DR-042 Phase A: expedition-driven win — move win-driver off base-siege survival, kill the AFK path

Design-review-gated (wf_ebef4e81-dba, GREEN-WITH-CHANGES). The win-driver moves
from "survive N base sieges" to "clear N expeditions". The review overturned the
literal plan: credit on RETURN, not at the clear edge (clear-edge crediting arms
the undefended final base siege -> uncontestable terminal Loss).

- ExpeditionGateSystem: now the sole production writer of GoalProgress.Charge —
  a clamped +1 per cleared expedition folded into the existing once-per-epoch
  reward block, reusing the LastRewardedEpoch latch (Ore + Charge share fate) +
  a SaveRequest checkpoint. No new latch, no new GhostField, no ordering change.
- CyclePhaseSystem: deleted the survived-siege +1 (the AFK win path). Victory
  latch unchanged; GoalReached still arms the final base siege at cap.
- CycleDirectorAuthoring + CycleDirector.prefab: ScheduleEnabled baked OFF
  (retaliation-only). A serialized prefab bool ignores the C# field initializer,
  so the value is flipped in the prefab, not just the code default.
- Tests: re-pointed CyclePhaseSystemTests + EndgameWinLoseTests survived-siege
  assertions; extended ExpeditionGateRewardTests (+1, no-double-credit, clamp).
  389/389 EditMode green; clean netcode Play smoke (no sort-cycle, Schedule=0).

SaveData stays v5. Docs: DR-042 build record + forks resolved, CLAUDE.md
base-loop line, Backlog (A done).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Docs/Vault/06_Roadmap/Backlog.md

@@ -29,7 +29,7 @@ Last decluttered 2026-06-08 (removed all shipped `[x]` items; their context is p
 
 A 5-subsystem loop evaluation found the loop has **two conflicting win-models bolted together**: the only path to victory is "survive 4 base sieges" (passively/AFK-reachable — scheduled sieges auto-arm + a 60 s timeout auto-clears them), while the **expedition (the stated combat spine, where all the new enemy variety lives) advances nothing toward winning.** Root cause: the END-1/END-2 base-siege win is a leftover from the superseded jam slice ([[DR-035_End_Of_Month_Slice_Adoption]]/[[DR-036_END2_Final_Siege_Win_Lose]]) that DR-037 never retired. **Operator chose (2026-06-24): commit to the expedition-driven vision.** Build order:
 
-- **A — Coherence core (first):** move the win-driver from *base sieges* → *expedition clears* (`GoalProgress.Charge` +1 on a cleared sortie); **kill the AFK win** (disable the blind scheduled siege as a progression source); final beat reached through the spine. *Netcode-touching → design-review first.*
+- **A — Coherence core — ✅ BUILT + validated (2026-06-25):** win-driver moved *base sieges → expedition clears* — `ExpeditionGateSystem` is now the sole production writer of `GoalProgress.Charge` (+1/clear, credited **on RETURN** — the review overturned credit-on-clear, which would arm the undefended final siege → uncontestable Loss; credit-on-return keeps the player home). **AFK win killed** (`ScheduleEnabled` baked OFF, retaliation-only; survived sieges no longer credit). Final beat = the END-2 final **base** siege (operator-locked). 389/389 EditMode + clean Play smoke (no sort-cycle, `ScheduleEnabled=0`). Design-review-gated (`wf_ebef4e81-dba`). Fun-gate playtest pending. *Next: **C**.*
 - **B — Retaliation connect:** post-expedition siege becomes THE base-siege source (fix the interference) — defending what you built becomes a *consequence* of sortieing, not the goal. *Netcode-touching → design-review.*
 - **C — Legibility fixes:** walls actually block (structures on the enemy collision filter); Aether-upgrade HUD button + cost; Biomass sink (or cut); cold-start ledger seed; hide dead Harvester/Conveyor/Pylon; expedition objective UI + gate prompt; reward scales with depth.
 - **D — Persistent meta (= Slice 4):** SaveData v6 + between-runs growth (above).

Docs/Vault/07_Sessions/_Decisions/DR-042_Loop_Reshape_Expedition_Driven.md

@@ -75,10 +75,22 @@ The game has **two disconnected win-models that fight each other**:
 - **[[DR-037_Procedural_Expedition_Spine_Two_Classes_Persistent_Meta]]:** this DR is the concrete loop the redirect implied. Slice 4 (Persistent Meta) is **re-framed**: the loop-coherence work (A–C) comes first; the meta/SaveData-v6 (D) is the final phase.
 - **Roadmap:** [[Path_to_Fun]] (Path A/B) was already historical post-DR-037; this is the current operative loop plan. The committed list is in [[Backlog]] under the Co-op Roguelite Redirect.
 
