# Aetherbound Guild Pre-Development Readiness Audit

Date: 2026-08-11
Status: `PHASE5_REFERENCE_FIDELITY_CANDIDATE_READY`
Integrated product source: `e7cdbfd8d4205d98d60302d38adc654c12077dfe`
Frozen specification: `.agent-taskgraph/spec.md` revision `design-spec-r3`

## Decision

`e7cdbfd8d4205d98d60302d38adc654c12077dfe` is the unique current
pre-development product candidate. It integrates the complete design
authorities, deterministic validators and simulations, the 50-page landscape
HTML prototype, seven Owner review groups, and the 15-entry reference fidelity
map. The reference map transfers functional structure only; it contains no
commercial screenshots, media, names, prose, formulas, code or copied layout.

All local machine gates pass. The package is ready for Owner review of product
direction, comprehension, visual composition and choice quality. It is not yet
Owner-approved and does not authorize engine development, final asset/audio
production, packaging, device work, store submission or release.

Earlier R4 and R5 failures and repairs remain below as append-only design
history. They are not the current status. In particular, the seed-first
run-creation transaction, hidden-profession art bindings, deterministic
outcomes, anti-dominance fixture and validator negative cases are integrated
and covered by the current 28-case self-test.

## Integrated Candidates

| Workstream | Candidate | Scope |
|---|---|---|
| Core systems | `b0f9033`, repair `e598eed` -> `f55906e`, `9be1927` | four files under `docs/product/` |
| Content matrix | `4e07a91` + `6a41189` -> `65f316d` + `30fb0b9` | `docs/content/` only |
| UX and screen states | `c364a5e`, repair `c64b99e` -> `dc8d654`, `ec4863a` | two named presentation files |
| Art, animation and VFX | `ca4d3df` -> `4b6270a` | art catalog only |
| Audio | `1d51137`, repair `ff4101a` -> `8b26e0a`, `f0095af` | audio catalog only |
| Design validator | `493e714`, repair `e7f80d0` -> `b2d8f7b`, `83bba23` | validator contract and executable only |
| Deterministic outcome authority | `5a05ffc` -> `f2a3f3c` | battle systems authority only |
| Content outcome sync | `d5fd2fc` -> `edbd281` | four content catalogs only |
| Art outcome sync | `8c8c1b7` -> `f8d5bfe` | art catalog only |
| Audio outcome sync | `ea69bc7` -> `404e996` | audio catalog only |
| R3 onboarding and screen repair | `388d90b` -> `14ea46e` | two UX authorities only |
| R3 art schema/order repair | `7f8d4b4` -> `03547a7` | art catalog only |
| R3 audio coverage repair | `9f63e46` -> `7071887` | audio catalog only |
| R3 reference-scope repair | `565ba2e` -> `96ea49a` | transferable lessons only |
| R3 validator gate + H1 integration repair | `ec5e234` -> `8a6f0c8`; `49e128f` -> `bcb2931` | validator contract and executable only |
| Fixed-seed anti-dominance simulation | `cb53794` -> `b65646c` | economy authority, one fixture and one simulator |
| R4 seed/transaction repair | `ad0a57d` -> `bba0c43` | product, save/failure and Recruit-generation authorities only |
| R4 hidden-profession art repair | `fef9057` -> `8102bd2` | art catalog only; invalid `c558d19` excluded |
| Phase 5 reference fidelity | `e60191b` -> `e7cdbfd` | 15-entry structural map, integrated Market, Gallery review surface and browser verifier |

The content worker's exact identity ACK was observed at candidate close instead
of before implementation. Runtime configuration, isolated worktree, exclusive
write scope, two candidate commits and clean terminal are all independently
recorded. This is a process-order deviation for reviewer awareness, not hidden
evidence or a reason to upgrade product acceptance.

## Verified Envelope

| Domain | Verified count or boundary |
|---|---|
| Professions | 12 base + 24 advanced + 6 hidden = 42 |
| Recruit traits | 36 |
| Equipment | 320 |
| Run artifacts | 60 |
| Ordinary/elite enemies | 100 |
| Bosses | 16 |
| Regions | 8 |
| Player-facing pages | 50 complete screen definitions |
| Audio production IDs | 32 music, 76 ambience, 228 shared SFX, 32 voice intents |
| Integration bindings | 34 declarations; zero unresolved values |

The product authority now uses generated recruits, repeated shop decisions, a
single ordered horizontal party line, automatic combat, recruitment,
dismissal, permanent casualty and replacement. Fixed protagonists,
relationships, fixed six-person grid combat, three lanes and manual
Directive/Anchor/Cache battle authority are retired.

