|
|
|
@@ -0,0 +1,54 @@
|
|
|
|
|
# Goal: ABG-V2 Independent Recruit-Line Design Review
|
|
|
|
|
|
|
|
|
|
> Task ID: `ABG-V2-REVIEW`
|
|
|
|
|
> Frozen spec: `.agent-taskgraph/spec.md` revision `design-spec-r3`
|
|
|
|
|
> Graph node: `v2_independent_review`
|
|
|
|
|
> Fixed revision: `7ffd6b0`
|
|
|
|
|
> Repository: `/Users/wusumac/Documents/game/aetherbound_guild`
|
|
|
|
|
> Legal terminal: `ABG-V2-REVIEW PASS` or `ABG-V2-REVIEW FAIL: <earliest failed boundary>` or `ABG-V2-REVIEW INVALID: <review environment boundary>`
|
|
|
|
|
|
|
|
|
|
## Role
|
|
|
|
|
|
|
|
|
|
Act as a fresh independent reviewer. The fixed revision is immutable. Use a
|
|
|
|
|
reviewer-owned detached archive or worktree and isolated temporary outputs.
|
|
|
|
|
Never edit, commit, merge, push, generate media, package or repair.
|
|
|
|
|
|
|
|
|
|
## Required Gates
|
|
|
|
|
|
|
|
|
|
1. Prove the reviewed tree is exactly `7ffd6b0`, clean and contains the six
|
|
|
|
|
integrated authorities plus the readiness audit and repaired validator.
|
|
|
|
|
2. Run the validator self-test, human-readable full audit and JSON audit from
|
|
|
|
|
zero. Require exact counts, zero diagnostics and deterministic output.
|
|
|
|
|
3. Review the ten-minute loop as a player decision chain: shop offers,
|
|
|
|
|
recruit/buy, equip, reorder one horizontal line, choose risk, automatic
|
|
|
|
|
combat, casualty/reward and the next shop decision.
|
|
|
|
|
4. Check deployment capacity versus cumulative recruitment, dismissal,
|
|
|
|
|
permanent casualty, replacement, price pressure, retreat and recovery for
|
|
|
|
|
exploits or unrecoverable contradictions.
|
|
|
|
|
5. Check deterministic battle, five pacing profiles, no-decision gaps,
|
|
|
|
|
solvency, anti-dominance and complete save/failure transaction semantics.
|
|
|
|
|
6. Audit all 42 professions, 36 traits, 320 equipment, 60 artifacts, 100
|
|
|
|
|
ordinary/elite enemies, 16 bosses and eight regions for schema coherence and
|
|
|
|
|
decision-bearing differentiation. Use full mechanical checks plus targeted
|
|
|
|
|
semantic sampling; report any filler family precisely.
|
|
|
|
|
7. Audit 49 screen definitions, onboarding, EN/zh_CN, 130% text, touch and
|
|
|
|
|
controller input, empty/error/confirmation/interruption, reduced motion,
|
|
|
|
|
mute and low-power states. This is a design review, not runtime evidence.
|
|
|
|
|
8. Audit art, animation, VFX, music, ambience, SFX and generated-recruit voice
|
|
|
|
|
scope for feasibility, traceability, readability and absence of fixed-cast
|
|
|
|
|
assumptions.
|
|
|
|
|
9. Check reference claims distinguish confirmed observation, lower bound,
|
|
|
|
|
inference and original invention. Fail copied names, prose, formulas,
|
|
|
|
|
layouts, source or commercial assets.
|
|
|
|
|
10. Confirm old `prototype/site/` and `tools/validate_design_contracts.py` are
|
|
|
|
|
explicitly superseded and are not being presented as current evidence.
|
|
|
|
|
|
|
|
|
|
## Terminal Rules
|
|
|
|
|
|
|
|
|
|
- `PASS` only if every required gate passes on the fixed revision.
|
|
|
|
|
- `FAIL` for a product/design/evidence defect; stop at the earliest defensible
|
|
|
|
|
boundary and cite file/line evidence plus checks already completed.
|
|
|
|
|
- `INVALID` for archive, command, parser, environment, timeout or reviewer
|
|
|
|
|
process failure; do not turn it into product FAIL and do not retry silently.
|
|
|
|
|
- Human fun, comprehension, visual acceptance, listening, device, runtime,
|
|
|
|
|
packaging and release remain open even after PASS.
|