-## Open forks (operator, lock before/at the design review)
-- **Final beat:** a **final expedition** (the climax IS a deep sortie — purest expression of "expedition = spine") vs a **final base siege** (a defense climax, reuses END-2's final-siege machinery). *Recommend present both at the A-phase review.*
-- **Do base sieges stay in v1 at all,** or does retaliation become a later layer (B) so the first coherent build is purely sortie→reward→escalate? *(The recommendation keeps a light retaliation siege so the build pillar isn't orphaned.)*
-- Expedition reward shape + depth scaling; whether a soft-loss should cost `GoalProgress` (give the run real downside).
+## Open forks — RESOLVED (operator, 2026-06-25)
+- **Final beat:** **final base siege** (defense climax, reuses END-2's final-siege machinery) — *operator-chosen*. With credit-on-RETURN the capping return arms the climactic siege while the player is freshly home, so the build pillar pays off at the finish. (A *final expedition* was the alternative; deferred.)
+- **Base sieges stay in v1** as **post-expedition retaliation only** (the build pillar isn't orphaned). The blind scheduled source is disabled.
+- Expedition reward shape + depth scaling, and whether a soft-loss costs `GoalProgress` — deferred to phase C/tuning.
 
 ## Status
-Accepted + locked (direction). Build pending — phase A first, design-review-gated. Loop evaluation transcript: `wf_4cebbc74-216`. Combat substrate it gives purpose to: [[DR-041_Slice_Combat_Depth_Enemy_Variety_Impact]].
+**Phase A — BUILT + validated (2026-06-25).** Design-review-gated (`wf_ebef4e81-dba`, GREEN-WITH-CHANGES). Loop evaluation transcript: `wf_4cebbc74-216`. Combat substrate it gives purpose to: [[DR-041_Slice_Combat_Depth_Enemy_Variety_Impact]].
+
+### Phase A build record — what shipped (and the design correction)
+The review overturned the literal A1 sketch. **The win credit fires on the player's RETURN, not at the expedition-clear edge.** Crediting at the clear edge (while the player is still out in the expedition) would arm the climactic final *base* siege with nobody home — and `SiegeTimeout` is disabled in the final — so the undefended Core would breach into an **uncontestable terminal Loss**. Credit-on-return guarantees the player is teleported home the same tick the final arms.
+
+The implementation **relocates** the single writer of `GoalProgress.Charge` (it does not add a second writer):
+- **`ExpeditionGateSystem`** is now the sole *production* writer of `GoalProgress.Charge`: folded a clamped `+1` (`math.min(Charge+1, Target)`) + a `SaveRequest` checkpoint into the **existing** once-per-epoch reward block, reusing the same `LastRewardedEpoch` latch that gates the Ore reward (Ore + Charge share fate; co-op/re-entry de-dup is free). No new latch field, **no new `[GhostField]`**, no ghost re-hash. The credit is guarded independently of the ledger so it still lands in ledger-less worlds.
+- **`CyclePhaseSystem`** no longer credits Charge on a survived base siege (the AFK win path deleted); the final-siege Victory latch is unchanged. `GoalReachedSystem` still arms the climactic final siege at `Charge>=Target`.
+- **`ScheduleEnabled` baked OFF** (both the `CycleDirectorAuthoring` code default *and* the serialized `CycleDirector.prefab` value — the code default alone does not flip an already-serialized prefab field; Play-verified `ScheduleEnabled=0` at runtime). The schedule code path is kept as a config-inert reserved hook; `ScheduleSizePerWave` left non-zero (the final-siege size formula still reads it). Retaliation (`PostExpeditionEnabled`) stays on.
+- **No system-ordering change** → no sort-cycle risk (Gate is already `[UpdateBefore(CyclePhaseSystem)]`, so the credit lands before `GoalReachedSystem` reads the edge same-tick). SaveData stays **v5**; legacy v5 saves load as-is (their old siege-era Charge reads as clears — accepted for single-slot dev saves).
+- **Validation:** 389/389 EditMode (re-pointed `CyclePhaseSystemTests` + `EndgameWinLoseTests` survived-siege assertions; extended `ExpeditionGateRewardTests` with the +1, no-double-credit, and clamp cases) + a clean netcode Play smoke (world boots, no sort-cycle, `ScheduleEnabled=0`, no exceptions). **Fun-gate playtest still pending** (the base reads as inert until you walk to the gate — the gate-prompt UI is phase C, accepted caveat).
+
+### Next phases (unchanged order): C (legibility) → B (retaliation polish) → D (Slice 4 meta, SaveData v6).