# Long Expeditions — multi-session plan

> Goal: dungeon expeditions become real multi-day journeys (T1 ≈ 2 days → T5 ≈ 7 days). Player picks **camp supplies** at launch, then watches it play out. Camp pitch/break, elite engagement, and **boss engagement** all become autonomous. Foreground `!camp` / `!fight` survive as overrides, not the expected path.

## 1. Why the current shape is "too short"

| Tier | Zones | Rooms | Real-time exit |
|------|-------|------|----------------|
| T1   | Goblin Warrens, Crypt Valdris       | 6–7  | usually <1 calendar day |
| T2   | Forest Shadows, Sunken Temple       | 7–8  | usually 1 day |
| T3   | Manor Blackspire, Underforge        | 7–9  | 1–2 days |
| T4   | Underdark, Feywild Crossing         | 8–10 (×regions) | 2–3 days |
| T5   | Dragon's Lair, Abyss Portal         | 9–10 (×regions) | 2–4 days |

(Room counts from `internal/plugin/dnd_zone.go:70-134`; tier zones from memory `project_expedition_difficulty`.)

Because the autopilot walks ~3 rooms / 2h (`expedition_autorun.go:43,53`) and boss/elite stops are manual, a player who pays attention clears T1–T3 in an evening. Briefing/recap UTC ticks fire but rarely matter — the expedition is over before "day 2" ever lands.

## 2. Target shape

| Tier | Target duration | New room budget (single-region) | Notes |
|------|-----------------|---------------------------------|-------|
| T1   | 2 days  | 12–14 | one autopilot-camp |
| T2   | 3 days  | 16–20 | two camps |
| T3   | 4 days  | 22–26 | three camps; threat starts to bite |
| T4   | 5–6 days | 28–34 (split across 3–4 regions) | base camp emerges |
| T5   | 7 days  | 36–44 (split across 3–4 regions) | full siege/temporal pressure |

(Numbers are first-draft anchors; class-balance sim drives the actual dial in D7.)

Duration is measured in **expedition-days advanced by autopilot camp**, not real-time UTC days. Each autopilot night-camp = day++. Real-time UTC still drives the morning briefing DM cadence, but day-count and supply burn become event-driven (camp = sleep = day++ = burn).

The autopilot decides:
- when to camp (and which camp type)
- whether to engage elites and bosses, or to camp first and retry
- whether to extract early on starvation/HP collapse

The player decides:
- supply pack purchases at launch (only meaningful pre-expedition choice)
- whether to override (`!camp <kind>`, `!fight`, `!expedition extract`)

## 3. Phasing

Built across multiple sessions; each phase ships independently with tests and a small DM-side feedback loop.

### D1 — Length + room-count rework
**Files:** `dnd_zone.go` (per-zone Min/MaxRooms), `dnd_expedition_region.go` (multi-region region sizes), `dnd_zone_run.go:generateRoomSequence` (still terminates Entry → … → Elite → Boss, but with more Explorations).
**Work:**
- Re-pitch Min/MaxRooms per zone to the table in §2.
- Multi-region zones: extend per-region room budgets so the total stays in band when summed.
- Keep the fixed Entry / Trap / Elite / Boss anchors; only Exploration count scales.
- Consider 2 traps + 2 elites for T4–T5 to break up the long middle (deferred decision — see §6).
- Update existing zone graphs in `zone_graph_*.go` for the new lengths. *Auditing whether each zone has a hand-tuned graph or relies on the linear compiler is part of D1 scoping.*

**Exit criteria:** `expedition-sim` shows mean-room-cleared per zone in the new band; no test fixtures hard-code old counts.

### D2 — Autopilot camp scheduler

**D2-a (shipped 2026-05-27):** new `expedition_autocamp.go` with pure `decideAutopilotCamp` + `pitchAutopilotCamp` + dwell-window lifecycle (`shouldSkipAutoRunForCamp`, `breakAutoCampIfDue`). Wired into `tryAutoRun`: skip while inside `minAutoCampDwell` (4h), break + walk past it, post-walk scheduler pitches HP-low rest camps and region-boss base-camp waypoints. CampState gained `AutoPitched` so auto- vs player-pitched lifetimes diverge. Day rollover is still UTC-anchored; D2-b moves it event-anchored.

