# Goal: ABG-V2 Systems Integrity Review R2

> Stage: pre-development fixed-revision review
> Started: 2026-08-11T00:47:43+08:00
> Task ID: `ABG-V2-SYSTEMS-REVIEW-R2`
> Baseline: `7054fe6` on local `main`, clean, no configured upstream
> Accepted previous stage: integrated product source `ec4863a`; PMO review closure `7054fe6`
> Frozen spec: `.agent-taskgraph/spec.md` revision `design-spec-r3`
> Graph node: `.agent-taskgraph/graph.yaml#v2_systems_review` revision `graph-r5`
> Independent contract: `docs/reviews/PRE_DEVELOPMENT_READINESS_AUDIT.md`
> Context manifest: `context.md`
> Context revision: `1`
> Legal terminal: **ABG-V2-SYSTEMS-REVIEW-R2 PASS** or **ABG-V2-SYSTEMS-REVIEW-R2 FAIL: <earliest failed product boundary>** or **ABG-V2-SYSTEMS-REVIEW-R2 INVALID: <review environment boundary>**

## Objective

Independently audit immutable revision `7054fe6` for an executable and
internally consistent recruit-shop, ordered-line auto-battle, economy, pacing,
save and failure design. Recheck every gate from zero; do not inherit R1's
partial work or assume the final-Recruit repair is sufficient.

## Graph Input, Output And Write Scope

- Needs: `v2_integration`
- Consumes: immutable `7054fe6`, design-spec-r3 and all product/content
  authorities required by the checks below
- Produces: one read-only PASS/FAIL/INVALID terminal with exact file/line and
  command evidence
- Writes: none in the repository; reviewer-owned temporary files only
- Route: PASS -> `v2_review_synthesis`; FAIL/INVALID -> `failed`; max attempt 1

## Assignment

- Role ref: `role:systems-integrity-reviewer-v2`
- Lifecycle: task-scoped
- Assigned to: new isolated visible HAPI session
- Reason: independent systems expertise; R1 is retired and cannot be reused
- Continuity: fresh role; may read the R1 report only as append-only history
- Workspace: repository root for authority discovery, then reviewer-owned
  `git archive 7054fe6` for all judgment
- Runtime requested: `runtime=hapi; flavor=codex; model=gpt-5.6-sol; effort=xhigh; permission=bypassPermissions; visibility=visible`
- Runtime observed: `runtime=hapi; flavor=codex; model=gpt-5.6-sol; effort=xhigh; permission=yolo; visibility=visible`
- Runtime verification: `VERIFIED` pre-dispatch at 2026-08-11T00:54:54+08:00
- Session evidence: HAPI `7945e8ff-abe9-45ef-9f22-2838dacfa197`, PID
  `41548`, verification `31e6e871-6278-42df-afb5-a9388675a22a`,
  `runtime-evidence.json`, zero-message watermark
- Dispatch bootstrap: `dispatch.md`
- Dispatch message: `SENT: 2026-08-11T00:56:00+08:00 via HAPI peer 7945e8ff; dispatch:ABG-V2-SYSTEMS-REVIEW-R2:r1`
- Identity ACK: `VERIFIED: IDENTITY_READY dispatch_id=dispatch:ABG-V2-SYSTEMS-REVIEW-R2:r1 role=role:systems-integrity-reviewer-v2 team_revision=rev-4 goal=task:ABG-V2-SYSTEMS-REVIEW-R2 context_revision=1`
- Full launch method: HAPI Hub helper on machine
  `a15888ad-d405-4884-a407-c28ddd8b397c`

## Required Gates

1. Prove the archive is exactly `7054fe6`, contains `ec4863a`, and the
   canonical repository is clean. Use no later source for judgment.
2. Run `python3 tools/validate_predevelopment_design.py --self-test`, the full
   human audit and JSON audit from the immutable archive. Require self-test 15,
   `errors=0`, `ok=true`, exact counts, and deterministic repeated JSON.
3. Trace the first 10 minutes and every cycle boundary: offer inspection,
   recruit/buy, equip, reorder, choose risk, automatic combat, casualties,
   rewards and next Market. Each state must have at least one legal action.
4. Re-audit final-Recruit dismissal at preview/commit, stale confirm, duplicate
   confirm, interruption, no rebate/state mutation, casualty recovery and
   explicit abandon-run separation. Find any other zero-unit or no-currency
   dead state.
5. Recalculate battle examples and clamps including hit, mitigation, critical,
   minimum Raw damage, targeting/order, reach, movement, healing, status and
   deterministic tie-breaking. Reject undefined variables or contradictory
   formulas.
6. Recalculate shop/recruit/equipment prices, rerolls, sales, rewards, sinks,
   XP, levels, skill points and stat budgets. Require every solvency and
   anti-dominance gate to be executable from declared formulas and item rows.
7. Reconcile retreat, defeat, casualty, collapse and abandonment across product,
   economy and save authorities. Audit pending/applied transaction recovery,
   idempotency, corruption, clock rollback, interruption and bounded offline
   behavior.
8. Verify five named pacing profiles include explicit decision timestamps,
   chapter/run totals and no-decision gaps consistent with declared playtime.
9. Mechanically check all 42 professions, 36 traits, 320 equipment, 60
   artifacts, 100 ordinary/elite enemies, 16 bosses and eight regions; sample
   every family for counterfactual decisions and reject name-only filler or an
   obviously dominant family.
10. Confirm no copied names, prose, formulas, source, commercial assets or
    reference-specific expression entered the design.

## Terminal Rules

- PASS only when all ten gates pass.
- FAIL at the earliest defensible product/design contradiction; cite exact
  file/line evidence and checks completed before stopping.
- INVALID for archive, parser, command, environment, timeout or reviewer
  process failure. Do not retry silently or convert it into product FAIL.
- Never edit, repair, commit, push, generate media, package or release.
- Human fun/comprehension, runtime, device and release remain open after PASS.

## Estimate

- Review: 1.5-3h; hard upper bound 4h; confidence medium.