N3 widened the combat *roster* from 1 to N but never widened the action model,
the scaling model, or the test net to match. Building the hireable companion
walked into all three. Only one of the four defects found was the companion's;
the rest have been live in prod for every human party since N3.
The party golden did not exist (§5)
Solo combat is pinned exhaustively (7468 lines); the entire N-body layer had
nothing. That is why a healer class that cannot heal shipped without a test
going red. Adds party_characterization.golden (9 scenarios x 5 seeds, incl.
weak and dying seats) and TestPartyCharacterization_OneSeatIsStillSolo, so the
N-body path can never quietly stop being a superset of the balance corpus.
Regenerate only on purpose: -update-party.
No action could target another seat (§1)
Every heal in the engine was self-scoped. A party cleric could not put one hit
point on a friend. Adds turnActionEffect.AllyHeal/AllySeat, `!cast <spell>
@user` and `--target @user` -- the latter has been advertised in !help and
silently swallowed by parseCombatCast since SP2 ("reserved for SP3"). The auto
picker uses it too (simPickAllyHeal), so away-players and engine-driven healers
behave like competent ones. It will not raise the dead.
Corpses kept buffing the boss (§2b)
enemyActionsThisRound counted len(st.actors), dead included -- so a party that
lost a member kept paying for them, and the survivor faced a boss still swinging
at two-player cadence, alone. A death spiral with the arrow pointing the wrong
way. Now counts livingActors(). Party golden moved deliberately for this.
An engine-driven seat was a bool any command could clear (§3)
autoDriveCombat drives a party by dispatching each seat's turn AS that seat, so
a companion's own auto-played move arrived at beginCombatTurn looking like a
player returning to the keyboard and cleared the latch that was moving him. He
then stood in the fight doing nothing while the boss he had inflated killed
everyone. ActorStatuses.EngineDriven is now a persisted seat property that no
command clears, and the driver calls driveEngineSeat instead of impersonating.
"The party" could be empty (§4)
A solo expedition has no expedition_party rows, so asking the roster who was in
the party answered "nobody" -- and every caller fell back to something plausible.
That is how the companion got hired at level 1 for exactly the player the feature
exists for. expeditionParty()/partyHumans() always include the owner.
The companion himself (!expedition hire [class] / !expedition dismiss)
Day 1, leader only, costs coins, role-fills the gap, globally exclusive. He is an
NPC seat and must never become a player: no player_meta, no dnd_character, no
inventory, no DM room -- mint him a player_meta row and
ensureDnDCharacterForCombat will auto-build the news bot a real character, and
he starts appearing in the graveyard and filing death notices about himself.
Mail and seats are different sets: he fights, he does not get written to.
Measured on millenia, n=750/arm. Before these fixes he was -28pp -- worse than no
companion at all. After: solo 48.5% -> 63.9% clear (+15.3pp), with +28.0pp for
trailing players and +2.0pp for leaders. Help, never a carry.
The solo golden is byte-identical throughout: solo combat provably did not move,
and the balance corpus is intact.
Known gap: the companion cannot cast (castActionForSeat loads a sheet from the DB
and he has none by design), so a hired Cleric is still just a bad fighter.
Claude-Session: https://claude.ai/code/session_01J5SQZWoLmL3M3mw2XmHHdy
Run the actual Adventure module against a copy of the prod DB with no
Matrix client, to smoke-test before deploy.
- expedition-sim: -real-user @mxid runs an EXISTING character loaded from
-data's gogobee.db instead of a synthetic build. SimRunner gains
PrepareRealCharacter (heals to full + tops up bankroll; keeps real
race/class/subclass/level/gear/spells).
- plugin.SendReply now honors the MessageSink like SendMessage/SendDM.
Reply-based handlers (duels, !town, !rivals, !achievements) previously
bypassed the capture seam and hit a nil client under the sink. Prod
behavior is unchanged (sink is nil in production).
- exercise_prod_test.go (build tag: prodexercise) drives every N-series
feature — world boss, duels, Shadow, Renown, achievements, journal,
town registries, vault, gifting — against a prod DB copy with all
outbound messages captured. Gated on GOGOBEE_PROD_DB_DIR; never runs in
normal CI.
Claude-Session: https://claude.ai/code/session_017mEwUmmS7aQTP2NQXj6rUa
Mechanical `gofmt -w ./internal ./cmd`. Mostly struct-field realignment that
had drifted, plus a few trailing-newline fixes. No behaviour change — gofmt is
semantics-preserving, and build/vet/test are green either side.
Split out from the code-review fixes that follow so those stay reviewable
instead of hiding inside a wall of realignment.
Adds -party N / -party-classes to expedition-sim. Followers are seated
through the real !expedition invite / !expedition accept pair, so the
harness measures the tier gate, the busy guard and the supply pooling
rather than a roster hand-built to succeed. A follower who is refused
halts the run: a party reading taken from a walk that was secretly solo
is worse than no reading.
