# Engagement replacement Task 4 — automated qualification

## Current result — automated scope complete, human acceptance open

23 October 2026: the agreed bounded automated gates pass for frozen W inputs and
the matching U development builds, with explicitly compatible earlier evidence.
This does not complete Task4's human keep/tune/revert or authorize promotion.
`TestResults/engagement-task4-20261023-eeca51c70f1748ec903e8a537df4e5f8/qualification-final-w.json`
is the immutable final gate/identity/reuse manifest. Lead revalidated every selected
positive/negative result, full frozen hashes, contract compatibility and cleanup.

| Gate | Accepted proof |
|---|---|
| Full EditMode | U1653/1653 fresh, zero skips/errors |
| Current native gameplay | 660 distinct passing leaves:338 compatible baseline,172 compatible H,150 superseding; not a fresh full-suite pass |
| Controlled launcher contracts | Final evidence27/27; unchanged director/coordination L contracts explicitly reused |
| Builds/exclusion | U normal/C3 development succeeded; R release and its excluded inputs explicitly reused; executable/packed exclusion audit passes |
| Actual C3 coordination | R two-peer94checks compatible mechanical reuse; U four-peer126checks fresh, real20/2/3 and lifecycle tail |
| Arena/lifecycle | V intended negative valid, positive two/four-player reset2/failure-proof/combat/lifecycle pass |
| Normal entry/progression | W menu10phases and Verdant two/four-player23phases pass |
| Final state | All2503 frozen inputs unchanged, no players, clean idle Editor, original preferences restored |

Original RED results remain immutable, including one unexplained R four-peer
receipt3.605ms late. The passing U run is current bounded proof, not a causal fix
or repeatability guarantee. All peer execution is automated local processes with
real online sessions and75/15/1 simulation. Occluded captures do not prove natural
noticeability. MainMenu's legacy probe lacks independent GUID readback; its exact
executable/full built data is bound before/after to the public normal build report.

The32 owned source inputs remain a preserved dependent dirty checkpoint;2471 other
original inputs are unchanged. A scoped documentation checkpoint records this work
without accumulating uncommitted prerequisites or unrelated staged art. Human
Task3 moving diagnostics, remaining engagement/M6–M8 observations, keep/tune/revert,
representative hardware, separate networks and production promotion remain open
in the [playtesting checklist](../../PLAYTESTING_CHECKLIST.md).

## Subsequent source checkpoint

After this qualification, the user explicitly authorized committing all completed
outstanding enemy AI work. Implementation commit `674abeea` includes the dependent
earlier slices and qualification corrections with unchanged W input bytes. The
[commit record](2026-10-23-enemy-ai-checkpoint-commits.md) closes the source-commit
limitation recorded below while retaining human acceptance and promotion gates.
The original execution chronology and its then-current dirty-state descriptions
remain historical evidence; no tests were rerun merely to package commits.

## Authorization, scope and ownership

23 October 2026: the user explicitly requests completion of Task 4's automated
qualification. Execute the automated part of the [replacement plan](2026-10-23-engagement-pursuit-replacement.md#task-4--integrated-qualification-and-keeptunerevert-decision),
not human playtesting or production promotion. Human timing remains End/deferred
because the user is unavailable. Tasks 1–2 have user-reported gameplay acceptance;
Task 3 visual inspection is accepted, gameplay diagnostic acceptance remains open.
The [playtesting checklist](../../PLAYTESTING_CHECKLIST.md) retains those gates.

Player intent: coherent moving combat, a finite attempt to regain useful combat,
an observable escape opportunity and useful independent encounter pressure. The
automated question is whether current owners, lifecycle, evidence, physical safety,
deadlines, shared accounting and real peer execution preserve that contract.

Scope: full EditMode and PlayMode; current melee/lunge/engagement integration;
normal/debug/release build identity and release exclusion; affected real two/four-peer
coordination/combat/targeting/lifecycle/join/leave/reset/disconnect; common menu and
normal-stage regressions after confirmed player cleanup. Inspect harness coverage
before claiming new pursuit scenarios. Diagnose and correct demonstrated automated
blockers at their responsible layer; no acceptance weakening, policy redesign,
roster expansion, authored terrain/scene regeneration or NavMesh rebake.

