Design History

The balance arc was a nine-version push from “Treasury crushes everything” to “three viable economic paths.” Every change was driven by deterministic simulation evidence — 100-game or 400-game batches against fixed seed sequences — not by guesswork.

The two artifact streams

Sovereign exists as two artifacts:

1. The board game (v0.2 baseline) — a printable 34-sheet edition with the original Hamilton-system mechanics.
2. The solo / digital adaptation — a self-contained HTML simulator that runs the same rules locally with deterministic AI opponents.

The board game stays at v0.2 until human-table playtest evidence exists. The digital mode is at v0.10. The digital mode runs ahead because simulation is cheaper than reprinting.

Phase-by-phase build (Solo / Digital)

Phase	What it added	Audit
Phase 1	Concept doc: state model, 8 surfaces, 3 solo modes, 4 opponent profiles spec	—
Phase 2	Static clickable prototype proving the 8 surfaces and the 11-step turn loop	—
Phase 3	Local state machine: real reducer, full 7-lap game, all 40 spaces, all 24 cards, all 7 Acts, scoring from rules	3 bugs found + fixed
Phase 4	Scripted opponents: players[] array, two MVP profiles, auction mechanic, multiplayer rent, vote tallying	6 bugs found + fixed; Act-vote orchestration maintenance patch
Phase 5	Narration library, save / load with hash integrity, replay scrubber	PASS, no fixes
Phase 6	Local balance telemetry, batch simulation, Manufacturer profile	1 bug found + fixed (card-choice phase leak)
Phase 6.1	Telemetry hygiene: slot-indexed attribution, rentNet split, asset bucket cleanup	byte-identical to Phase 6

Every phase is fully additive: the reducer body, action set, scoring formula, opponent decision functions, and v0.2 anchors all remain byte-identical from one phase to the next. Phase 7 is “no new code, just data” — the architecture supports new telemetry without touching gameplay.

Balance arc

Version	Change	CANONICAL split (T / Mc / Mfg)
v0.2	Board-game baseline	81 / 18 / 1 — Treasury runaway
v0.3	NF Credit nerf · Manufactures capital event · Capacity ≥ 8 payment +50%	76 / 22 / 2 — Mfg cash-starved
v0.4	Capacity ≥ 10 milestone · industrial set bonuses	73 / 21 / 6 — Mfg gets scoring potential but Capacity track stays cold
v0.5	Capacity +1 on first industrial purchase	73 / 21 / 6 — Capacity moves in MFG-MIRROR, not CANONICAL
v0.6	Manufacturer set-completion gate relaxed	73 / 22 / 5 — profile fix didn’t move the needle
v0.7	Telemetry only. Acquisition funnel diagnostic	Same — but reveals slot 2 lands on industrial spaces less due to dice geometry
v0.8	Manufacturer Industrial Charter (starting Textile Works at setup)	68 / 21 / 11 — Mfg enters the 10–25% target band for the first time
v0.9	Telemetry only. Scoring decomposition + counterfactual	Same — but reveals cashIP is what’s keeping Treasury at 68%
v0.10	Cash scoring 1 IP per 200 TN → 1 IP per 400 TN	59 / 25 / 16 — all three profiles in target band · FROZEN

What the evidence taught

1. Don’t trust the obvious cause. v0.6’s hypothesis — that Manufacturer’s set-completion gate was suppressing buys — didn’t survive evidence. The gate was real, but not load-bearing. Real cause was slot-geometry (v0.7 diagnostic).

2. Telemetry hygiene matters. Phase 6’s first batch reported industrial buys per player: Treasury 0.65 / Merchant 0.59 / Manufacturer 0.40. That was a measurement artifact — name-keyed attribution failed in mirror batches. Phase 6.1 fixed it. Slot-indexed attribution revealed Mfg captures all the industrial assets it lands on; the limit is reaching them.

3. The right lever is sometimes a level lower. v0.4 added scoring rules for Capacity ≥ 8 and ≥ 10 thinking that would tip Mfg over. It didn’t — because the Capacity track wasn’t reaching those thresholds. The actual fix was v0.5 (Capacity +1 on purchase), v0.8 (Industrial Charter), and a track that climbs as industry is bought.

4. Treasury was never about NF Credit. The v0.3 NF Credit nerf was philosophically clean but mechanically minor. The v0.9 counterfactual showed cashIP (the generic “cash held” Influence source) was the load-bearing line for Treasury, not the Hamilton-thesis scoring. v0.10 nerfed cashIP and Treasury fell into band.

5. Don’t chase 33 / 33 / 33. Treasury is supposed to be the strongest profile. That matches the historical thesis: public credit + federal finance were Hamilton’s dominant lever. The target was Treasury 45–65, not Treasury 33. Treasury at 59% with two other viable paths is the win condition for the design.

Hard invariants preserved across all 9 versions

• 17-action reducer shape (Phase 3’s 16 + Phase 4’s AUCTION_BID).
• mulberry32(state.rngSeed) as sole RNG.
• Same seed + same human decisions = byte-identical ledger.
• No fetch, no XMLHttpRequest, no WebSocket, no sendBeacon, no postMessage to external origins.
• route ladder 25 / 50 / 100 / 150, Coinage Act effect, Revolutionary Debt bases (Continental 4 / Soldier Pay 6), Capacity ≥ 6 payment threshold (+25%) — all locked.

What’s deferred

• Failure events (Default, Rebellion, Bankruptcy). Fired 0 / 400 times in the v0.10 evidence pass. Currently decorative. May be revisited if the design ever pushes into a v0.5 fail-state pressure arc.
• Opportunist / Cash profile (the 4th concept-doc profile). Locked v0.10 set is 3 profiles.
• Human playtest. Balance is simulation-tested only. Strategic deviation by a human against scripted opponents may shift these rates.

Next

• Security — threat model and data handling.