., PERS(ch, victim), ...)` at `src/act_comm.c:891`/`924`/`935` wraps the same PERS call. Locked in by `tests/integration/test_tell_parity.py::test_tell_003_invisible_sender_renders_as_someone_to_target`.
+
+### Notes
+
+- Third of six TELL-NNN gaps from the 2026-05-22 `do_tell` re-audit. TELL-004 (PERS for target to sender) and TELL-005 (charcoal colour codes) remain open. TELL-006 deferred (MINOR).
+- Full suite at `4622 passed, 4 skipped, 3 deselected` (+1 vs 2.8.45; zero regressions).
+
+## [2.8.45]
+
+### Fixed
+- **`TELL-002` — `do_tell` TO_VICT wording drops the comma** (`src/act_comm.c:942`): ROM emits `act_new("$n tells you '$t'", ...)` — no comma between `you` and the open quote. Python's `_handle_buffered_tell` previously formatted `"{sender.name} tells you, '{message}'"` with an extra comma. Fix: drop the comma in `mud/commands/communication.py:_handle_buffered_tell` formatted string. Legacy assertions in `tests/test_communication.py` (×4) were baked to the buggy comma form and have been updated to the ROM-exact wording. Locked in by `tests/integration/test_tell_parity.py::test_tell_002_to_vict_wording_drops_comma`.
+
+### Notes
+
+- Second of six TELL-NNN gaps from the 2026-05-22 `do_tell` re-audit. TELL-003 (PERS for sender to victim), TELL-004 (PERS for target to sender), TELL-005 (charcoal colour codes) remain open. TELL-006 deferred (MINOR).
+- Full suite at `4621 passed, 4 skipped, 4 deselected` (+1 vs 2.8.44; zero regressions).
+
+## [2.8.44]
+
+### Fixed
+- **`TELL-001` — `do_tell` TO_CHAR wording drops the comma** (`src/act_comm.c:941`): ROM emits `act("You tell $N '$t'", ...)` — no comma between target name and the open quote. Python previously returned `"You tell {target.name}, '{message}'"` with an extra comma. Fix: drop the comma in `mud/commands/communication.py:do_tell` return. Legacy assertions in `tests/test_communication.py` (×2) and `tests/integration/test_communication_enhancement.py` were baked to the buggy comma form and have been updated to the ROM-exact wording. Surfaced by a 2026-05-22 `do_tell` re-audit that found the prior "100% VERIFIED" claim incorrect (same audit-doc inflation that hit `do_say` and `do_emote` this session). Locked in by `tests/integration/test_tell_parity.py::test_tell_001_to_char_wording_drops_comma`.
+
+### Notes
+
+- First of six TELL-NNN gaps from the 2026-05-22 `do_tell` re-audit. TELL-002 (TO_VICT wording comma), TELL-003 (PERS for sender to victim), TELL-004 (PERS for target to sender), TELL-005 (charcoal colour codes) remain open and will close in sequence. TELL-006 (uppercase first char of buffered tells — MINOR cosmetic) is intentionally deferred.
+- Full suite at `4620 passed, 4 skipped, 5 deselected` (+1 vs 2.8.43 baseline 4619/4; deselected = the still-open TELL gaps; zero regressions).
+
+## [2.8.43]
+
+### Fixed
+- **`EMOTE-002` — `do_emote` TO_CHAR `$n` renders as "You", not the actor's own name** (`src/act_comm.c:1092`): ROM's `act("$n $T", ..., TO_CHAR)` substitutes `$n` to `"You"` on the self branch — the actor reads `"You smiles happily"`, not `"Alice smiles happily"`. Python previously returned `f"{char.name} {args}"` so the actor saw their own name as the emote subject. Fix: `mud/commands/communication.py:do_emote` now returns `f"You {args}"`. Legacy `tests/integration/test_communication_enhancement.py::test_emote_broadcasts_custom_action` was baked to the buggy form (per AGENTS.md — tests asserting Python behavior contradicting ROM are bugs in the **test**, not the implementation) and has been updated to the ROM-exact wording. Locked in by `tests/integration/test_emote_parity.py::test_emote_002_self_message_renders_you_not_actor_name`.