High-risk system qualification. Lead selects checks, diagnoses and integrates;
one delegated test operator owns exclusive Editor/test/build/player operations
during each frozen batch. Lead is sole source writer. Independent technical,
gameplay and design/UX final review is required; significant runtime corrections
receive the relevant pre/post delta review. At most lead plus three active agents.
Model policy: Sol6.1/Ultra lead, Sol6.1/Medium operator/reviewers.

## Preserved checkpoint and prerequisites

Starting HEAD `8e0d3e0` plus the existing dirty tree is the actual checkpoint.
Preserve all prior implementation, tests, art and authoring. Baseline scripts,
metas, package/settings/tools, full input hashes and Git status are under
`TestResults/engagement-task4-20261023-eeca51c70f1748ec903e8a537df4e5f8/`.
The manifest has2501 files; tool virtual environments/cache/dependencies are excluded
from the copy. An earlier incomplete snapshot directory ending `722eb52b…` is not
used as evidence or a rollback baseline.

Lead preflight: connected to the correct project on pinned Unity6000.6.4f1; package
pins unchanged (Behavior1.0.16, Navigation2.0.14, NGO2.13.3, URP17.6.0, Tests1.8.0).
Runner idle with no active request, Console0errors. The only playing session was
the owned paused Task 3 sandbox; stop requested through its supported cleanup.
Operator must confirm stopped, ready Editor and restored authoring/play settings
before discovery. Saved/live/cached VSync was consistently0/false; retain and
verify that profile. Preserve any unsaved authoring; no blind SaveAssets or rebuild.

## Execution sequence and acceptance

1. Operator confirms readiness and positive exact discovery through the existing
   Foundation runner; run full `ProjectI.Foundation.EditModeTests`, then full
   `ProjectI.Foundation.PlayModeTests`, once each on frozen input. Labels begin
   `engagement-task4-baseline-…`. Even an assertion-RED first suite does not excuse
   omitting the second, provided ownership/readiness is clean. Invalid discovery,
   source drift, infrastructure or uncertain cleanup returns to lead first.
2. Lead classifies actual failures: product, fixture/setup, compile/discovery/
   identity, infrastructure, interrupted ownership or incomplete coverage. Preserve
   every original report. Reproduce relevant failures, obtain required delta review,
   correct the cause and repeat only invalidated checks. Historical failures are
   neither a waiver nor proof of current failure.
3. Audit native and built-player coverage against the thirteen scenario rows in
   replacement-plan section6, particularly real moving melee/lunge opportunity,
   exact Breakoff, target swaps, source/lifetime cleanup and group conservation.
   Add only missing meaningful coverage through existing fixtures/harness seams.
4. Freeze corrected source and select matching builds. Task 3's successful build
   provenance is reusable only for demonstrably unchanged compatible inputs; source/
   art authoring has since changed, so compare actual inventories rather than HEAD.
   Use unique/preserved outputs. Active coordination needs the documented temporary
   fixture/Resources packaging, not only a generic debug build.
5. Execute selected supported two/four-peer cohorts with existing ceilings,
   stability predicates, read-back75/15/1 simulation and required failure-path proof.
   Confirm all owned processes have exited before another cohort or common normal
   menu/Verdant progression regression. No Natural-arrival rerolls/new budgets.
6. Inspect every manifest: exact requested/discovered/executed selection, current
   source/compiled/build identity, terminal match, zero unexplained failures/skips
   and preserved cleanup errors. Inspect required captures separately from tool
   success. Finish technical/gameplay/design final review and update this record
   and brief with exact automated result and remaining human/hardware/network gates.

Duration unknown: full PlayMode and online packaged cohorts may be substantial;
record actual phase durations, never add overlapping phases as elapsed totals.
No new campaign budgets or indefinite retries. Stop/escalate on ownership ambiguity,
authoring loss, incompatible evidence, an external service blocker without changed
conditions, or a required new design decision. A timeout is not termination.

Automated completion requires actual agreed gates, not merely executing them.
If a gate cannot be satisfied, report its concrete retained failure/incomplete
coverage; do not mark qualification GREEN. Human readability/feel, representative
hardware and independently controlled separate-network acceptance remain open.
No accumulated implementation commit or promotion through existing RED/human gates.

