Skip to content
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion qa/SCORECARD.md
Original file line number Diff line number Diff line change
Expand Up @@ -42,7 +42,7 @@
| **— gates-work read —** | 2026-05-29 | — | — | — | — | — | — | — | — | **★ The proactive-gate strategy is PROVEN: the new FATAL gates caught 2 real defects the noisy LLM lenses would have blurred.** Both fixed in 6e7c3b0 (update_character skills alias + duo PC-as-player brief). LESSON: the 16GB host CANNOT run 2 duos + 3 audit agents concurrently (memory pressure silenced swA's DM) — revalidation runs SEQUENTIALLY (1 heavy DM at a time) for clean, non-degraded scores. Revalidation (ow-cs3 sprint + 1 duo) in flight. |
| ow-cs3-133001 | 2026-05-29 | baldurs-gate | (combat-sprint, pre-seeded) | sonnet | 1 fight | GREEN | — | — | **3.7** | **★ Combat fidelity 3.0→3.7 (+0.7) — the e5d651f combat cues VALIDATED.** The combat-adherence audit found the 3.0 (ow-cs2) was scored at 04:15, ~8h BEFORE the cues shipped (12:32); re-measuring on current main climbs. Residual WARNs (still adherence, not missing docs): ranged-in-melee disadvantage + Parry reaction unused. New engine-side combat gaps queued (mech audit): Magic Missile auto-hit via cast_spell (combat.md nudge LANDED), surprise=Surprised condition, Mage Armor AC recompute, maneuver-rider saves. |
| ow-rv1-134258 | 2026-05-29 | baldurs-gate | Dal (canon) | sonnet | 9 | **RED** | 2.5 | 2.5 | 2.5 | POST-6e7c3b0. **no_rejected_tool_calls PASSED — the update_character skills fix VALIDATED in vivo** (no extra_forbidden this run). Still RED on 2 gates: (1) **player_in_party** — the brief fix worked HALFWAY (Dal is now kind="player" ✓, not companion) but load_canon_character(kind="player") defaults add_to_party=False → PC sat OUTSIDE party (only Astarion in). Caught LIVE via the preview repointed at the running session. → ENGINE INVARIANT fix: kind="player" ⇒ always in party (both creation paths, +test). (2) **dm_voices_characters** 0/9 quoted dialogue — a flat-DM sampling run (gate matches straight+curly quotes, so not a false-fail; swA/most runs PASS it). |
| ow-v103-reval | 2026-05-30 | baldurs-gate | Dal Lightspark (canon Harper evoker) | sonnet | 8 | **GREEN** | **4.1** | **4.1** | 3.2 | **★ POST-v1.0.3 / post-#162 REVALIDATION — ALL 20+ gates GREEN incl. the previously-failing `player_in_party`** (#162 kind="player"⇒in-party + canon-PC bind VALIDATED in vivo; ow-rv1's RED closed — PC seated via load_canon_character(kind="player"), the 2 create_character calls were NPCs, no Caelar regression). Clock day1→day3, ≥2 locations (Siltwharf Steps→Heapside Warrens), Director fired beat 5, real costs (failed History 14 v DC17, nat-20 Insight). Story 4.1 (short-3act, progressed; lever = final-act info density). Mech 4.1 (no hallucinated rolls; gaps: Mage Armor state, Imaskari conjunction clock). angrydm 3.2 = social-only (0 fights/saves/conditions) + **NEW [HIGH]: canon ability scores load flat 10/10/10 — Dal + 11 NPCs (Withers excepted); caster PC broken at INT 10, no engine warning → fixed in this PR.** |
| ow-v103-reval | 2026-05-30 | baldurs-gate | Dal Lightspark (canon Harper evoker) | sonnet | 8 | **GREEN** | **4.1** | **4.1** | 3.2 | **★ POST-v1.0.3 / post-#162 REVALIDATION — ALL 20+ gates GREEN incl. the previously-failing `player_in_party`** (#162 kind="player"⇒in-party + canon-PC bind VALIDATED in vivo; ow-rv1's RED closed — PC seated via load_canon_character(kind="player"), the 2 create_character calls were NPCs, no Caelar regression). Clock day1→day3, ≥2 locations (Siltwharf Steps→Heapside Warrens), Director fired beat 5, real costs (failed History 14 v DC17, nat-20 Insight). Story 4.1 (short-3act, progressed; lever = final-act info density). Mech 4.1 (no hallucinated rolls; gaps: Mage Armor state, Imaskari conjunction clock). angrydm 3.2 = social-only (0 fights/saves/conditions) + **NEW [HIGH]: canon ability scores load flat 10/10/10 — Dal + 11 NPCs (Withers excepted); caster PC broken at INT 10, no engine warning → fixed in this PR.** **⚠ [SUPERSEDED — #305]: Dal Lightspark is canonically DEAD, so this run's STORY/MECH scores (4.1/4.1/3.2) are SUSPECT as a clean canon-PC baseline (ran portrait-less on a dead PC). The engine-plumbing validations still hold (the #162 kind="player"⇒in-party invariant, clock/travel/world-peopled — char-agnostic). The clean-canon-PC revalidation duo is OWED on a LIVING canon PC via the #329 living fixture — do NOT cite 4.1/4.1 as the canon-PC baseline until then.** |

## Change log (run ↔ engine delta)
- **#137** wandering-encounter system (travel/camp combat risk) — merged.
Expand Down
Loading