codex
b5730f280b
Records absent from central carried random pre-ADR-007 identifiers minted by the retired local hub, which C-06 refused as stale references. Deriving from the canonical record id takes no identity from anything. Refs CUST-WP-0068-T06 Assistant: claude-code Assistant-Model: opus Assistant-Process: 2583210@bnt-lap001 Assistant-Session: f2bff2d5-e9b2-4338-92ca-10282a927006
3.4 KiB
Mode playtest notes — engine (GR-E03 / GR-E04)
Date: 2026-08-08
Workplan: GROUND-WP-0003
Source: GROUND-RPT-0003 (clay-borg attack-value panel + full-mode playouts)
Variant: ground-darvo-r0 (baseline)
These notes close the engine half of WP-0003 (“human table or full engine playout”). They are not a substitute for human felt-play if we want table texture later; they do answer “do the modes score to the end, and what does ATTACK do?”
Runs covered
| Task | Mode | Seats | Evidence |
|---|---|---|---|
| T01 | MODE_SEMI | 2p | attack-value cells; modes playable to scoring |
| T02 | MODE_SEMI | 3–4p (and 6p) | same panel |
| T03 | MODE_COALITION | 2p | same |
| T04 | MODE_COALITION | 3–4p (and 6p) | same |
| — | MODE_COOP | all | control cells |
Identical-play differentiation (seed 1, 4p, 37 commands):
| Mode | Winners |
|---|---|
| SHARED GROUND | all four (mastery 4) |
| COMMON PROBLEM (semi) | P3 (top personal) |
| BONDED COALITIONS | P1+P2 (best Bond network) |
Encoding of Modes.csv behaves as three distinct games.
Driver note (clay-borg): GR-E03/E04 were previously unplayed because
cb-play forced SHARED GROUND — fixed on their side; not a modes defect.
Outcomes checklist
Shared total vs threshold / personal scores
- Group success rates under “never attack” competent-ish policy (wins per
200 games, 2/3/4/6 seats):
- SHARED: 132 / 165 / 190 / 200
- SEMI (seat-0 among winners framing differs; personal race present): lower individual “among winners” counts
- COALITION: high “among winners” when not attacking (network still intact)
- Exact seed-by-seed personal score tables were not shipped in RPT-0003; mode differentiation and ATTACK value are the load-bearing results.
Stress / DARVO
- Under competent non-attacking play on baseline r0: DARVO armed 0 times in the RPT-0003 competent aggregate (500 games cited).
- Later CB-EV-0031: peak Stress held never exceeds start 2 when policies never select ATTACK — Stress only falls. Inbound pressure is ATTACK-only for those policies.
ATTACK value (all three modes)
| mode | never | sometimes | always |
|---|---|---|---|
| SHARED | 132/165/190/200 | identical | 0 |
| SEMI | 59/52/48/44 | 59/52/48/34 | 0 |
| COALITION | 131/134/132/116 | 59/52/48/34 | 0 |
(2/3/4/6 seats.) ATTACK never improves outcomes; in coalitions it roughly halves “among winners.” E04 attacking numbers match E03 when Bonds break (GR-A07 + network scoring).
Underdetermined rules exposed?
- No new U-items from mode scoring itself.
- Design gap (not rules ambiguity): ATTACK/DARVO irrelevant under competent baseline play — tracked experimentally via WP-0006 (H1, reject) → WP-0007 (H2 scoped stress, in flight).
Follow-ups
| Item | Where |
|---|---|
| Restore inbound Stress without flat table tax | H2 h2-scoped-problem-stress |
| Human felt-play of semi/coalition (optional) | residual note below; not blocking |
| Opening-action economy (INVESTIGATE first) | clay-borg offer; deferred |
Residual (optional, not a new workplan unless scheduled): one human 2p semi and one 3–4p coalition session for “feels like one game” texture. Engine checklist for T01–T04 is satisfied.