**D2-b (shipped 2026-05-27):** event-anchored day rollover. `dnd_expedition_cycle.go` adds `eventAnchoredCutoff` + `isEventAnchored`, splits the briefing body into `nightRolloverBurn` + `nightRolloverDrift` (+ a convenience `processNightCamp` that runs both back-to-back). For expeditions started ≥ cutoff, `deliverBriefing` skips mutators and posts a re-engagement DM; if `last_briefing_at` is older than `nightSafetyNet` (28h) it force-fires `processNightCamp` so a stalled autopilot doesn't freeze the expedition. The autopilot scheduler grew a `Night` flag — `decideAutopilotCamp` sets it when `time.Since(LastBriefingAt) ≥ nightCampWindow` (16h); `pitchAutopilotCamp` then runs the burn → camp → rest → drift sequence in one pitch. Manual `!camp <type>` runs the same flow when it's the first camp since the last rollover. Legacy UTC-anchored expeditions (start_date < cutoff) keep the original mutator flow via the same staged helpers, preserving rest-before-drift ordering through `processOvernightCamp`. Tests fence cutoff to year 9999 in `TestMain` so existing legacy assertions stand; new tests exercise the night decision, night-pitch rollover, event-anchored skip, and safety-net force.


**Files:** new `expedition_autocamp.go`; hooks in `expedition_autorun.go:tryAutoRun`; reuse `dnd_expedition_camp.go:campPitch` / `applyCampRest`.
**Heuristics (in priority order):**
1. **Day-budget pacing** — pitch when rooms-walked-this-day ≥ tier-specific day-room target. Standard if room cleared, rough if not.
2. **Resource gates** — pitch if HP < 35% of max, or supplies projected to cover <1 more day's burn (`exp.Supplies.Current < exp.Supplies.DailyBurn * harshMod`).
3. **Post-elite breath** — auto-pitch a standard camp in the cleared elite room.
4. **Region boss cleared in multi-region zone** — auto-pitch base camp at first eligible site.
5. **Don't pitch when** mid-fight, in a trap/boss room, in a fork-pending state, or already camped.

Auto-break already happens on move (`autoBreakCampOnMove` in `dnd_expedition_camp.go:401`). Autopilot just needs to *not* break a freshly-pitched camp by immediately walking on — gate by `time.Since(camp.EstablishedAt) > minCampDwell` (15min real-time? or simply: don't auto-walk while camped).

**Day-rollover semantics change (decided event-anchored):** today, day++ happens at the 06:00 UTC briefing (`dnd_expedition_cycle.go:236`). After D2, the autopilot **night-camp pitch** is the event that fires day++, supply burn, overnight rest, and threat drift — in one atomic rollover the camp scheduler triggers. The 06:00 UTC briefing tick survives only as a re-engagement DM anchor (and a safety net: if for some reason no camp pitched in N hours, force one). Concretely:
- Move the body of `deliverBriefing` (`dnd_expedition_cycle.go:186-`) into a `processNightCamp(exp)` helper called from the autopilot camp scheduler when it pitches a *night* camp (rough/standard/fortified — not mid-day breath stops).
- Distinguish "night camp" (advances day) from "rest stop" (doesn't). Heuristic: a camp pitched at the end of the day-budget room count is night; one pitched purely for HP/SU recovery mid-day is a rest stop. Both apply `applyCampRest`; only night also runs supply burn + day++ + threat drift.
- The existing UTC briefing ticker becomes: if `last_briefing_at` is older than ~20h *and* the user is active, post a "Day N — here's where you stand" re-engagement DM. It calls **no** mutators.
- Manual `!camp <type>` from the player still does what it does today (apply rest immediately) but **also** counts as a night camp if invoked with no other camp today — keep parity with the autopilot behavior so the override doesn't accidentally stall day advancement.