## Execution ledger

- Full EditMode `engagement-task4-baseline-edit-20261023a`, request
  `c0a5a479ae9d47d491990339189820fa`: **1646/1649 passed**,3failed,0skipped/
  inconclusive, completed/matched terminal, no error. 05:44:00–05:44:37UTC.
  Source `0D50C66C6C9284B0653A2F7C3B2C1BBE3052DFCD753493C5A95A2E7D22892065`;
  compiled `D485F9A188F478688CDB49597296B46477320B4F1FD63899292F2BDCA6CB4B97`.
  All three failures are SharedWeaponMotionTests: two warm-cache variants and the
  historical neutral-migration comparison. Original RED artifacts remain intact.
- Full PlayMode `engagement-task4-baseline-play-20261023a`, request
  `5b18d47294f54b18837703dbb8a4e16e`: **642/649 passed**,7failed,0skipped/
  inconclusive, completed/matched terminal. Execution GUID
  `422d6f31-af2e-4a4d-9620-c12ce0faae24`;05:45:16–06:14:31UTC,
  execution1750.81s. Source identity matches EditMode; compiled identity
  `47D0F2BB0AFCCB767049A044236D8DEF90256437046F7AD3C8D676605239AE95`.
  Full failure names/messages/stacks and phase timings are preserved in the
  baseline directory's `initial-full-suite-summary.json`. Failures: two retirement/
  recovery cases, firing-fallback travel, and four first-person presentation cases.
  Operator confirmed ready/idle/stopped and all2501baseline hashes unchanged.
  Three Console errors are deliberately injected by passing cleanup/observer/
  sandbox failure-path tests, not unresolved cleanup errors.
- Operator owns a subsequent diagnostic batch: each failed first-person case once
  in isolation against unchanged input. No retries or corrections during that batch;
  source remains frozen until explicit handback.
- Source baseline2501file hashes matched after stopped-session restoration and
  after EditMode. Original start scene/play options were restored; VSync remained
  consistent. No implementation changes or player launches during this batch.
- Task 3 evidence remains historical scoped proof: rules93/93; native154/155 with
  its remaining fixture correction and six affected reruns passing; successful
  development/release builds with explicit release-exclusion limits. It does not
  replace the full suites or current matching-build cohorts selected here.

## Reviewed correction and coverage decisions (not yet applied)

- All three independent final perspectives identified/confirmed one P2 blocker:
  `EnemyMeleeExecution.TryBeginServer` accepts a rooted target at3m inside lunge
  admission range3.5m, disables translation, but pays/confirms Fight although its
  stationary strike cannot physically reach. Repetition can restart pursuit.
  Require the existing owner `ContactOpportunity` before **new rooted admission**,
  with target-directed heading and current zero-distance fallback. Preserve valid
  close stationary hits, paid miss semantics and an already-paid lunge's remaining
  stationary contact window when Root arrives later. Add a real paid/cycle RED
  regression for no payment/confirmation/deadline refund, plus positive contact.
- No other concrete source blocker emerged. A suspected shared DirectDamage
  source-death issue was checked against the pre-task source and accepted alert/
  report contract: generic retained pack damage knowledge is unchanged and is not
  being silently assigned alert/report revocation rules. No speculative fix.
- Motion fixture diagnosis: copied Sunskipper gains are authored0.67, so warming
  then expecting gain2 to equal twice that baseline is invalid. Set the **owned
  disposable probe** to neutral gains before warming; retain both save variants,
  warm references/state, absolute0/2 assertions and tolerances. Hand-cannon inline
  profile is inactive after shared authoring; its sprint.1899 versus current shared
  .35 is not a runtime migration defect. Preserve migration coverage using an owned
  shared definition/view populated from the captured legacy baseline, with neutral
  multipliers and the full240-action/output comparison. No weapon runtime retuning.
- Coverage gaps: current outside-sphere test uses setup warps, and existing melee
  probes use direct stationary admissions. Add actual supported traversal across
  the former boundary and autonomous paid crawler/lunge/cycle coverage using real
  owners, including opportunity loss and nonrefunded search. Do not claim warps or
  substituted center-range probes prove those moving behaviors.
- C3 currently exports bolt `phase`, not engagement phase. Extend the existing
  peer observer/harness with fenced cycle/target/deadline/Breakoff/owner-work
  receipts and a bounded real20/2/3 qualification phase. Observe full Breakoff with
  returning legal combat but no target-seeking work; genuine target swaps/damage
  cannot move the current attempt deadline. Keep420s ceiling and existing native,
  wound/finance/lifecycle/client-authority checks. Legitimate combat confirmation
  may accrue during Breakoff, but cannot end its movement restriction early.
- Existing clear operation's synchronous generation/deadline preservation remains
  required. Its prolonged stability assertion is obsolete across legitimate Fight/
  loss transitions. Retain lifetime, wound, paid recovery, finance and sole pressure
  contribution conservation; test no deadline refund within a captured Pursue
  generation instead of forbidding legitimate later generations.
- Corrections wait for the owned full PlayMode terminal. Lead is sole source/
  task-record writer; operator writes operation evidence under TestResults only.