+
+### Notes
+
+- Second of three gaps from the 2026-05-22 `do_emote` re-audit. PMOTE-001 (`do_pmote` not implemented in Python) remains open as a missing-function gap, tracked for a future session.
+- Full suite at `4619 passed, 4 skipped` (+1 vs 2.8.42 baseline 4618/4; zero regressions).
+
+## [2.8.42]
+
+### Fixed
+- **`EMOTE-001` — `do_emote` PERS substitution: invisible emoter renders as `"someone"`** (`src/act_comm.c:1091`, `src/handler.c:2618-2664`): ROM's `act("$n $T", ..., TO_ROOM)` substitutes `$n` per-listener through `PERS()` so an invisible emoter renders as `"someone"` to listeners without `DETECT_INVIS`. Python previously hardcoded `char.name` into one TO_ROOM message and broadcast it to everyone in the room, leaking the invisible emoter's identity exactly like the pre-fix `do_say` (SAY-002). Fix: refactored `mud/commands/communication.py:do_emote` to render TO_ROOM per-listener using `mud.world.vision.pers(char, listener)` (the helper added in 2.8.41 for SAY-002). Tests: `tests/integration/test_emote_parity.py::test_emote_001_invisible_emoter_renders_as_someone_to_unaided_listener` and `::test_emote_001_visible_emoter_renders_real_name_to_listener`.
+
+### Notes
+
+- First of three gaps from the 2026-05-22 `do_emote` re-audit. EMOTE-002 (TO_CHAR `$n` → "You" instead of actor's own name) and PMOTE-001 (`do_pmote` not implemented in Python) remain open.
+- Full suite at `4618 passed, 4 skipped` (+2 vs 2.8.41 baseline 4616/4; zero regressions despite the per-listener broadcast refactor).
+
+## [2.8.41]
+
+### Added
+- **`mud.world.vision.pers(target, observer)`** mirroring ROM's `PERS()` macro (`src/comm.c` via `#define` and `src/handler.c:2618-2664 can_see`). Returns `"someone"` if `observer` cannot see `target`, otherwise the target's NPC `short_descr` or PC `name`. No aura prefixes (that's `act_info show_char_to_char_0`) and no self-handling (callers route TO_CHAR vs TO_ROOM separately).
+
+### Fixed
+- **`SAY-002` — `do_say` PERS substitution: invisible/hidden speakers appear as `"someone"`** (`src/act_comm.c:776`, `src/handler.c:2618-2664`): ROM's `act()` substitutes `$n` through `PERS(ch, looker)`, which gates on `can_see(looker, ch)`. So when an invisible speaker says something, listeners without `DETECT_INVIS` see `"someone says '<msg>'"`, not the speaker's real name. Python previously built one TO_ROOM message string with `char.name` hardcoded and broadcast it to everyone in the room, leaking invisible/hidden PC identities. Fix: added `pers()` helper in `mud/world/vision.py` (uses existing `can_see_character`), then refactored `do_say` to render the TO_ROOM string per-listener with `pers(speaker, listener)` substituted for `$n`. Closes the four-gap `do_say` re-audit cluster opened on 2026-05-22. Tests: `tests/integration/test_say_parity.py::test_say_002_invisible_speaker_renders_as_someone_to_unaided_listener` and `::test_say_002_invisible_speaker_seen_by_detect_invis_listener`.
+
+### Notes
+
+- Fourth and final SAY-NNN gap from the 2026-05-22 `act_comm.c` re-audit. `do_say` is now ✅ 100% re-audited.
+- Analogous CRITICAL gaps almost certainly exist in `do_tell` / `do_shout` / `do_yell` / `do_emote` / `do_pose` / `do_pmote` — those audits are now unblocked (the helper exists). Tracked as next-session candidates in `docs/sessions/SESSION_STATUS.md`.
+- Full suite at `4616 passed, 4 skipped` (+2 vs 2.8.40 baseline 4614/4; zero regressions despite the per-listener broadcast refactor).