**Migration:** in-flight expeditions at deploy time keep their existing 06:00 UTC day++ until they end (fence by `start_date`). New expeditions use the event-anchored model.

**Exit criteria:** sim shows autopilot pitches the right kind of camp at the right room with no player input; supply economy still survives a 7-day T5.

### D3 — Autonomous elite + boss engagement
**Files:** `dnd_zone_cmd.go:497-535` (the prev==RoomElite / RoomBoss branch).

**Shipped 2026-05-27.** `dnd_zone_cmd.go` adds `stopBossSafety` + `bossSafetyGate(uid, exp)` (HP < 80%, supplies < daily burn, exhaustion ≥ 3 — the "active low status" interpretation). The elite/boss doorway branch drops the `prev == RoomBoss || !compact` carve-out: in compact mode bosses fall through to the same `resolveCombatRoom` path elites have used, after the gate clears. `resolveCombatRoom` selects monster + label + loot-drop by `run.CurrentRoomType()` so the same call site handles boss kills (zone.Boss bestiary, "👑 Boss — name down", boss-loot drop, elite-tier threat bump). Loss on auto-resolved boss falls through to the existing player-death narration / forceExtractExpeditionForRunLoss path — no special treatment.

The walk loop (`dnd_expedition_cmd.go:717`) only breaks at the elite/boss doorway when `!compact`; compact lets the next iteration auto-resolve. `stopBossSafety` is excluded from the rooms-walked tally and its `autopilotFooter` returns "" — `res.final` already carries the held-back line.

When the gate trips, `tryAutoRun` calls `pitchBossSafetyCamp(exp)` (new in `expedition_autocamp.go`): force-pitches Standard (or Rough fallback) regardless of `decideAutopilotCamp`'s HP threshold or its RoomBoss room-type block, with the same `Night` event-anchored rollover handling. Dwell expires → next tick retries the boss. Foreground `!fight` and foreground `!expedition run` are unchanged: both still stop at the boss doorway for the player who wants to watch. Sim path is unaffected — `stopBossSafety` falls into `expedition_sim.go`'s default branch (tick-a-day, retry next walk).

**Open question:** do we want a "set autopilot off" toggle for players who actively want to play the boss themselves? My take: default-on, single-flag per expedition (`autopilot_engage`) settable at start or via `!expedition autopilot off`. Defer until D3 lands and we see how it reads.

**Exit criteria:** sim runs a full T5 expedition launch → boss kill with zero player commands besides `!expedition start`.

### D4 — DM volume + day surfaces
The big risk: a 7-day T5 with autopilot walking, camping, fighting elites, and killing bosses will spam DMs unless we bundle. The user has already pushed back on recap chatter (`feedback_skip_recaps`).

**D4-a (shipped 2026-05-27):** autopilot DM bundling. `expedition_autorun.go:tryAutoRun` now drops per-tick auto-walk DMs in compact mode. New surface rules:
- Fork / death / run-complete still DM the walk stream — interactive + climax beats.
- A Night=true camp pitch flushes the day as an end-of-day digest (`renderEndOfDayDigest` in new `expedition_autorun_digest.go`) followed by the camp block.
- Boss-safety camp pitch gets a short "holding before the boss" header + camp block (walk stream dropped).
- Non-night auto-camp (mid-day rest / base-camp waypoint) surfaces the camp block by itself.
- Everything else — uneventful walks, preflight pauses, harvest interrupts — goes silent.
Each successful background walk now logs a `walk` entry (`appendExpeditionLog(... "walk" ...)`) so the digest can count rooms walked from structured logs (the raw `r.stream` narration is no longer persisted across ticks). Digest groups counts of walk/harvest/interrupt and inlines threat/milestone/narrative lines for the prior day (`dayExpeditionLog` helper). `maybeAutoCamp` + `pitchBossSafetyCamp` now return the `autoCampDecision` so `tryAutoRun` can branch on `dec.Night`.

