# After H2: deal influx, end conditions, sequence salience **Date:** 2026-08-08 **Context:** H2 measured (RPT-0005) — scoping works; felt “more interesting.” **Status:** design discussion — not yet a packaged variant Maintainer questions: (1) fixed problem deal vs random influx; (2) fixed 5 rounds vs clear-board / stressed-out ends; (3) GROUND/DARVO as sequences with significant effect — sim artifact or gameplay gap? --- ## 0. What H2 already proved - Scoped stress is the right *pressure* shape (all-global control = H1 collapse). - DARVO can arm for a seat that still sometimes wins (unlike H1). - ATTACK is affordable under H2, still not profitable. - Bond joint-SOLVE incentive is **untested by bots** (they do not model scopes). H2 does **not** yet answer deal variety, pacing, or “do sequences *feel* significant.” Those are the next design layer. --- ## 1. Problem deal: start set + random influx ### Current Fixed deal at setup: Surface + hidden 1..k by seats. Owners assigned once. Solution deck is the only ongoing draw (INVESTIGATE draws Solutions). ### Proposal Keep a **starting set** of Problems, then **random draws** that can add personal / bond / global Problems to play (or into a hand that can be played/attached). ### Strengths - Replay variety without new scenarios every time. - Fiction: issues *arrive* mid-relationship; hand/draw = what lands on you. - Difficulty dial = composition of the **draw bag** (how many personal vs bond vs global), cleaner than fixed priority mapping alone. - Supports “my hand gave *me* a personal problem” without full EXT_PERSONAL shadow goals. ### Challenges | Risk | Mitigation | |------|------------| | Available points / threshold float mid-game | Threshold as % of *currently in play* claimed target, or score only starting set, or drawn problems are stress-only (no point value) | | Global flood rebuilds H1 | Cap globals in bag; max N open globals | | Too long / never clears | Soft cap on open Problems; draw only on triggers (missed SOLVE, DENY, Round End if board empty of hidden) | | Ownership chaos | Drawn personal → drawer is owner; bond → drawer’s network; global → none | | Teach complexity | Phase 1: **draw adds to board face-down**, not a second hand of problems | ### Lean experiment shape (if packaged later) **H3-deal (sketch):** After setup of Surface + 1–2 starters, remaining scoped Problems form a **Pressure deck**. At End of round (or on INVESTIGATE miss), draw 0–1 if open unclaimed count < cap. Drawn card enters play with scope as printed; owner = drawer or Lead-round-robin. Do **not** put Problems in the Solution hand first — that collides with suit management. Separate “issue arrives” channel is clearer. --- ## 2. End conditions: beyond fixed 5 rounds ### Current Always 5 rounds, then threshold + mode scoring. Compact (5–15 min) was explicit intent (2026-07-30 exploration). ### Why 5 feels arbitrary It is a **timer**, not a story beat. Under H2 stress, a clear board early still forces empty rounds; a meltdown at round 3 still limps to 5. ### Proposal family | End | Meaning | Risk | |-----|---------|------| | **A. Board clear** | No unclaimed Problems left → immediate group success check | Short games; INVESTIGATE race | | **B. Group break** | All seats at Stress 5 and/or in DARVO → group fail (or “stressed out”) | Need precise definition; soft lock if one seat holds out | | **C. Max rounds** | Still a ceiling (4–6) so games terminate | Keeps product length | | **Hybrid** | End on A or B, else at round max | Best of both | ### Recommendation **Hybrid** is the least arbitrary and keeps the product box promise: 1. If **no unclaimed Problems** remain after a round’s Solve → end; apply mode scoring (threshold may be automatic success if all claimed value ≥ threshold, or redefine success as “cleared the table”). 2. If **collapse**: e.g. every seat is at Stress ≥4 **or** ≥ half the table is in active DARVO → end as **group failure** (SHARED) / no personal win (semi). 3. Else after **Round Max** (keep 5 or try 6) → current threshold scoring. “All stressed out” as pure unanimous Stress 5 is harsh and rare; prefer **majority in DARVO** or **no seat below Stress 4** as the fail line. Threshold vs clear-board: if success = clear all, thresholds become redundant; if success = clear *enough*, keep threshold and treat early clear as auto-meet if points ≥ threshold. --- ## 3. GROUND and DARVO as sequences — sim vs design ### What the rules actually are | System | In r0 today | “Sequence”? | |--------|-------------|-------------| | **DARVO** | DENY → ATTACK → REVERSE over **consecutive rounds** once armed at Stress 5 | **Yes** — three mandatory stages | | **GROUND** | One action, **one of three modes** (GR / OU / ND) chosen after reveal | **No** — not a multi-round sequence | | INTENT / early design | GROUND as G-R-O-U-N-D practices; DARVO as binding multi-turn chain | Aspirational / partial | So if the desire is “GROUND is also a sequence,” that is **missing design**, not a clay-borg bug. ### Why sequences feel weak in the simulator **Mostly gameplay + policy, not “sim is broken.”** 1. **Competent bots regulate.** Greedy GROUNDs when gated → Stress often never hits 5 → **no DARVO stage chain to watch**. H2 reactive shows arms; greedy play will not. 2. **Full DARVO chain needs three rounds.** With max 5, a late arm gets DENY only (or DENY+ATTACK). REVERSE’s drama is rare. 3. **Exits are strong.** Bond Support cancels the stage and ends the sequence; GROUND—GR lets current stage fire then ends the rest. So sequences are designed to be **interruptible** — good for teaching, bad for “unstoppable set piece” feel. 4. **GROUND modes are situational.** GR (−2 Stress) is the default bot use. OU/ND only matter if DENY/Attack/Blame already happened — rare under peaceful greedy play. Sim never “chooses the dramatic mode.” 5. **Bots optimize scores, not narrative.** Sequence significance is a human-felt property; aggregates report arms and wins, not “that REVERSE mattered.” **Where the sim *is* limited:** no multi-seat-aware bond motivation; no felt-play; watching greedy H2 underplays DARVO by construction. ### If we want sequences to *matter* more (gameplay levers) | Lever | Effect | |-------|--------| | Longer max rounds or early arm | More full DENY→…→REVERSE completions | | Softer exits (Bond Support cancels only *this* stage, not whole sequence) | Harder to snuff DARVO | | DARVO stages alter scoring more (Blame already −1; DENY already blocks points) | Make completion costly for the *group* | | GROUND as optional **multi-step practice** (e.g. commit GR this round, may chain OU next if still regulating) | True GROUND sequence — large design change | | Scenario “Focus cards” that force a stage | Guaranteed set piece for teaching | Prefer measuring **full-sequence completion rate** (DENY+ATTACK+REVERSE all fired) under H2 before changing exits — if completion ≈ 0 even when arms > 0, the chain is too long for the clock. --- ## 4. Suggested order (do not stack into one silent H2 edit) 1. **Keep H2 stress routing** (`keep-as-experiment`; candidate for r1 core). 2. **Table smoke** for bond joint SOLVE (criterion 4). 3. Package **one** next experiment, not three: - either **H3-end** hybrid end conditions on top of H2, **or** - **H3-deal** pressure-deck influx on top of H2, - measure sequence completion rate as a side metric either way. 4. Only then consider GROUND-as-sequence or DARVO exit tuning. Promoting H2 into baseline before deal/end redesign is optional; stress routing can be kept even if deal changes later.