+
+## [2.8.40]
+
+### Fixed
+- **`SAY-003` — `do_say` wraps output with ROM colour codes (`{6...{7$T{6'{x`)** (`src/act_comm.c:776-777`): ROM frames the say channel in cyan/green (`{6`), switches to white (`{7`) for the message body, returns to `{6` for the closing quote, and resets with `{x`. The ANSI translation layer at `mud/net/ansi.py` consumes those tokens on websocket send. Python previously emitted no codes, so the say channel rendered in the player's default terminal colour and the framing-vs-body contrast ROM relies on was lost. Fix: `mud/commands/communication.py:do_say` now wraps both TO_CHAR and TO_ROOM strings with the ROM colour sequence. Legacy assertions in `tests/test_commands.py`, `tests/test_command_abbrev.py`, and `tests/integration/test_communication_enhancement.py` were baked to the codeless plain-text form (per AGENTS.md, tests asserting Python behavior contradicting ROM are bugs in the **test**, not the implementation) and have been updated. Locked in by `tests/integration/test_say_parity.py::test_say_003_to_char_wraps_rom_color_codes` and `::test_say_003_to_room_wraps_rom_color_codes`.
+
+### Notes
+
+- Third of four SAY-NNN gaps from the 2026-05-22 `act_comm.c` re-audit. SAY-002 (`$n` PERS for invisible speakers) remains open — requires building a Python `pers()` helper that mirrors ROM's `PERS()` macro.
+- Full suite at `4614 passed, 4 skipped` (+2 vs 2.8.39 baseline 4612/4; zero regressions).
+
+## [2.8.39]
+
+### Fixed
+- **`SAY-004` — `do_say` delivers TO_ROOM broadcast exactly once (INV-001 SINGLE-DELIVERY)** (`src/act_comm.c:776`): ROM's `act(..., TO_ROOM)` iterates `ch->in_room->people` and delivers to each target exactly once. Python called BOTH `char.room.broadcast(message, exclude=char)` AND `broadcast_room(char.room, message, exclude=char)`. The two helpers do identical work — iterate `room.people`, fire-and-forget websocket send, append to `char.messages` — so every `say` was delivered twice. Players received "<name> says 'hi'" twice; transcripts and message queues both diverged. Fix: dropped the redundant `broadcast_room` call in `mud/commands/communication.py:do_say`, keeping only `char.room.broadcast`. Locked in by `tests/integration/test_say_parity.py::test_say_004_listener_receives_broadcast_exactly_once`.
+
+### Notes
+
+- Second of four SAY-NNN gaps from the 2026-05-22 `act_comm.c` re-audit. SAY-002 (`$n` PERS for invisible speakers) and SAY-003 (ROM colour codes) remain open.
+- Full suite at `4612 passed, 4 skipped` (+1 vs 2.8.38 baseline 4611/4; zero regressions).
+
+## [2.8.38]
+
+### Fixed
+- **`SAY-001` — `do_say` wording matches ROM (`says '$T'` / `You say '$T'` — no comma)** (`src/act_comm.c:776-777`): ROM emits `act("$n says '$T'", ...)` and `act("You say '$T'", ...)`. Python emitted `"says, '..'"` and `"You say, '..'"` with a comma the player did not type. Surfaced by a 2026-05-22 re-audit of `act_comm.c` that found the previous "100% VERIFIED" claim was incorrect. Fix: `mud/commands/communication.py:do_say` now produces the ROM-exact wording. Legacy `tests/test_commands.py`, `tests/test_command_abbrev.py`, `tests/integration/test_communication_enhancement.py` assertions were baked to the buggy comma form (per AGENTS.md — tests that assert Python behavior contradicting ROM are bugs in the **test**, not the implementation) and have been updated to the ROM-exact wording. Locked in by `tests/integration/test_say_parity.py::test_say_001_room_broadcast_drops_comma` and `test_say_001_to_char_drops_comma`.
+
+### Notes
+
+- Three additional `do_say` gaps surfaced by the re-audit are stable-IDed in `docs/parity/ACT_COMM_C_AUDIT.md` for sequential closure: SAY-002 (`$n` PERS substitution for invisible speakers), SAY-003 (ROM colour codes), SAY-004 (double-delivery via `room.broadcast` + `broadcast_room` — INV-001 SINGLE-DELIVERY violation).