**D4-b (shipped 2026-05-27):** morning re-engagement DM anchored to user activity, not 06:00 UTC. `fireExpeditionBriefings` now skips event-anchored expeditions whose `last_activity` is older than today's 06:00 UTC threshold (and we're still inside `nightSafetyNet` — stalled-autopilot force-fires still win). New `maybeDeliverDeferredBriefing` runs at the top of `AdventurePlugin.OnMessage` on every inbound message in any room: it stamps `last_activity` (so the next ticker pass sees the player as present, even from non-bot chatter) and posts the deferred briefing if one is owed (CAS-gated by today's 06:00 threshold; pre-06:00 lazy fires are suppressed to avoid double-emit with the ticker sweep that follows). Day-1 inbounds don't trigger a briefing before the first night camp has happened.

**Remaining work:**
- TwinBee voice + no-recap copy pass on the digest once the shape settles.

**Exit criteria:** a 7-day T5 produces ≤ 10 DMs across the whole expedition (1 launch + 6 end-of-day + 1 boss + 1 run-complete + 1 emergency).

### D5 — Supplies economics retune

**D5-a (shipped 2026-05-27):** per-tier pack caps. `dnd_expedition_supplies.go` retires the global `SupplyPackStandardMax`/`SupplyPackDeluxeMax` constants in favor of `supplyPackCaps(tier) (std, dlx int)` — T1/T2: (2,1); T3: (3,1) unchanged; T4: (5,1); T5: (7,2). `SupplyPurchase.Validate` is now `Validate(tier ZoneTier)`; both call sites (`!expedition start`, `!resume`) pass the resolved zone's tier. The cap clears `DailyBurn(raw) × intendedDays × 1.3` for every tier under the §2 target durations even with harsh×3 layered on top, so a player who wants to buy for the long shape can. DailyBurn / harsh-multiplier / `phase5BDailyBurnRatePct` are unchanged here — they're a D7 lever, alongside the sim work that should unblock empirical baselining. The help text now describes the cap as "scales by zone tier"; the holiday +1 standard pack still bypasses the cap on purpose (it's a freebie on top).

> **Note:** the original "Empirical (sim-driven)" path was blocked: under D2-b event-anchored expeditions, `SimRunner.TickDay` calls `deliverBriefing` → `deliverBriefingEventAnchored`, which reads `time.Now().UTC()` for its safety-net check, so synthetic ticks never advance `CurrentDay` and `DaysAtEnd` stays at 0. Re-baselining off real day-counts requires teaching the sim to drive the event-anchored rollover (D7).

**D5-b (shipped 2026-05-27):** "Pick your loadout" preset prompt at launch. `!expedition start <zone>` with no pack arg now DMs a Lean / Balanced / Heavy menu (sized per zone tier via new `loadoutPurchase(tier, SupplyLoadout)` in `dnd_expedition_supplies.go`); player replies with `!expedition start <zone> lean|balanced|heavy` (short forms `l`/`b`/`h` also work). `!resume` got the same treatment. Raw `Ns Md` syntax remains the advanced override — `resolveLoadoutOrParse` in `dnd_expedition_cmd.go` tries a single-token preset first, falls back to `parseSupplyArgs`. Heavy always equals `supplyPackCaps` (the D5-a harsh×3 ceiling); Lean covers intended days at raw burn; Balanced sits between. `parseSupplyArgs("")` still returns 1 standard for any tier — it's no longer reachable from `!expedition start` (the empty path is intercepted to prompt), but the `1s` default is preserved so future callers don't get surprise zero-pack purchases. Help text updated; two existing scenario tests now pass `lean` explicitly, three new tests cover preset resolution + the prompt-vs-start guard.

**Remaining work (D5-c+):**
- Forage and Ranger forage re-baseline against the new caps.
- DailyBurn / `phase5BDailyBurnRatePct` revisit once D7 sim measures real elapsed-day counts.

**Exit criteria:** balanced loadout completes intended-duration expedition in ≥ 85% of sim runs without starvation extraction.

