mirror of
https://github.com/prosolis/gogobee.git
synced 2026-07-15 08:32:41 +00:00
f2c2d774d4
The post-J1 sweep had the casters clustered at 19–22% L12 clear, vs martials at 70–80%. A per-round trace across 240 boss-room fights showed why: autoResolveCombat dispatched !attack only — zero spell_casts, zero mid-fight consumable uses across every caster class. The entire "caster cliff" was the sim measuring a strawman where casters couldn't cast. J2a teaches the sim's autoResolveCombat to mirror a competent prod player: heal at low HP if an inventory consumable is available, otherwise cast the highest-EV damage spell (slot or cantrip), otherwise swing. BuildCharacter now seeds the known-spell list via ensureSpellsForCharacter so the synthetic spellbook is populated. A -trace flag on the cmd attaches the raw CombatEvent stream to the last combat of each run for post-hoc diagnostics. A first re-baseline (n=100, all 10 classes) showed Ranger regressed -35.8pp — the picker was burning L3 slots on lightning_arrow when Ranger's weapon chassis (Hunter's Mark + Extra Attack) was the better play. Added simMartialFirstClass to gate the picker off for Ranger and Paladin (whose default kit is also weapon-first / no damage spells). J2c experimented with widening the picker to control + heal spells; heal-spell preempt cost druid 10pp (slot heals are 10HP vs 40HP consumables) and control-spell scoring at 22 cost warlock 6.6pp. Both reverted. Corpora retained under baseline_j2c*.jsonl for the post-mortem in sim_results/j2b_findings.md. Post-J2 L12 leaderboard (baseline_j2a_v2_all10.jsonl, n=100): fighter 80.0, ranger 80.0, paladin 78.4, rogue 76.8, druid 61.6, mage 53.4, sorcerer 50.6, warlock 48.2, bard 40.4, cleric 39.0. The caster cluster is dissolved; martials are within ±5pp of J1 (sweep noise). Bard/cleric still trail, but it's no longer a sim artifact — their defaultKnownSpells damage rosters cap at L2 and the picker can't pick spells they don't have. That's a prod-level fix, deferred. J3 trace (sim_results/j3_findings.md): T5 dragons_lair walls every solo class at 0% (Infernax 546 HP vs solo player HP 110–175; ~25% boss HP eaten before TPK across all classes). Per the J3 plan menu, this is party-shaped content the engine doesn't yet have parties for. Surface a TwinBee-voiced heads-up in handleDnDExpeditionCmd's start path and a matching tag in !expedition list — players see "raid-shaped — solo runs not yet survivable" before they spend outfitting coin. No combat or class balance changes. Files: cmd/expedition-sim/main.go +trace flag; expedition_sim.go picker + SimCombatSummary.Events + spellbook seed; dnd_expedition_cmd.go raidContentWarning + list tag. All baselines + traces + findings checked in under sim_results/.
68 lines
6.3 KiB
Markdown
68 lines
6.3 KiB
Markdown
# J2a — autoResolveCombat cast+consume picker (verification sweep)
|
||
|
||
**Date:** 2026-05-17
|
||
**Corpus:** `sim_results/j2a_traces.jsonl` — 240 runs (6 classes × {manor_blackspire, underdark} × L12 × n=20)
|
||
**Pre-state corpus:** `sim_results/j2_traces.jsonl` (same shape, picker disabled). See [j2_findings.md](j2_findings.md) for the strawman diagnosis that motivated this fix.
|
||
|
||
## What J2a shipped
|
||
|
||
- `internal/plugin/expedition_sim.go`:
|
||
- New `simPickCombatAction(uid, sess)` decision tree:
|
||
1. HP < 40% MaxHP + heal consumable in inventory → `!consume <heal>`
|
||
2. Caster with a usable damage spell (slot or cantrip) → `!cast <spell>`
|
||
3. Otherwise → `!attack`
|
||
- New `simPickSpell(c, uid)` ranks prepared damage-effect spells by slot level (descending) then expected damage from the dice string. Reaction-cast spells excluded; non-cantrips require an available slot at their native level (no upcasting in v1).
|
||
- `BuildCharacter` now calls `ensureSpellsForCharacter` so the synthetic spellbook is populated (prod players hit this on character creation; the sim path skipped it).
|
||
- `autoResolveCombat` dispatches via the picker instead of always swinging.
|
||
|
||
Built on top of the per-round `-trace` plumbing landed earlier in J2 (`SimCombatSummary.Events`, `SetSimIncludeTrace`).