It immediately found what it was built to find. Only elite and boss
doorways seat the roster; exploration rooms, patrols and harvest
interrupts all resolve through SimulateCombat against ctx.Sender, and
P6d made the walk commands leader-only. So on a 38-room T5 expedition
the party fights 2-3 rooms together and the leader solos the rest -
then dies alone, tearing down the run rows only they own.
Measured at L15/16 over dragons_lair + abyss_portal, party of 3, n=15
per cell: zero TPKs and zero member deaths across 240 seats. Every
failure is the leader falling while two untouched members stand at full
HP. Fighter clears 100% (solo: 47-67%), cleric 33-53% (solo: 13-47%).
The band is unreadable until inline combat seats the party, so the C1
contingency - +35% monster HP per member - stays on the shelf; it would
punish the trash the leader already fights alone.
Hence the outcome vocabulary grows a third word. "tpk" used to mean any
run-ending event, which is how a leader dying beside a healthy party
stayed invisible through P5 and P6. Now: tpk (roster dead),
leader_down, fled (run over, leader alive) - read off the death flag,
not off HP, which the close-out leaves anywhere. A one-seat roster
makes leader-dead and all-dead the same predicate, so solo keeps its
labels for a real death.
Two bugs found in review before this landed:
- An unknown class was not an error anywhere: -class fightr built a
1-HP character and reported an ordinary outcome for it. Guarded in
BuildCharacter, not the flag parser, since the leader had the bug
long before parties did.
- The roster short-rest healed the dead - handleDnDShortRest does not
gate on the death flag, so a member killed in a won boss fight got
rested back above 0 and stopped counting as a casualty.
Golden byte-identical; go test ./... green.
- expedition-sim matrix worker now captures child stderr and dumps
runErr/stderr/stdout-snippet on failure so halted rows have a cause.
- simPickSpiritualWeapon walks slots 2..5 and upcasts when L2 is spent
instead of silently skipping the spell on high-level clerics.
- advanceOnceWithOpts !inlineBossCombat branch now emits the same
"Room X/Y — Boss/Elite. Type !fight to engage." line as foreground.
Adds inlineBossCombat alongside compact in runAutopilotWalk and
advanceOnceWithOpts. Production background autorun keeps both true
(inline auto-resolve), foreground stays both false (manual !fight), the
sim now uses compact=true + inlineBossCombat=false so the boss/elite
doorway returns stopBoss/stopElite after the safety gate — autoResolveCombat
+ simPickCombatAction / simPickSpell drive the fight via the turn-based
engine. The picker (and D8-b upcasting) has been dead since D3's
compact-inline boss rooms; this re-wires it.
n=50/cell L10 smoke vs d7d (zones T1-T3):
bard forest_shadows 61 → 100 (+39)
bard manor_blackspire 10 → 34 (+24)
cleric forest_shadows 15 → 96 (+81)
cleric manor_blackspire 0 → 54 (+54)
fighter T1-T3 100 (unchanged)
Also parallelizes matrix mode via subprocess workers (each child has its
own SQLite — db package globals preclude in-process parallelism). New
-jobs flag, defaults to runtime.NumCPU(). 8 workers gave ~7x speedup on
the smoke matrix.
cmd/expedition-sim -days N caps runs by synthetic day rollovers
(Outcome="day_capped"). SimResult.DaySnapshots traces HP/SU/threat/rooms
at start, every Night-camp rollover, and end-of-run — unblocks empirical
D5-d retune of phase5BDailyBurnRatePct against per-day SU draws.
Synthetic sim chars are vanilla base-class (no pet, no subclass), so the
per-round pet attack/deflect/whiff path was never exercised in the sim.
Add -pet-level N (1-10) which stamps a base Massive Dog (no armor) onto the
AdventureCharacter via the normal save path before the run, so combat's
DerivePlayerStats sees HasPet()==true. 0 (default) stays petless.
Measured lift (n=40, 10 classes x L3/7/12 x 3 zones): overall clear-rate
38.3% petless -> 47.8% at pet L10 (+9.4pp); biggest gains go to the caster
trailers (bard +13.3, warlock +11.9, cleric +11.4), narrowing class spread.
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/.
-matrix sweeps the cartesian product of -classes × -levels × -zones
with -runs replicates per cell, fresh sqlite DB per run, emitting one
JSON object per stdout line. Log field is suppressed by default in
matrix mode (override with -log=true) since a single matrix cell's
log can easily dwarf the per-run summary.
Default -cap shrunk from 200 to 50 — RunExpedition now counts
autopilot bursts (not !expedition run invocations), and 50 bursts is
enough to clear a tier-1 zone with headroom.