### D6 — Player-facing surface cleanup
- Help text rewrite — frame expedition as "supply pack + watch it play out".
- Hide / soft-deprecate `!camp <type>` from primary help, keep it documented as an override.
- `!fight` still in help, framed as override.
- Status DM (`!expedition status`) shows: current day / expected days, rooms walked / total, supplies, last 3 events.

### D7 — Sim + class re-baseline
- `cmd/expedition-sim` extensions: multi-day mode, autopilot camp, autopilot boss.
- Re-run the class win-rate corpus (cross-link `project_class_balance`, `project_j2_sim_artifact`).
- Re-check trailers (bard/cleric per `project_j1_post_sweep_findings`) — autopilot camp pacing may relieve their HP-cliff issue.
- New balance memory entry once corpus stabilizes.

## 4. Cross-cutting risks

- **Existing in-flight expeditions** at deploy time. Either fence by `start_date` (old expeditions finish under old rules) or write a one-shot migration that retunes their room counts. Lean: fence — these are typically <2 days.
- **The 24h zone-run inactivity timeout** (`dnd_zone_run.go:254`) — must be raised, or the autopilot must self-poke `last_action_at` per tick. With multi-day expeditions, a quiet stretch overnight is now expected.
- **Idle reaper** (commit `0f72484`, see `expedition_autorun.go` references) — already holds streak on autopilot; verify it tolerates 5-day expeditions.
- **Babysit perk** — `BabysitSafeRest` already upgrades Standard → Fortified at rest (`dnd_expedition_camp.go:241-245`); confirm interaction with autopilot picking Standard (no change needed; the upgrade fires inside `applyCampRest`).
- **Multi-region travel still unwired in autopilot** (memory `project_multiregion_travel_unwired`). For T4/T5 to actually take 5–7 days, autopilot **must** auto-advance regions on region-clear. D2/D3 covers this implicitly but call it out — likely a small dedicated commit.
- **Pet arrival** (memory `project_pet_event_reachability`) — emergence seam is at run-complete; longer expedition just delays the roll, no code change needed.
- **Twinbee voice + secret NPC buffs** (memories `feedback_twinbee_voice`, `feedback_npc_buffs_are_secret`) — none of the new copy can break these.

## 5. Schema impact

Likely none. Existing columns cover it:
- `current_day` → autopilot increments via `advanceExpeditionDay`.
- `camp_json`, `supplies_json`, `region_state`, `threat_*` already in place.
- One **possible** new column: `expected_days` int (set at launch from zone tier + loadout) so status/end-of-day DMs can render "Day 3/5". Cheaper to derive than store — defer.

## 6. Open questions for next session

1. ~~Day-advance semantics~~ — **decided 2026-05-27: event-anchored.** day++ fires on autopilot camp-pitch (autopilot night-camp = sleep = day burn + briefing). UTC clock becomes a re-engagement-DM anchor only, not a state mutator. See §3-D2 for implementation impact.
2. **Boss autopilot default**: always-on, opt-out, or opt-in? My lean: always-on with a per-expedition opt-out flag; revisit after D3 reads.
3. **Per-tier room counts**: the table in §2 is a starting point. Worth a sim pass before D1 commits — or fine to ship a guess and refine in D7?
4. **Extra anchor rooms**: T4/T5 with one elite + one trap may feel thin spread over 30+ rooms. Add a second elite mid-zone, or rely on multi-region + patrols + threat for variety?
5. **Mobile/AFK friendliness**: should the autopilot-DM cadence aim for "1 DM per real-time evening" regardless of room count, or "1 DM per in-game day" (which could be hours apart if the player set the pacing fast)?

## 7. Sequence

D1 → D2 → D3 (independent of D4/D5) → D5 → D4 → D6 → D7.

D2 and D3 can land in either order; D4 needs the volume problem to be real (so after D2/D3 at minimum). D5 wants D1's new lengths but can run in parallel with D2/D3. D7 always last.

Estimated sessions: 5–7. D2 and D5 are the biggest single-session chunks.