DUEL_DESIGN v1.3: balance v1 = secondPlayerCitations 1 + turnCap 40; Location rider rejected (#64)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
2026-11-18 00:03:38 -05:00
co-authored by Claude Opus 5.5
parent 6e97e2c8ee
commit 5121fd0418
+4 -4
View File
@@ -1,6 +1,6 @@
# WikiDeck Duel: design v1 # WikiDeck Duel: design v1
**Status:** v1.2, Night 2 (2026-11-17): v1 plus night-review's nits on PR #66, duel-cards' implementation choices from PR #70, and the balance v1 second-player rule (#64). Owner: duel-design (issue #60). **Status:** v1.3, Night 2 (2026-11-17): v1 plus night-review's nits on PR #66, duel-cards' implementation choices from PR #70, and the balance v1 rules `secondPlayerCitations` 1 and `turnCap` 40 (#64). Owner: duel-design (issue #60).
**Base:** `docs/night2/BRIEF.md` (Eric's locked decisions). Where this file and BRIEF disagree, BRIEF wins. Reviews used: designer (rules), data §2–4 (derivation), critic S7 (tone). **Base:** `docs/night2/BRIEF.md` (Eric's locked decisions). Where this file and BRIEF disagree, BRIEF wins. Reviews used: designer (rules), data §2–4 (derivation), critic S7 (tone).
**Consumers:** duel-core (`shared/duel/` engine and `types.ts`), duel-cards (`classify/derive/effects/text.ts`), sim-balance (bots, `DUEL_BALANCE.md`), ui-duel / ui-pvp (faces, recap), host-pvp (rulesets). **Consumers:** duel-core (`shared/duel/` engine and `types.ts`), duel-cards (`classify/derive/effects/text.ts`), sim-balance (bots, `DUEL_BALANCE.md`), ui-duel / ui-pvp (faces, recap), host-pvp (rulesets).
@@ -63,8 +63,8 @@ Classification order (duel-cards owns it): P31 → P279 anchor classes → P31 l
### 2.6 Clock and length ### 2.6 Clock and length
- `turnSeconds` **60** per turn plus `bankSeconds` **300** per player. The turn timer runs first; when it expires, the player's bank drains; when the bank is also empty, the turn passes (`timeout` action). The bank does not refill. Three timeouts in a row concede. - `turnSeconds` **60** per turn plus `bankSeconds` **300** per player. The turn timer runs first; when it expires, the player's bank drains; when the bank is also empty, the turn passes (`timeout` action). The bank does not refill. Three timeouts in a row concede.
- No turn cap by default. If the sim shows stalls, add `turnCap` (suggested 20 turns each; at the cap the higher Notability wins, then the higher total power in play, then the second player). - **`turnCap` 40** (20 turns each, balance v1). At the cap the higher Notability wins, then the higher total power in play, then the second player. In sims it decides about 4% of games and removes every game past 20 turns each, without moving the median.
- **Sim targets** (agreed with sim-balance, #64): median 8–12 turns each; first-player win rate 48–54%; < 5% of games past 20 turns each. Balance v1 baseline with Doctrines working: median 8, p90 12, 3.4% past 20, so no `turnCap` is needed. - **Sim targets** (agreed with sim-balance, #64): median 8–12 turns each; first-player win rate 48–54%; < 5% of games past 20 turns each. **Balance v1 result** (`secondPlayerCitations` 1 + `turnCap` 40; 3,000 heuristic-mirror matches, engine 4623930, duel-cards 067de83; run by ui-duel for sim-balance, #64): first player **54.7%** [53–57], turns each 6 / 8 / 12, 0% past 20, about 8 minutes median.
### 2.7 Determinism ### 2.7 Determinism
All randomness (shuffles, coin flip, template variants) comes from the match seed via `shared/game/rng.ts`. The same seed and actions always replay to the same state, log and recap. All randomness (shuffles, coin flip, template variants) comes from the match seed via `shared/game/rng.ts`. The same seed and actions always replay to the same state, log and recap.
@@ -276,4 +276,4 @@ Sample log:
## 11. Open questions for the sim (not blockers) ## 11. Open questions for the sim (not blockers)
1. Disaster T4 at k−1 may be weak against 40-card boards; try k. 1. Disaster T4 at k−1 may be weak against 40-card boards; try k.
2. Stub at +2 counters on cheap bodies may be too strong (33% of Figures); try max 1. 2. Stub at +2 counters on cheap bodies may be too strong (33% of Figures); try max 1.
3. **Locations (balance v1 finding):** Location-heavy decks win ~20% and Locations are the weakest cards. `territoryCap` 3, `locationsPerTurn` 2 and `garrisonStep` 15 move them ±3 points at most. Proposal under test: T2 gains "When this enters, draw a card" (a plain `onEnter` + `draw` effect, no condition), so a Location replaces itself. 3. **Locations (balance v1 finding):** Location-heavy decks win ~20% and Locations are the weakest cards. `territoryCap` 3, `locationsPerTurn` 2 and `garrisonStep` 15 move them ±3 points at most. Tested and **rejected for v1**: a T2 rider "When this enters, draw a card". It lifts the Location card (47 → 52%) but not Location-heavy decks (18 → 17%), pushes the first player back to 58.4%, and brings back library-out losses. Open for v1.1: a lever that helps Location-heavy decks without extra card flow, e.g. Territory that scales (+1 Citation per two Locations) or Garrison Locations that can block Featured.