+- Full suite at `4611 passed, 4 skipped` (+2 vs 2.8.37 baseline 4609/4; zero regressions).
+
+## [2.8.37]
+
+### Fixed
+- **`PROMPT-CMD-003` — `do_prompt` runs `smash_tilde` on custom template before storing** (`src/act_info.c:945`, `src/db.c:3663-3672`): ROM calls `smash_tilde(buf)` before `ch->prompt = str_dup(buf)`, replacing every `~` with `-` so the stored template cannot corrupt the player file on save (ROM treats `~` as the end-of-string marker on disk). Python's `do_prompt` previously stored the raw argument including any tildes the player typed. Fix: `mud/commands/auto_settings.py:do_prompt` now calls `mud.utils.text.smash_tilde` on the custom-template branch before assigning `char.prompt`. Locked in by `tests/integration/test_prompt_cmd_parity.py::test_prompt_cmd_003_smash_tilde_on_custom_template`.
+
+### Notes
+
+- Closes the PROMPT-CMD cluster opened by 2.8.35's NANNY-SAVELOAD-002 probe (PROMPT-CMD-001/002 landed in 2.8.36).
+- Two ROM-divergence corners remain on `do_prompt` but are now stable-IDed for future hardening: PROMPT-CMD-004 (50-char truncation), PROMPT-CMD-005 (`%c`-suffix → append trailing space). Both are corner cases — no behavioral impact for typical players — and tracked in `docs/parity/ACT_INFO_C_AUDIT.md`.
+- Full suite at `4609 passed, 4 skipped` (+1 vs 2.8.36 baseline 4608/4; zero regressions).
+
+## [2.8.36]
+
+### Fixed
+- **`PROMPT-CMD-001` — `do_prompt` preserves trailing whitespace in custom templates** (`src/act_info.c:944`): the dispatcher (`mud/commands/dispatcher.py:process_command`) used `core = trimmed.rstrip()` which stripped visible trailing whitespace from every command before reaching its handler. ROM `src/interp.c:interpret` only strips line-ending whitespace, leaving visible trailing whitespace intact. Fix: dispatcher now uses `core = trimmed.rstrip("\r\n")` and computes log-line trailing whitespace from `trimmed.rstrip()` separately so the existing per-command log convention (`tests/test_logging_admin.py::test_logging_logs_alias_expansion`) is preserved. Also removed `arg = args.strip()` in `mud/commands/auto_settings.py:do_prompt` so the handler now uses the raw arg (matching ROM's raw `argument` usage at `src/act_info.c:944`). Now `prompt MYTAG> ` correctly stores `"MYTAG> "` and `bust_a_prompt` renders the trailing space.
+- **`PROMPT-CMD-002` — `do_prompt` success reply echoes the stored template** (`src/act_info.c:953-954`): Python previously emitted the truncated `"Prompt set."`; ROM emits `"Prompt set to %s\n\r"` with the stored template. Fixed both the `"all"` branch and the custom-template branch in `mud/commands/auto_settings.py:do_prompt`. Existing helper tests (`tests/test_player_prompt.py`) and the live-websocket round-trip in `tests/integration/test_nanny_saveload_runtime_path.py` updated to the ROM-exact wording.
+
+### Notes
+
+- New enforcement tests in `tests/integration/test_prompt_cmd_parity.py` lock in both gaps on the live websocket path.
+- Full suite at `4608 passed, 4 skipped` (+2 vs 2.8.35 baseline 4606/4; zero regressions). Surfaced and fixed in a single session after being flagged in 2.8.35's NANNY-SAVELOAD-002 probe.
+
+## [2.8.35]
+
+### Changed
+- **`NANNY-SAVELOAD-001` — wimpy round-trips through reconnect** (`src/act_info.c:2800-2830`, `src/save.c:fwrite_char/fread_char`, `src/act_info.c:1548-1549`): added `tests/integration/test_nanny_saveload_runtime_path.py::test_nanny_saveload_001_wimpy_round_trips_through_reconnect` exercising `wimpy <n>` mid-session, then asserting the value comes back via `score` on the first command after reconnect. Verified the live runtime path; no implementation drift found.