SimRunner re-uses production plugin code paths against a fresh sqlite DB —
SendDM no-ops without a Matrix client, so ground truth comes from the DB.
BuildCharacter persists adv+dnd character rows with class-appropriate
ability scores and spell slots; RunExpedition loops !expedition run with
stall detection until the autopilot halts or the expedition closes.
The cmd binary is a thin orchestrator: one flag-driven run, JSON output.
Per-room structured logging, day-cycle fast-forward, and matrix batching
are next session.
B4: Slot classifier no longer treats "Springing" / "Snaring" / "Devouring"
as ring matches; tokenises by word boundary instead of raw substring.
Adds DnDSlotCloak so cloaks/capes/mantles/wings stop evicting body armor
from the chest slot. Regenerated magic_items_srd_data.go: boots_of_*,
gloves_of_missile_snaring, bag_of_devouring, talismans, and 6 cloaks all
land in the right slot.
R4: equipMagicItem swap-back returns the prior occupant at full Value
instead of half — swapping a curio shouldn't tax it.
R5: Attunement count is recomputed *after* the swap-back so freeing the
prior occupant's bond opens the slot for the incoming item.
R1: Inventory tags magic_item rows with 🔮 + rarity label and prints a
single equip-magic footer when any are present.
R6: Sheet's Magic Items block marks unbonded items as **(inactive)**
with the reason (cap full vs unbonded), so over-cap items aren't silent.
R7: New activeMagicItemsLine surfaces a one-shot "your curios stir: …"
at combat-start in both the dungeon path and !fight, mirroring the way
class passives are surfaced.
R8/R9: dropMagicItemLoot pretty-prints rarity, drops "wondrous", calls
attunement "needs bonding", appends "auto-uses in combat" for
potions/scrolls, and routes persistence errors to slog instead of
leaking %v into chat.
R2/R3: Curios shelf now shows "Very Rare" not "very_rare", drops the
bare "wondrous" word (the effect line carries it), renders the codified
magicItemEffectSummary above the SRD desc, and ends with a one-line
plain-language "what is bonding" footnote.
R10: Curios stock day flips at 06:00 UTC instead of midnight so EU
players don't see a fresh shelf at 1 a.m. mid-session.
R11: Curios buy resolver disambiguates fuzzy matches — typing "ring"
when several rings are on the shelf lists candidates instead of
silently selling the first.
P1: Greeting grid pairs Curios with an Exit chip so the 2-column
emoji layout doesn't dangle.
P2: Equip-magic empty state trimmed to one line.
P4 (back-from-curios reprompt) deferred — the existing back-flow is
correct, just verbose; not worth the surface-area expansion this
session.
Tests: word-boundary classifier, cloak/chest coexistence, full-value
swap-back. go test ./... + go vet clean.
R22: replace race copy that promised mechanics the engine doesn't deliver.
- Tiefling: wire FireResist as a CombatModifier. Enemy main attack is
halved when monster is FireAttacker-tagged; aoe_fire abilities are
halved unconditionally; fire-tagged traps deal half damage to Tieflings.
DnDMonsterTemplate carries FireAttacker; toCombatStats propagates it.
Hand-authored fire entries tagged in dnd_bestiary.go (flameskull,
magmin, azer, salamander, fire_elemental, emberlord_thyrak,
young_red_dragon, infernax, belaxath).
- Open5e tuned generator derives FireAttacker from the highest-AvgDamage
attack's DamageType (threshold AvgDamage>=5). 19 tuned monsters tag.
Regenerated bestiary_tuned_data.go.
- Elf: drop "immune to sleep" (no sleep mechanic); reframe as keen
senses + trance flavor.
- Half-Elf: drop "two bonus skill proficiencies" (no skill system);
reframe as adaptable cross-cultural know-how.
- Tiefling copy: drop "bonus on CHA checks" (no checks); keep fire
resistance with flavor framing.
R23: DnDRaceInfo grows a BestFit field; renderRaceMenu emits an
"_best with: …_" hint per race so spiky stat spreads (Orc -1/-1/-1)
read as specialist picks rather than a brick of penalties.
R24: dnd.go header comment for the caster classes now reflects the
shipped state (Playable=true, spell lists populated) instead of the
pre-Open5e scaffold language.
Tests: TestApplyRacePassives gains a FireResist column; new
TestTieflingFireResistance asserts ~0.5x ratio over a 300-trial sweep
against a FireAttacker enemy. Full suite green.
Plumbed through the open5e importer so regen stays safe:
- New cmd/open5e-import/desc_overrides.go holds two per-ID override
maps (spellDescOverride, magicItemDescOverride) and a regex-driven
cleanDesc sanitizer. Override wins outright; otherwise the SRD
first-sentence runs through cleanDesc, which strips the phrases
the S3 acceptance criteria forbid (saving throw[s], spell slot,
within range, 5-foot, DC <n>, "(save DC X)" parentheticals,
"constitution score is N", "out to a range of N feet"). A small
post-pass repairs the orphan stubs the strippers leave behind
(" and." trailers, "must make." after the saving-throw object
is gone).