### Diagnostic results and correction rationale

Four original first-person failures were each isolated once on unchanged baseline:
ADS, grip and delayed-contact expectations reproduced; LowCeiling passed. All four
had exact one-leaf matched terminals, unchanged source/compiled identity and restored
Editor state. `first-person-isolation-summary.json` retains those artifacts.

Subsequent test-only diagnostics preserve the original failing assertions. Rooted
3m admission reproduced `Running` instead of `Rejected`; close rooted contact passed.
The rejection test's eventual long tail will use one genuine30s Snare, because its
owner correctly rejects attempts to refresh an already-active Root. No status rule
or resistance is changed. Rooted rejection failed on its first admission as intended.

The fallback's actual Ready plan is a supported transit prefix: section0.365605m,
full route12.814211m, action allowance0.5m, original Investigation context,30candidates.
Technical review approves paired migration: retain the original0.5m case as real
bounded-travel positive proof; preserve the failure/throttle assertions with a valid
0.05m action allowance below the unchanged0.1m minimum useful prefix. Geometry and
production tuning stay unchanged; this distinguishes budget infeasibility from a
valid partial approach rather than declaring all prefixes illegal.

Both coarse recovery traces reproduce repeated short routes with an unused captured
arrival-facing point and no earned fire. Non-personal plans became transit, while
the executor suppresses arrival facing on all transit. Reviewed correction: carry
an immutable full-route-consumed fact through slicing/projection/copies and realize
captured facing only on the final section. Near-goal distance is insufficient to
distinguish a truncated section, so no new geographic tolerance is introduced.

ADS diagnostics confirm saved weapon cosmetic strength0, raw Euler1.125002degrees
and presented Euler0. The fixture used raw full-strength motion as its oracle while
only selecting camera strength. It will own and restore its cosmetic inputs and
retain the0.02degree bound. Current sight rotation is identity; authored sight
rotation alone was not the measured cause. Grip needs actual sprint blend endpoints
under current0.35s tuning. The delayed-shot tracer and endpoints passed; its failed
assertion expects the retired impact globe, which current accepted effects explicitly
hide. Migrate to the real contact-effect owner, preserving tracer and expiry proof.
LowCeiling's original full-suite failure remains unresolved, despite isolation GREEN;
retain its bound and gather physical/contact and motion samples before correction.

### Applied changes and current verification

Rooted-away admission RED and close-contact GREEN were retained before the guard.
Additional geometry RED:3/3 cases failed on missing final-section metadata, and the
native intermediate/final transit case failed its actual final-facing angle. The
minimal rooted guard and immutable `CompletesApproach` fact are now applied. They
do not alter attack tuning, paid Root-after-lunge semantics, route ceilings or
coarse information authority.

Added actual supported moving-fight scenarios for shooter, large elite and melee.
They physically move the player85m at2.5m/s while the autonomous paid enemy graph
follows/fights, requiring native enemy traversal beyond the original48m anchor,
actual attacks and one ledger contribution. Melee additionally observes immediate
director confirmation, owned straight lunge travel and actual hits. Temporary owned
actors use low outgoing damage for the long mechanical probe; authored assets and
human damage/fairness acceptance are unchanged. Trace files are immutable/unique.

