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/.
44 lines
4.2 KiB
Markdown
44 lines
4.2 KiB
Markdown
# J3 — T5 dragons_lair universal wall (initial trace)
|
||
|
||
**Date:** 2026-05-17
|
||
**Corpus:** `sim_results/j3_traces.jsonl` — 60 expeditions (6 representative classes × L12 × dragons_lair × n=10), per-round traces enabled. Hypothesis-sift only; not a balance sweep.
|
||
**Pre-state:** post-J2 picker (`baseline_j2a_v2_all10.jsonl` shows 0% clear across all 10 classes at L12 dragons_lair).
|
||
|
||
## Boss-room aggregates (L12 vs `boss_*` of dragons_lair)
|
||
|
||
| class | n | reachBoss | won | rounds | pDmg | eDmg | enemyHPMax | enemyHP@end | playerHPMax | playerHP@end | casts | consume |
|
||
|---------|----|-----------|-----|--------|-------|-------|------------|-------------|-------------|--------------|-------|---------|
|
||
| fighter | 10 | 10 | 0 | 8.8 | 122.5 | 207.8 | **546** | 423.5 | 173.6 | 0 | 0 | 14 |
|
||
| ranger | 10 | 10 | 0 | 8.8 | 133.4 | 197.1 | **546** | 412.6 | 131.0 | 0 | 0 | 18 |
|
||
| paladin | 10 | 10 | 0 | 8.3 | 76.0 | 200.0 | **546** | 470.0 | 169.4 | 0 | 0 | 16 |
|
||
| rogue | 10 | 9 | 0 | 6.6 | 63.7 | 168.6 | **546** | 482.3 | 133.2 | 0 | 0 | 11 |
|
||
| druid | 10 | 9 | 0 | 4.8 | 122.4 | 136.9 | **546** | 423.6 | 131.0 | 0 | 34 | 9 |
|
||
| mage | 10 | 6 | 0 | 5.2 | 92.7 | 129.2 | **546** | 453.3 | 112.5 | 0 | 26 | 5 |
|
||
|
||
## Diagnosis
|
||
|
||
T5 dragons_lair boss is at **546 max HP**. Solo characters (player HP 110–175) take 5–9 rounds to die, dealing 60–130 player damage in that time — they chew down ~10–25% of the boss before being TPK'd. This is **party content tuned for solo by accident**.
|
||
|
||
Per-round math: clearing 546 HP in ~10 rounds requires ~55 DPR sustained. Solo player DPR caps around 15–25 even on fighter (122/8.8 = 13.9 mean) and druid (122/4.8 = 25.5 — burst before dying). A 4-character party at the same DPR clears in ~7 rounds, which is the survivable window the boss seems tuned around.
|
||
|
||
Plan §J3 listed three hypotheses; this trace rules them out cleanly:
|
||
|
||
1. ~~`autoResolveCombat` unfair (only swings).~~ **Closed by J2a.** Druid/mage are now casting (34/26 casts in 10 fights) and still 0% clear.
|
||
2. ~~Boss HP pool just too deep / multi-phase.~~ **No multi-phase complication observed** — fights are linear monotone HP descent, the boss just doesn't have a phase mechanic exposed in the trace. The HP pool *itself* is the issue.
|
||
3. **Party scaling missing — boss is tuned for 4-character parties.** Confirmed by the consistent ~25% boss-HP-eaten-at-player-death across every class.
|
||
|
||
## Recommended J3 lever
|
||
|
||
Per `gogobee_harvest_charges_plan.md` §J3 likely-levers, the difficulty memo allows *lifting trailers without pushing monster scaling*. **Reducing boss scaling at low party counts is in-bounds.**
|
||
|
||
- **Auto-scale boss HP (and probably boss damage) by participant count.** A simple `bossHP *= (0.4 + 0.15*participants)` style ramp would put solo at 0.55× (≈ 300 HP from 546) and 4-up at full. That brings a 5–9 round solo fight in line with the boss-HP curve already validated at T3/T4 (post-J2 manor: 60%+ clear for top classes; underdark: 30–100%).
|
||
- **Alternative:** Document dragons_lair as raid content. Surface a verb-style warning in `!zone` help ("Bring friends — this dragon eats solo adventurers."). Cheaper but doesn't lift the trailer end of the solo population, which `feedback_difficulty_target` says we should.
|
||
|
||
Recommendation: **try party-scaling first** — it's the lever that both keeps the boss difficult for parties AND respects the "lift trailers" feedback. Document-only is the fallback if scaling proves intrusive.
|
||
|
||
## Open question for the operator
|
||
|
||
What's the *intended* T5 boss audience — solo or party? If solo, scaling is required. If party, dragons_lair should be flagged as raid content and not appear in solo-play recommendations. The sim has been running solo throughout, which the plan's J-phase open questions already flagged needs an operator answer.
|
||
|
||
Out of scope here: implementing the scaling lever. That's a prod change in the boss HP / damage path. Picking up that work would need the operator's call on the audience question first.
|