+- **`NANNY-SAVELOAD-002` — custom prompt template round-trips through reconnect** (`src/act_info.c:919-955`, `src/comm.c:1420-1595`): added `tests/integration/test_nanny_saveload_runtime_path.py::test_nanny_saveload_002_prompt_template_round_trips_through_reconnect` exercising `prompt <template>` mid-session, then asserting the first in-game prompt after reconnect renders the saved template (no ROM-default `<Nhp Nm Nmv>` fallback). Verified the live runtime path; the persistence layer round-trip is intact. Probe also surfaced two **separate** `do_prompt` command-side parity gaps (not save→reload bugs): (a) the dispatcher strips trailing whitespace from `prompt <template>` args before `do_prompt` sees them, so a template with a trailing space loses it; (b) the success reply is the truncated `"Prompt set."` instead of ROM's `"Prompt set to <template>\n\r"` which echoes the stored template. Both captured as follow-ups in `docs/sessions/SESSION_STATUS.md`.
+- **`NANNY-SAVELOAD-003` — per-character aliases round-trip through reconnect** (`src/alias.c:102-220`, `src/save.c:fwrite_char/fread_char`): added `tests/integration/test_nanny_saveload_runtime_path.py::test_nanny_saveload_003_alias_round_trips_through_reconnect` exercising `alias k kill` mid-session, then asserting the alias listing on the first command after reconnect re-renders the saved entry with ROM listing format `"    k:  kill"`. Verified the live runtime path through the DB JSON column; no implementation drift found.
+
+### Notes
+
+- Full suite at `4606 passed, 4 skipped` (+3 vs 2.8.34 baseline 4603/4; zero regressions).
+- Plan Task 4 (`docs/superpowers/plans/2026-05-21-parity-trust-rebuild-reaudit.md`) — save → reload → retained-state runtime-path bullet — is now complete. Remaining trust-rebuild work shifts to Plan Task 5 (re-audit other high-risk user-visible command families).
+
+## [2.8.34]
+
+### Fixed
+- **`INV-009` REGISTRY-DISCONNECT-CLEANUP** (new cross-file invariant): `character_registry` was accumulating duplicate `Character` entries per player name in two places:
+  - **In-session promote-from-bare-row duplication.** The level=0 bare-row Character loaded at `_select_character` (during the nanny name/password phase) was appended to `character_registry` and never replaced. After creation finalised, a fresh level=1 Character was loaded via `load_character` and appended on top, leaving both in the registry. `next((c for c in character_registry if c.name == X), None)` could return the stale level=0 entry (e.g. `hit=20`) while the live session ran against the level=1 entry (`hit=100`).
+  - **Cross-session leak on clean disconnect.** The websocket and telnet disconnect `finally` blocks already saved the character, removed them from their room, and released the account — but never removed them from `character_registry`. Each disconnect-then-reconnect cycle added a new entry without removing the old one.
+- **Fix (`mud/account/account_manager.py:load_character`)**: drop any prior `character_registry` entry with the same name before appending the freshly-loaded one (dedup-by-name on append).
+- **Fix (`mud/net/connection.py` websocket + telnet disconnect cleanup)**: remove the Character from `character_registry` in the `if char: if not forced_disconnect:` block, matching the `save + char_from_room + release_account` quit semantics already present. Forced disconnects (descriptor takeover) skip removal — `_disconnect_session` transfers the live Character to the new descriptor.
+- **Test**: `tests/integration/test_inv009_registry_disconnect_cleanup.py::test_inv009_registry_has_single_entry_after_disconnect_and_reconnect` reproduces both cases on the live websocket path and asserts: (a) zero `Regista` entries after a clean disconnect, (b) exactly one `Regista` entry while a reconnected session is active.
+- **Tracker**: new INV-009 row in `docs/parity/CROSS_FILE_INVARIANTS_TRACKER.md`.
+
+### Notes
+
+- Full suite at `4603 passed, 4 skipped` (+1 vs 2.8.33 baseline 4602/4; zero regressions).
