Gate Check: Concept → Systems Design

Date: 2026-09-14 Checked by: gate-check skill Review mode: lean (all four directors run — phase gates are what lean mode reserves them for)

Required Artifacts: 3/3 present

• [x] design/gdd/game-concept.md — 39KB, all 15 concept-template sections, zero placeholders
• [x] Game pillars defined — 3 pillars + 4 anti-pillars, in the concept doc (no separate file needed)
• [x] Visual Identity Anchor — present, with a one-line visual rule and 3 supporting principles

Recommended (not blocking): 0/1

• [ ] Concept prototype with a PROCEED verdict — the prototype returned PIVOT. Its PROCEED verdict was withdrawn the same day: the ruleset admitted a dominant strategy (15 spears beat any 2d6 raid, bankable two turns before the earliest one) and the plough payoff could not generate regret. The revised design — failure floor, production-halting muster, per-worker income, damaged gear — is entirely untested.

Quality Checks: 3/4 passing

• [x] Core loop described — 4 tiers (30s / 5-15m / 30-120m / long-term)
• [x] Target audience identified — Target Player Profile with 7 attributes
• [x] Visual Identity Anchor has a one-line rule and ≥2 supporting principles (3 present)
• [ ] Design review verdict — MAJOR REVISION NEEDED on the first pass. All five blockers were resolved, and a second pass resolved six more findings. But the gate asks for a verdict that is not MAJOR REVISION NEEDED, and no re-review has been run since the revisions landed.

Director Panel Assessment

Creative Director:  CONCERNS
Technical Director: CONCERNS
Producer:           CONCERNS
Art Director:       CONCERNS

Creative Director — The vision is coherent and well-argued on paper, and the Anti-Pillar 2 rewrite genuinely constrains where the old wording did no work. But the only empirical data this concept ever produced was a dominant strategy that collapsed its central decision. Recommends advancing the systems that do not depend on the muster's unresolved numbers, while gating the Muster, losses/income model and threat-band GDDs behind a second adversarially-reviewed prototype — specifically, not letting Formulas and Tuning Knobs be written for those before the retest.

Technical Director — Decomposition is sound and correctly layered. Two binding conditions, neither blocking the start of Systems Design: (1) the tick-phase contract must be pinned in the settler-simulation ADR as a single shared enum before GDD #14 (The Muster) is drafted — it is 13 systems away; (2) Threat Forecasting UI needs an explicit camera-yaw dependency, since directional threat must stay correct under arbitrary rotation. Also flagged that Calendar/Time must explicitly own tick-phase ordering, not just day/season counting.

Producer — The hardest numbers in this gate. The 19 MVP systems total roughly 40–45 design sessions, which at one focused session per working day is **8–9 calendar weeks of design alone**, before any implementation. Against the stated "3–5 weeks, aspirational" MVP timeline that is 2–3× the entire budget spent on documents. Recommends re-prototyping first, authoring only the prototype-blocking subset (Calendar/Time, Resource/Stockpile, the Armoury/Muster/Role cluster, Failure State), and gating the remaining ~13 GDDs behind a second PROCEED verdict.

Art Director — The anchor is unusually rigorous for this stage: real design tests, an explicit lighting-leads-UI-confirms ordering rule, named accessibility mitigations. Two fixable, non-blocking concerns: the rotating-camera threat indicator is acknowledged but unsolved and risks quietly becoming the UI element the ordering rule forbids; and the FLUX keyframes read as painterly 2D illustration rather than the "chunky 3D model, painterly shading" register the text describes — a reference-vs-target confusion that costs one clarifying sentence to prevent.

Fixed During This Gate

The producer's count exposed a real inconsistency. The enumeration table listed 19 MVP systems; the design order table listed 18. **Failure State / Run End had been added to the enumeration but never to the design order** — the system that defines how the game ends had no place in the sequence that says when to design it. Now inserted at #17 (Presentation layer, after Combat Resolution, which it reads), tail renumbered, tracker reconciled at 19.

Also applied from TD-PHASE-GATE: the shared tick-phase enum contract is now written into the Circular Dependencies section, and Threat Forecasting UI carries its camera-yaw dependency.

Blockers

• The core mechanic is untested. The prototype's PROCEED was withdrawn; the replacement design has never been played. Four directors independently flagged authoring GDDs on it as the central risk.
• Schedule is not reconciled with scope. 8–9 weeks of design against a stated 3–5 week MVP. The number is not wrong — the framing is, and it should be corrected before more design debt accumulates.

Recommendations

• Re-prototype the muster per prototypes/muster-mechanic-concept/PIVOT-NOTE.md. Days, not weeks. It must verify the 2-settler reserve prevents the death spiral, and that no dominant strategy survives scaling raid strength and gear decay.
• Author only the prototype-blocking subset meanwhile: Calendar/Time, Resource/Stockpile, the Armoury/Muster/Role cluster, Failure State.
• Gate the remaining GDDs behind a second PROCEED.
• Re-baseline the timeline honestly in the concept doc's Scope Tiers.
• Pin the tick-phase contract before GDD #14; add the keyframe caveat to the anchor.

Verdict: CONCERNS

All required artifacts are present and of good quality. The concept document is thorough, honest about its own gaps, and has survived two adversarial review passes. The gap is not in the documents — it is that **the mechanic they describe has never been played in its current form**, and four directors independently reached the same conclusion about what that means for sequencing.

CONCERNS rather than FAIL because the prototype is recommended, not required, at this gate, and because advancing is legitimate provided the sequencing is deliberate: design what the prototype needs, prototype, then design the rest.

Chain-of-Verification: 5 questions checked — verdict unchanged. One artifact inconsistency found and fixed during verification (the missing Failure State row).