ADR-0009 — Turn structure: plan-months, two decision speeds, day as resolution tick

• Status: ACCEPTED
• Date: 2026-07-12

Context

Workshop #1's intent was a week-based planning loop; the build drifted to day-turns with the week surviving only as display (game_state.gd TURNS_PER_WEEK, "Day 3/5" HUD). The drift's real shape: decisions attached to the resolution tick. Evidence forcing the ruling: exploit-finder sweep (loan terms absurd at day cadence; only safety-spam beats passive), and Pip's playtest notes (repetitive event loop, weirdly slow pacing, instantaneous rewards from fundraising breaking time-feel, salary cadence off).

Pip fixed the fiction anchors this session: 2017 start; ~2027 early-competent loss, 2033 decent, 2037 hard, 2040 ≈ the 1% run. Wall-clock ruling: a decent run is a Civ/X-com-campaign 6–8 hours (Pip: "60% 6–8 hours, 30% evening, 10% other"). Grain arithmetic against those anchors: week ≈ 500+ turns to an early loss (dead); month ≈ 114 / 186 / 234 / 270 turns; quarter ≈ 38 / 62 / 78 / 90.

Decision

1. The turn is a month (decision cadence — a plan LAYER over day-grain sim ticks, not a re-grain; see clarification below). Fixed grain, v1. Plan phase: allocate staff and founder hours, queue strategic actions, explicitly set reserve (slack). Implementation clarification (2026-07-14, PR #636): "the turn is a month" means the decision cadence — implemented as a plan layer over day-grain simulation ticks, not a literal re-grain of the sim step. Point 2 already requires this (days remain "sim substrate, calendar, animation, score resolution"), and the coarse-hands/ fine-scoreboard principle depends on day-grain scoring surviving. Recorded because the headline wording was misread as a hard re-grain during L1 build.
2. The day is demoted to resolution tick — sim substrate, calendar, animation, score resolution. Guard rule: no mechanic may hang a routine decision on the day tick. Days are for physics and comedy.
3. Two decision speeds (MtG taxonomy, none of its machinery): offers a costed menu: HANDLE from reserve (painless — what insurance was for) · HANDLE by cannibalizing (delay/kill planned WIP) · DEFER (mints a Liability Ledger entry with named terms) · IGNORE (stated immediate consequence at list price). Defer/ignore pricing is the ledger beat's to detail (ADR-0010, this workshop).
   • Plan speed ("sorcery") — only castable at the month boundary.
   • Response windows ("instant") — an event fires mid-month and, if a window opens,
4. Reserve is crisp. Unspent slack evaporates at month end. No banking ever (Pip: "I don't think we can let players bank time"). Overcommitting is a legal gamble.
5. Strategic actions have durations. Nothing strategic resolves instantaneously; effects land mid-period or at the month review (fundraising is a campaign, hiring is a search → offer-interrupt → onboarding drag).
6. The badge is the date. Public score = the exact calendar day the run died ("I made it to March 2034"); internals unchanged — lexicographic (days survived, doom-integral) per ADR-0002. Ladder texture Pip named: players overtake each other by days while decisions lock in months — coarse hands, fine scoreboard.

Beacons served / violated

• Rams #10 (as little design as possible): one cadence ruling folds the loan-term absurdity, salary cadence, event pacing, and banking (deleted) — and the response menu reuses the ledger as its payment rail instead of adding a system.
• Rams #6 (honest): durations match fiction; a muted channel's events resolve honestly at list price; the insane 25%/turn loan becomes a nameable desperation tier at month denomination instead of a bug.
• MaRo Interaction: the reserve gamble makes the plan and the event stream interlock. (Does not by itself fix the dominant-line finding — that's payoffs, beat 2.)

Interaction contract

Reads/writes ≥2 existing systems: calendar/game_state (month boundary, day tick), Liability Ledger (DEFER intake — the flagship gains an intake valve every run hits), SA channels (window gating: SA buys response windows, ADR-0004 cashed out; muting a channel = auto-resolve without a window, priced), replay artifact (response records), exploit-finder (policies become plan-policy × response-policy), payroll/loans (re-denominated monthly).

Rejected alternatives

• Week (the original intent): dead against the fiction window — ~500 turns to an early loss can't carry fast replays or badge legibility.
• Quarter: strong candidate (evening-length runs; native quarterly-report/runway fiction; turn numbers matching the old badge intuitions). Lost to Pip's wall-clock preference (6–8 hr campaigns) and early-game texture. Revisit trigger: month-grain playtests showing decent runs well over ~8 hr, or empty/repetitive mid-run months.
• Variable/coarsening cadence (month → quarter as the org scales): honest fit to the delegation fantasy ("They can't go to all these weekly stand-ups, after all!") — parked until fixed grain demonstrably fails in playtest (Rams #10 discipline).
• Soft reserve fallback (idle time does low-efficiency generic work): rejected v1 — crisp evaporation keeps the commitment-loss legible.

Consequences / open questions

• Re-denomination pass: salaries → monthly payroll; loans → %/month with cost-of-debt as a function (org type / hype / reputation — merges with the DEFER pricing engine, agenda 3+4); event density per month; review_period_weeks conversion.
• Replay schema bump: record (event_id, window, choice, payment_source) per response (ADR-0001/ADR-0006 artifact versioning; adjacent to DQ-6).
• Exploit-finder rework: bots need plan-level and response-level policies — real work, and it upgrades the instrument (next sweep question: which response policy dominates?).
• UI: day-tick playback needs speed control + auto-pause-on-window; the month review screen is the natural home for world-state progression display (Pip's playtest note: progression feel is weak).
• Risk (~40%): late-game dead air if interrupt density doesn't scale with era — expect a per-era density tuning pass.
• Partially closes ADR-0008's deliberately-open wall-clock item: decent run 6–8 hr; consistent with its fixed constraints (deep-run tail "several hundred turns" ≈ 270 months to 2040; a noob death ~2 fiction-years ≈ 24 turns still costs well under an hour).
• Save/fork/divergence: Pip is open ("I don't see strong reasons why we shouldn't"); legal under ADR-0006's verification law (a legal replay is a legal run however produced). Parked to backlog — it's ladder norms, not rules.