The first independent review of fixed revision `7ffd6b0` failed because an
ordinary dismissal could remove the final live Recruit and expose an undefined
zero-Recruit Market. The forward repair now evaluates one canonical
`live_recruit_count` at preview and commit, rejects final-survivor, stale,
duplicate and interrupted dismissal without rebate or state mutation, and
keeps casualty recovery and explicit run abandonment as separate
transactions. UX exposes the rejection and a distinct `OUT-006` abandonment
confirmation that routes directly to run end rather than an empty Shop.

Systems R2 then failed fixed revision `7054fe6` because `PF-A04` depended on
undefined probabilistic hit/miss/evasion semantics. The forward repair keeps
the product rule of no hidden miss chance, defines 12 deterministic `AO_*`
terminal outcomes, and binds Turnblade, affected equipment/traits/enemies,
VFX shapes, SFX and captions to those outcomes. Experience R2 was `INVALID`
because its context named a nonexistent reference file; R3 must use the actual
authorities `docs/reference/REFERENCE_RUNTIME_CENSUS.md` and
`docs/reference/TRANSFERABLE_STRUCTURE_LESSONS.md`.

Systems R3 failed fixed revision `d454120` only because the declared
anti-dominance promise lacked executable fixed-seed inputs and per-policy
outcomes. Five exact states now replay four legal policies with winner spread
`refresh_first=1`, `hire_first=2`, `dismiss_first=1`, `no_replace=1`; a
deliberately dominant in-memory matrix is rejected. Experience R3 failed the
same revision on provisional choose-two presentation, party direction,
recruit-art schema mapping, `OUT-006`/50-screen audio coverage and obsolete
fixed-cast reference lessons. Those authorities are repaired. The first
validator integration candidate then produced three parser false negatives
against the real repaired document shapes; H1 now accepts semantic-equivalent
staged rows, endpoint clauses and four-column crosswalks while preserving all
five negative diagnostic families.

## Reproducible Checks

Run from the repository root:

```bash
python3 tools/validate_predevelopment_design.py --self-test
python3 tools/validate_predevelopment_design.py
python3 tools/validate_predevelopment_design.py --json
python3 tools/simulate_anti_dominance.py --self-test
python3 tools/simulate_anti_dominance.py
git diff --check
```

Expected markers and counts:

```text
ABG_PREDEVELOPMENT_DESIGN_SELF_TEST_OK cases=28
ABG_PREDEVELOPMENT_DESIGN_OK errors=0
ABG_ANTI_DOMINANCE_SELF_TEST_OK seeds=5 dominant_matrix_rejected=1
ABG_ANTI_DOMINANCE_OK no_weak_dominance=1
ABG_R6_BROWSER_OK screens=50 flow_states=23 reference_maps=15 review_groups=7
```

The JSON result must contain `"ok": true`, an empty `diagnostics` array and
the exact counts above.
The combined fixed-seed replay result SHA-256 is
`17154371d4b93035296d1219f012f381af2e10a5f2c1b3f66bcfcd7ad4d85352`.

The current `prototype/site/` and self-contained Gallery bundle are the
Owner-facing representation of this candidate. They are a design prototype,
not runtime evidence, final art, an engine build or proof of human acceptance.
`tools/validate_design_contracts.py` remains historical; the current authority
is `tools/validate_predevelopment_design.py` plus the browser verifier named in
`docs/prototype/INTERACTIVE_PROTOTYPE_SPEC.md`.

## Review Questions

1. Does the ten-minute loop remain immediately understandable: inspect offers,
   recruit or buy, equip, reorder, choose risk, watch automatic combat, read
   casualties/rewards, then make the next shop decision?
2. Does party order materially change protection, reach, movement, healing and
   replacement without recreating a hidden lane grid?
3. Do permanent loss and price pressure create recoverable adaptation instead
   of an unrecoverable run collapse or a trivial dismissal exploit?
4. Are all 42 professions, 36 traits, 320 equipment, 60 artifacts and 116 enemy
   identities decision-bearing rather than name-only filler?
5. Are the five pacing profiles, save/failure transaction rules and postgame
   loop executable and mutually consistent?
6. Can the 50-page landscape UX teach the loop without relying on prose-heavy
   documentation, unexplained icons or manual skill controls?
7. Are the modular art and audio scopes feasible, original and consistent with
   generated recruits rather than fixed protagonists?
8. Do any transferred structural lessons reproduce commercial expression,
   names, formulas, layouts, source or assets?

Questions 1-5 remain covered by deterministic design validation and simulation;
questions 6-8 require Owner judgment through the current Gallery. Historical
reviewer terminals remain evidence for their fixed revisions only and are not
substitutes for Owner approval of `e7cdbfd`.

## Open Gates

- Owner review of the 15 reference mappings, especially integrated Market,
  battle payoff, Guild Charter and completion-versus-continuation choice.
- Owner acceptance of game direction, comprehension, choice quality and scope.
- Final art/audio generation, integration, human visual review and listening.
- Engine implementation, simulations, runtime, physical device and persistence.
- Packaging, signing, stores, submission, publishing and release.