+- Discovered during NANNY-RECONNECT-003 debugging in 2.8.33 (see SESSION_SUMMARY_2026-05-22_NANNY_RECONNECT_RUNTIME_PATH_PARITY.md "follow-ups").
+
+## [2.8.33]
+
+### Changed
+- **`NANNY-RECONNECT-001` — `score` after reconnect transcript parity** (`src/act_info.c:1477-1507`, `src/nanny.c:760`): added a websocket-runtime-path regression in `tests/test_websocket_server.py::test_websocket_reconnect_score_matches_rom_act_info_lines` that asserts ROM-exact title (`"You are <name> the Apprentice of Magic, level 1, ..."`), race/sex/class line (`"Race: elf  Sex: male  Class: mage"`), and hit/mana/move at full after `reset_char` on login. Verified the live runtime path; no implementation drift found.
+- **`NANNY-RECONNECT-002` — `look` after reconnect matches live room registry** (`src/act_info.c:1037-1116`): added `tests/test_websocket_server.py::test_websocket_reconnect_look_matches_room_registry_not_cached_snapshot` which reconnects an elf mage, snapshots the live `character_registry` entry's `room.name` and description, then asserts the `look` transcript on the **first command after reconnect** contains both — guarding against stale pre-disconnect cached snapshots. Verified the live runtime path; no implementation drift found.
+- **`NANNY-RECONNECT-003` — first in-game prompt after reconnect reflects loaded resources** (`src/comm.c:1437-1443`, `src/nanny.c:760`): added `tests/test_websocket_server.py::test_websocket_reconnect_initial_prompt_reflects_loaded_resources` which captures the first post-MOTD prompt on the live websocket, then sends `score` on the same session and asserts the ROM `<Nhp Nm Nmv>` token values equal what `score` reports (self-consistent), and that `hit == max_hit` after `reset_char` on login. Verified the live runtime path; no implementation drift found.
+
+### Notes
+
+- Full suite at `4602 passed, 4 skipped` (+3 vs prior baseline 4599/4; zero regressions).
+- During NANNY-RECONNECT-003 debugging, a `character_registry` reconnect-duplication oddity surfaced (stale pre-disconnect Character returned by name lookup). Recorded as a follow-up in `docs/sessions/SESSION_STATUS.md` for a dedicated future slice.
+
+## [2.8.32]
+
+### Fixed
+- **`nanny.c` happy-path new-character creation prompts**: `mud/net/connection.py` now emits ROM-exact `"New character."`, `"Give me a password for <Name>:"`, `"Please retype password:"`, and ROM race/sex/class/weapon prompt wording on the live runtime path. Removed the non-ROM `"Creating new character 'X'."` line and the non-ROM stat-reroll and hometown prompts (ROM `nanny.c:476-478` derives `perm_stat` from `pc_race_table[race].stats[i]` and has no such prompts).
+- **`nanny.c` pre-login greeting parser**: the greeting no longer leaks the embedded MOTD block before `Name:`.
+- **`NANNY-RETRY-001` — race invalid wording** (`src/nanny.c:460-471`): `_prompt_for_race` now sends `"That is not a valid race."` (was `"That's"`); race listing prefix is `"The following races are available:\n\r  "` to match ROM telnet line endings.
+- **`NANNY-RETRY-002` — race "help" branch wording** (`src/nanny.c:444-453`): verified ROM-exact; regression test added.
+- **`NANNY-RETRY-003` — class invalid wording** (`src/nanny.c:538-539`): `_prompt_for_class` now sends `"That's not a class."` (was `"That's not a valid class."`) and switches the retry prompt to `"What IS your class? "`.
+- **`NANNY-RETRY-004` — customize Y/N retry** (`src/nanny.c:626-628`): `_prompt_customization_choice` now sends `"Please answer (Y/N)? "` on invalid entry (was the generic `"Please answer Y or N."`). `_prompt_yes_no` gained an optional `retry_message` kwarg so other Y/N callsites keep their existing wording.
+- **`NANNY-RETRY-005` / `NANNY-RETRY-006` — weapon prompt CRLF** (`src/nanny.c:611-624`, `:638-649`): the weapon-pick prompt and its invalid-entry retry now use ROM `\n\r` line endings before `"Your choice? "` (were bare `\n`).
