feat(sim): make subject rebellions possible for war-subjugated vassals

[csheldrick]

Summary

• War losers forced into subjugation now start at loyalty 0.30 instead of 0.55, reflecting that they were defeated, not willing
• While loyalty is below 0.45, the power-imbalance stability bonus is replaced by a resentment penalty (−0.0005/tick), creating a realistic window for rebellion within ~100–400 ticks
• Voluntary protectorates (maybeVoluntaryProtectorate) retain the original 0.55 starting loyalty and positive drift

Problem

The subject rebellion mechanic was structurally dead. The loyalty drift formula was always net-positive for vassals (~+0.00014/tick), so loyalty never fell below the 0.25 rebellion threshold. Result: 0 rebellions in 10,000 ticks across 122 subject relations created. Subjects were decorative — they only ended by overlord collapse (liberation), never by their own action.

Fix

Two small changes:

src/sim/Subjects.ts — add optional startLoyalty parameter to createSubjectRelation; add resentment drift when loyalty < 0.45 (suppresses the power-bonus, applies −0.0005/tick grudge instead).

src/sim/Diplomacy.ts — pass startLoyalty: 0.30 in maybeSubjugateLoser (the war-defeat path only).

Verification

npm run report                         # default 3000-tick run
npm run report -- --milestones 1000,3000,10000 --no-determinism

Before: Subjects: … 0 rebellions at every milestone through tick 10 000
After: Subjects: … 5 rebellions at tick 3 000 · 12 rebellions at tick 10 000 (10% of 120 created)

Determinism check passes unchanged.

Dogfood receipt

This PR was produced under Noet dogfood protocol (PR270 post-merge run).

• Initial slice: slice-2026-07-03-150114
• Initial allowed surface: src/sim/Subjects.ts, src/sim/Diplomacy.ts
• New files: (none)
• SERs created: 1 (1 accepted — verification scope expansion, one-shot via noet expand)
• Final slice: slice-2026-07-03-150121
• Files changed: src/sim/Subjects.ts, src/sim/Diplomacy.ts
• Tests: npm run report — determinism passed, 12 rebellions observed at tick 10 000 (manually bound; external-repo verification binding is a known gap — see OBS-2)
• Surface gate result: ✓ PASS
• Hash valid: ✓ yes
• Observations: 2
  • OBS-1: PR270 expand felt lighter — SER filed and accepted in one shot with no clean-tree guard interruption (protocol-win)
  • OBS-2: External-repo verification binding gap — noet verify run cannot bind evidence for a command that lives in the target repo, not the noet repo; finish correctly identifies it as declared-only rather than silently passing (protocol-gap)

---

Generated by Claude Code