07c Game Level Review

07c — Independently review game spaces

Follow the protocol, including its source-loading rules. Inputs through profile.sources: baseline, all applicable 07b packages, action/camera rules, catalogs, 07a, UX, art/audio, technical constraints, current calculations from 07, earlier findings, and changes. A unit's author cannot perform its independent final review. Use a fresh-context Codex subagent under the subagent review protocol.

Record source revisions. Changes to movement/turn rules, resources, camera, objects, or accessibility can invalidate a report even when the plan is unchanged. For absent, check evidence of inapplicability. unknown and missing required packages cannot receive N_A.

1. Build a matrix of all required authored units and procedural families × required representations × checks. Read every package and view every included visual plan. Assess substantive placements and rules rather than file counts. Explicitly identify partial scope; it cannot support an overall positive verdict.
2. Reconcile IDs, units/grid, scale, axes/layers, boundaries, dimensions, and agreement between plans and structured data. Recalculate distances, widths, adjacency, costs, or angles that affect actions. Trace each required object/node from catalog and requirement to placement and permitted state. Identify the canonical representation and how derived copies stay synchronized.
3. For authored spaces, trace required paths and returns from optional branches with access conditions and state changes. Check allowed bodies/objects, turns, obstacles, and vertical transitions, or valid board moves and occupied cells. Identify cycles such as a key behind its own locked door, hidden mandatory resources, and unspecified dead ends. Do not require a physical corridor for a discrete graph rule.
4. For procedural families, check complete inputs, connection rules, required and forbidden combinations, boundary configurations, output acceptance conditions, retry limits, and fallback. Test documented examples and counterexamples against the rules. Investigate global failures not prevented by individual local constraints. Future validation must distinguish invariants, statistical quality, and sampling. A small sample cannot prove properties of every seed.
5. For every required encounter/task, check entry, outcomes, reset, permitted solutions, and recovery. Assess view/available information, landmarks, teaching before challenge, pacing, and accessibility. For competition, check starting conditions and geometry exploits; for puzzles, accepted solution multiplicity, dead ends, and hints; for combat, threats, means of action, and resources. Review selected mechanics without adding new ones.
6. Where space depends on resources, verify a current repeat of 07 after 07b and matching node IDs/order. Independently recalculate material entry → actions → exit → recovery sequences for supported unfavorable states. Sufficient totals do not excuse intermediate shortages; an encounter bypass remains a permitted variant. Missing required calculations are gaps. Where resources do not affect space, justified inapplicability is sufficient.
7. Check applicable narrative/learning transitions, revisits, gain/loss of key objects, reset/load, participant changes, and post-completion states. Reconcile zone joins, events, recovery of necessary information, and production constraints with their owners. Separate geometry conflicts from perception hypotheses requiring later testing.

For irreversible transitions, check information before action, affected promised objectives, and the basis for deliberate surprise. A reachable ending does not close this check.

Output: level-review.md in profile.paths.state_dir, with each unit's author/reviewer, source snapshot, matrix, scenarios, calculation sources, findings, and limitations. Findings include exact IDs/coordinates/nodes, consequences, the corrective stage, and closure checks. The coordinator owns shared logs.

Completion: required packages specify the intended structure unambiguously, all material findings are closed by recheck, and future gameplay evidence remains clearly separate from D0. Unavailable independent review remains pending. Route spatial errors to 07b, narrative errors to 07a, and rule/calculation errors to 06/07. Then update 05, 04, and affected review rows.