+
+### Changed
+- **Stricter creation transcript coverage**: `tests/integration/test_nanny_login_parity.py` adds six transcript-parity tests (`NANNY-RETRY-001..006`); full suite now at `4599 passed, 4 skipped` (+6 from prior baseline 4593/4, zero regressions).
+
+## [2.8.31]
+
+### Fixed
+- **`nanny.c` trust-rebuild: `CON_READ_MOTD` now emits the ROM welcome line**: `mud/net/connection.py` now sends `Welcome to ROM 2.4.  Please don't feed the mobiles!` at the moment a normal login actually enters the game, matching the ROM branch instead of skipping straight to MOTD/help/look output.
+- **Forced reconnects no longer run the normal login tail**: duplicate-session takeover reconnects now skip the normal welcome/look/board path and only emit the ROM reconnect-side output, which restores parity with `check_reconnect(..., TRUE)` and keeps telnet reconnect sequencing stable.
+
+### Changed
+- **Stricter reconnect/login runtime coverage**: `tests/test_websocket_server.py` now asserts the ROM welcome line on first entry, and the full suite recertifies clean with the reconnect split in place (`4593 passed, 4 skipped`).
+
+## [2.8.30]
+
+### Fixed
+- **`save.c` trust-rebuild: reconnect now preserves the school outfit**: `mud/models/character.py` now restores equipped items from DB-canonical `equipment_state` without silently dropping them on load, and recomputes carry totals after inventory/equipment restore so the first `score` after reconnect matches the first `score` after character creation.
+- **`nanny.c` trust-rebuild: login now emits the ROM board summary**: `mud/net/connection.py` now mirrors the tail of `CON_READ_MOTD` by sending a blank line and `board` output after the initial room `look` on both connection paths.
+
+### Changed
+- **Stricter runtime-path persistence coverage**: `tests/test_websocket_server.py` now asserts both reconnect outfit persistence and post-login board summary output on the real WebSocket path, and `tests/integration/test_inv008_persistence_coherence.py` now locks equipment/carry-state save-load round trips directly.
+
+## [2.8.29]
+
+### Fixed
+- **`nanny.c` trust-rebuild: returning level-1 players no longer replay first-login outfit flow**: `mud/net/connection.py` now treats the persisted `newbie_help_seen` flag as the durable first-login marker, so reconnecting or relogging level-1 characters no longer get re-equipped by Mota or reclassified as brand-new.
+
+### Changed
+- **Stricter runtime-path login coverage**: `tests/test_websocket_server.py` now asserts that a real WebSocket reconnect does not replay the one-time school outfit path, and `tests/test_connection_motd.py` now locks the `_is_new_player()` helper contract directly.
+
+## [2.8.28]
+
+### Fixed
+- **`act_info.c` trust-rebuild: `do_look` dark-room and container-detail parity**: `mud/world/look.py` now mirrors ROM dark-room output more closely by appending raw visible character lines without a Python-only `Characters:` label, and `look in` now uses the ROM drink-container wording with liquid color plus the correct `CONT_CLOSED` bit for closed containers.
+- **`check_blind()` / `do_look` blind-gate parity**: `mud/rom_api.py` now rejects `AFF_BLIND` correctly, and `mud/world/look.py` now returns the ROM blind message `"You can't see a thing!"` instead of falling through to normal room output.
+
+### Changed
+- **Stricter `do_look` regression coverage**: `tests/integration/test_do_look_command.py` now includes exact ROM-style assertions for dark-room visible occupants, drink-container liquid wording, closed-container gating, and blindness; `tests/test_rom_api.py` now locks the blind helper contract directly.
+
+## [2.8.27]
+
+### Fixed
+- **`act_info.c` trust-rebuild: `do_look` autoexit parity**: `mud/world/look.py` no longer emits a manual exits line on every room look; exits now appear only through the ROM `PLR_AUTOEXIT` branch.