- gen.go (spells) + magicitems.go now call spellDescription /
magicItemDescription instead of raw firstSentence; same hand-
authored override pattern, same cleanDesc fallthrough.
- Override coverage: the 19 SRD-only spells that show up in
defaultKnownSpells (call_lightning, charm_person, vicious_mockery,
…) plus ~35 high-visibility magic items (Amulet of Health, every
Belt of Giant Strength variant, Cloak of Displacement, etc.).
Tone is outcome-first second-person with bite — these surface
in the spellbook and the curio shop, so they get to be funny.
- tuned.go (R21) + magicitems.go strip "(...)" from emitted Names
via stripNameParenthetical. Slug keeps the variant; only the
display text loses the qualifier. Two bestiary entries
(giant_rat_diseased, deep_gnome_svirfneblin) and stone_of_good_luck
affected.
- Regenerated all three data files. Acceptance grep is clean:
zero hits for any banned phrase in Description/Desc fields.
- New cmd/open5e-import/desc_overrides_test.go covers the
sanitizer regressions, orphan-repair, name-strip, and the
override-wins-but-fallthrough-still-sanitizes path.
Conflicts: none. S4 (magic-item UX) wanted this in first so the
new curio renderer consumes clean text — ready for it now.
The tuned bestiary previously left every generated entry with a nil
Ability. abilityFromTraits now classifies each creature's SRD trait
names against a priority-ordered rule table, mapping the most
combat-defining trait onto a MonsterAbility effect (death_aoe,
regenerate, spell_resist, evade, enrage, ...). Creatures whose traits
are all non-combat stay nil. 165 of 322 entries get an ability.
fetch/gen magicitems subcommands vendor data/open5e/magicitems.json (237
SRD items) and classify them into a generated registry. magic_items.go
holds the MagicItem struct + Kind enum + an init-time overlay merge where
a hand-authored entry wins on ID collision, mirroring the spell and
bestiary imports. Not yet wired into zone loot or the shop — that
integration is a deliberate follow-up.
Adds `gen tuned` to cmd/open5e-import: a deterministic formula that scales
every raw SRD stat block down to an engine-ready DnDMonsterTemplate. HP/AC/
AttackBonus are verbatim SRD (AC clamped to the engine min 10); the Attack
stat is interpolated from attackByCRPoints, a CR→Attack anchor table lifted
from the hand-tuned dndBestiary (CR is the calibration axis the 2026-05-10
rebalance used — raw SRD per-hit damage is ignored). Speed/BlockRate are
coarse baselines from SpeedWalk/AC.
bestiary_tuned.go merges the 322 generated templates into dndBestiary, but
hand-authored roster entries win — the merge only fills IDs the roster does
not already define, so playtested numbers and wired abilities are untouched.
Abilities are deliberately not wired: every generated entry has a nil
Ability, with the SRD multiattack/trait text parked in Notes as raw material
for the follow-up ability-wiring pass.
fetch|gen bestiary subcommands vendor data/open5e/monsters.json (322
SRD monsters) and generate bestiary_srd_data.go — all 322 as raw SRD
stat blocks (HP/AC/ability scores/CR + per-attack damage dice).
This is a balance-baseline reference, not an engine roster: raw SRD
damage one-shots the solo player, so nothing here feeds combat. It's
what the future tuning pass reads against when deriving dndBestiary /
srdProfiles entries. XP is derived from CR (Open5e has no XP field).
Rebuilds the cmd/open5e-import CLI (fetch/gen × spells) to vendor
data/open5e/spells.json (319 SRD spells) and generate
dnd_spells_srd_data.go (237 after the level>5 filter). mapClasses unions
the API's incomplete structured spell_lists field with the complete
free-text dnd_class field so all eight casters get spells.
dndSpellRegistry loads buildSRDSpellList() first; the hand-authored
buildSpellList() overlays it (hand wins on ID collision). Playable=true
flipped for Druid/Bard/Sorcerer/Warlock/Paladin, each with a
defaultKnownSpells case. TestDefaultKnownSpellsExistInRegistry now covers
all eight classes.
.gitignore: vendor data/open5e/ while keeping the rest of data/ ignored,
and anchor the open5e-import binary pattern so it stops swallowing the
cmd/open5e-import source dir. NOTICE adds CC-BY-4.0 / SRD attribution.
Replaces hardcoded tip scenarios with solver-frequency-backed decisions, adds
equity range display, fixes bet-size matching tolerance (25% threshold), and
adds comprehensive test coverage for scenario validation.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