|
||
|
||
## Verification (n=20 per cell, L12 only — same cells as j2_findings)
|
||
|
||
| class | zone | n | reachBoss | won | lost | rnd | pDmg | eDmg | slotRuns | casts | consumeRuns | eHP@loss |
|
||
|----------|------------------|----|-----------|-----|------|------|-------|-------|----------|-------|-------------|----------|
|
||
| mage | manor_blackspire | 20 | 15 | **12** | 3 | 7.2 | 133.0 | 74.3 | 15 / 15 | 93 | 12 | 83.3 |
|
||
| mage | underdark | 20 | 8 | 3 | 5 | 7.0 | 143.8 | 78.6 | 8 / 8 | 54 | 2 | 34.2 |
|
||
| cleric | manor_blackspire | 20 | 6 | 2 | 4 | 11.3 | 101.3 | 146.2 | 6 / 6 | 58 | 5 | 64.8 |
|
||
| cleric | underdark | 20 | 4 | 0 | 4 | 7.0 | 74.5 | 89.5 | 4 / 4 | 28 | 0 | 87.5 |
|
||
| sorcerer | manor_blackspire | 20 | 13 | **9** | 4 | 8.0 | 130.9 | 87.3 | 13 / 13 | 91 | 8 | 53.2 |
|
||
| sorcerer | underdark | 20 | 7 | 1 | 6 | 5.9 | 89.6 | 67.4 | 7 / 7 | 40 | 1 | 86.2 |
|
||
| warlock | manor_blackspire | 20 | 12 | 6 | 6 | 13.2 | 131.0 | 140.6 | 12 / 12 | 141 | 9 | 45.5 |
|
||
| warlock | underdark | 20 | 15 | 0 | 15 | 6.8 | 74.3 | 82.5 | 15 / 15 | 101 | 1 | 87.7 |
|
||
| bard | manor_blackspire | 20 | 12 | 0 | 12 | 10.6 | 68.1 | 128.2 | 12 / 12 | 117 | 6 | 75.9 |
|
||
| bard | underdark | 20 | 12 | 0 | 12 | 11.2 | 74.3 | 108.8 | 12 / 12 | 132 | 2 | 87.7 |
|
||
| druid | manor_blackspire | 20 | 15 | **14** | 1 | 6.9 | 142.6 | 84.5 | 15 / 15 | 87 | 11 | 79.0 |
|
||
| druid | underdark | 20 | 18 | 3 | 15 | 5.1 | 123.3 | 82.3 | 18 / 18 | 85 | 5 | 47.5 |
|
||
|
||
`slotRuns` = boss fights where ≥1 spell_cast fired. `casts` = total spell_cast events. `consumeRuns` = fights with ≥1 use_consumable. `eHP@loss` = mean enemy HP remaining when player lost.
|
||
|
||
## DoD check
|
||
|
||
- **J2a DoD: ≥80% of boss fights produce a cast or consumable use.** **MET** — every cell shows `slotRuns == reachBoss` (100% cast rate).
|
||
- Manor mean clear (5 trailer casters): mage 60%, sorcerer 45%, warlock 30%, bard 0%, cleric 10%. Mage and sorcerer clear the ≥40% J2 target on the strength of the sim fix alone; warlock/cleric are close, bard is the one outlier still at zero.
|
||
- Druid jumped 0% → 70% manor / 0% → 15% underdark. Pre-J2a it was the lone mid-band class on weapon-attack alone; with cast on top, druid pulls into the top tier.
|
||
|
||
## What's still broken and why
|
||
|
||
- **Bard 0% clear despite firing 117 casts/cell.** Bard's prepared damage spell registry is thin: `vicious_mockery` cantrip (4d4 at L11) and `shatter` L2 (3d8) are the only damage-effect options in `defaultKnownSpells`. Every other bard spell is control/buff/heal, which the v1 picker skips. Bard's average pDmg 68 in 11-round fights = 6/round — that's accurate for the spell pool, not a picker bug. Fix candidates: (a) extend bard's default known-spell list with a damage L3+; (b) widen picker to score `EffectControl` (hold_person, hypnotic_pattern) which trivializes enemy turns and effectively converts to damage by buying free player rounds; (c) include healing spells in the consume-priority path. **(b)** has the highest expected ROI and lifts cleric and warlock too.
|
||
- **Cleric still bottom-of-table.** 10% manor / 0% underdark. Cleric's damage-effect spells: `sacred_flame` cantrip (4d8 at L11) — that's it. Same shape problem as bard. `spiritual_weapon`, `bless`, `guiding_bolt`, `cure_wounds` aren't reachable by the v1 picker. Fix: same as bard — score control spells and recognise heal spells as alternatives to consumables (cleric has 40-HP healing word at low slot cost).
|
||
- **Warlock underdark 0%.** Different shape: high cast rate (101 casts/15 fights = 7/round), high player damage (74), but loses anyway because enemy damage averaged 82.5 and warlock has no defensive lever. Underdark boss is dropping warlock before warlock can drop it. This is a class-balance problem the post-J2b corpus needs to confirm before we move on.
|
||
|
||
## Implications for J2b (the big re-baseline)
|
||
|
||
A full n=100 all-classes baseline (`sim_results/baseline_j2a_all10.jsonl`) is kicked off; that replaces `baseline_j1_all10.jsonl` as the canonical post-fix corpus. Expected shape:
|
||
|
||
- Mage / sorcerer / druid pull into mid-band (40–70% L12 clear range).
|
||
- Bard / cleric stay at bottom, but the *reason* is a thin damage-spell pool, not a sim artifact.
|
||
- Warlock somewhere middle; partial-clear depending on zone.
|
||
- Martials should be unchanged (picker is a no-op for non-casters with no low-HP heal trigger).
|
||
|
||
If that picture holds, the right J2 follow-up is **picker widening** (control + heal scoring) rather than per-class balance riders. Cleaner than five class buffs.
|
||
|
||
## Files touched
|
||
|
||
- `internal/plugin/expedition_sim.go` — picker, BuildCharacter spell-seed, dispatch in autoResolveCombat.
|
||
- `cmd/expedition-sim/main.go` — already had `-trace` from earlier in J2.
|
||
|
||
No prod paths changed. The fix is sim-scoped.
|