+- **`act_info.c` trust-rebuild: `do_look` room rendering parity**: room contents and visible occupants are now appended as raw lines instead of Python-only `Objects:` / `Characters:` labels, matching ROM `show_list_to_char` / `show_char_to_char` behavior more closely.
+
+### Changed
+- **Direct `do_look` regression coverage**: added ROM-exact room-look tests for autoexit gating and raw room content / occupant line formatting.
+
+
+## [2.8.26]
+
+### Fixed
+- **`act_info.c` trust-rebuild: `do_who` switched-session parity**: `mud/commands/info.py` now mirrors ROM `do_who` by preferring the switched descriptor `original` character for output and title rendering instead of always showing the shell character.
+
+### Changed
+- **Stricter `do_who` output coverage**: `tests/integration/test_do_who_command.py` now includes exact ROM-style output assertions for switched sessions and representative fully-flagged player rows.
+
+
+## [2.8.25]
+
+### Fixed
+- **`act_info.c` trust-rebuild: `do_equipment` slot order parity**: `mud/commands/inventory.py` now mirrors ROM `do_equipment` by iterating wear slots in numeric `MAX_WEAR` order instead of Python dict insertion order.
+- **`act_info.c` trust-rebuild: `do_where` private-room parity**: `mud/commands/info.py` mode 1 now applies the ROM room-owner/private-room gate on the live session path, so mortals no longer see descriptor-backed players in genuinely private rooms.
+
+### Changed
+- **Stricter `inventory` / `equipment` / `where` regressions**: added ROM-exact layout assertions for `do_inventory` and `do_equipment`, plus a real descriptor-path private-room regression for `do_where`.
+
+
+## [2.8.24]
+
+### Fixed
+- **`act_info.c` trust-rebuild: `whois` formatting parity**: `mud/commands/info_extended.py` now mirrors ROM descriptor-path `whois` behavior more closely, including race/class display, immortal rank labels, AFK/KILLER/THIEF flags via enums, and switched-descriptor `original` handling.
+- **`score` player-race mapping**: fixed the player-facing race-name helper so playable race id `0` renders as `human` instead of the base-race sentinel `unique`.
+
+### Changed
+- **Stricter `act_info.c` regressions**: `tests/test_player_info_commands.py` now replaces several weak smoke assertions with ROM-exact `score` and `whois` output checks.
+- **Trust-rebuild handoff updated**: `ACT_INFO_C_AUDIT.md` and the session docs now explicitly record that `act_info.c` is under observable-behavior revalidation rather than relying on legacy structural-audit completion claims.
+
+
+## [2.8.23]
+
+### Fixed
+- **`score` parity and load-session hydration**: fixed ROM-visible `score` output so loaded characters now refresh live session `logon`, render race/class/title correctly, and use ROM-style low-level AC wording instead of stale helper-path values.
+- **Runtime-path persistence regression coverage**: added a full WebSocket create → disconnect → reconnect test to prove completed character creation persists and the second login reaches `Password:` instead of restarting character creation.
+
+### Changed
+- **Verification language downgraded to match evidence**: README and session guidance now mark ROM parity as under active revalidation rather than fully certified, after a live `score` bug showed gaps in prior observable-behavior coverage.
+- **Trust-rebuild program documented**: added a re-audit plan and a differential-testing design spec covering ROM-exact output, runtime-path verification, and data parity for future audit closure work.
+
+
+## [2.8.22]
+
+### Fixed
+- **Area JSON boot/reset parity**: repaired the checked-in `data/areas/*.json` dataset and the ROM area loaders so world boot no longer emits the invalid O/P/D/G/E reset warning flood. This includes ROM-accurate room-door lock decoding, object `F` affect parsing, first-tilde string termination, and JSON room-sector preservation.
+
+### Changed
+- **Area dataset regeneration**: regenerated the checked-in area JSON files from their `.are` sources after the loader fixes, restoring missing object prototypes and corrected room topology across the shipped world data.
+- **Stale test cleanup after area repair**: updated area/scan/mobprog parity tests to stop depending on previously broken topology or ambient RNG state.
+- **Full-suite recertification**: the suite now reruns clean at `4567 passed, 4 skipped`.
+
 ## [2.8.21]
 
 ### Changed