The failure-budget fixture migration is paired with real0.5m prefix execution and
retains original captured-ramp geometry. Motion migration uses an owned neutral
shared definition populated with the captured inline legacy profile. First-person
fixture corrections preserve pose/grip tolerances, own cosmetic settings, replace
retired impact-globe proof with actual contact effect/expiry, and wait for authored
sprint endpoints. LowCeiling provenance (16October plan126,19October directional
landing/Step5 acceptance) requires no invented ceiling landing, not an invariant
40mm composed offset after stronger landing authoring. The migrated case observes
actual motor Above/Below contacts, zero ceiling-created landing, one descending
floor return/onset, authored channel cap,120mm composed safety envelope, authored
attack timing and the unchanged2mm settling bound. No runtime feel retuning.

The C3 development observer now reports actual engagement phase separately from bolt
activity and retains ledger group identity through assignment-free Breakoff. A
fenced host command exercises genuine20/2/3 timing, distinct peer-lifetime peeks,
accepted incoming damage, legal fire during full Breakoff and a new attempt.
Controlled acceptance tests reject foreign identities, missing samples, gaps,
shortened/refunded deadlines, invalid sight-only peeks and incomplete confirmations.
This is controlled accepted placement, not moving-fight or human escape proof.

Correction batch `e` stopped **before discovery** on two CS1061 test-access errors;
no cases ran. Tests now use the established reflection access to the internal
safety service; no production API widening. Retained compile-blocker artifact.
Fresh `f` full EditMode is **1652/1652 GREEN**,0skipped/inconclusive, matched terminal;
request `a3e63df0…`, source
`BEE63CE991A4BD1AB6C6E8C32511260A0FEC7238EB6D365678569B6A505D678A`, compiled
`81767B12688FC02B74CCAF8E4CA845402C5B27C29B27A99A0621F0CF42BE1BE4`.
The `f` batch completed14exact PlayMode cases:8passed/6failed,0skips, all matched
terminals. Rooted pair, both real recovery cases, positive firing prefix, ADS,
contact-qualified ceiling and actual delayed-contact effect passed. The six failures
were fixture setup: cycle read before paid registration, unattainable mixed selection
in the elite-only choice, intermediate transit with travel-facing mistakenly enabled,
facing-route allowance greater than the sub-minimum travel allowance, and the remaining
250ms sprint-release wait under350ms authoring. Source, failures and timings remain
in `corrected-f-summary.json`. Corrections wait for confirmed handback; no test
assertions, runtime bounds or authored geometry were weakened.

Final delta review: gameplay found no blocker. Technical requires an explicit
bounded acknowledgement window for the new longer C3 command (existing default25s
cannot own20s+full3s+setup); preserve25s for old commands, existing60s operation and
420s cohort ceilings. Design requires actual admission/release counters to remain
unchanged throughout the sight/damage-only Pursue interval, with exported negative
contracts; a range check only at the sight receipt is insufficient. Both were applied
after handback and independently rechecked without further blockers. C3 starts after
the prior real burst reaches Recovery, then freezes/export-checks both attack counters
through the original attempt's expiry. Extracted actual launcher function tests prove
the25s default, explicit60s engagement window and61s rejection. Controlled phase
contracts reject admission/release counter deltas. Whole-cohort420s stays unchanged.

The corrected `g` full EditMode is **1652/1652 GREEN**, exact1652,0skips/inconclusive,
matched terminal, source
`F6AC7BADB491C2275879E184B255364702452B99DB745B3585B94C690780F6B8`, compiled
`C185A59E266557F76446E6D3F11741C6EB9538CA28AFF08BFFB9A6617A6C0453`;
request `2cab28bbcf40480bbcc79bea01dffacd`. Six corrected native cases are now owned
by the operator on frozen source. No packaged-player gate has yet executed on this
source. Automated qualification remains open pending actual moving-fight/affected
native proof, matching builds and real cohorts.

### Final freeze and remaining execution packet

`g` finished1652/1652EditMode and all six corrected native cases GREEN, with matched
terminals and no skips; `final-g-summary.json` retains each request. The independent
technical review subsequently approved the new C3 far-control precondition: supported
distinct peer positions must be farther than attack range plus the complete20s
physical travel allowance and2m margin. This refuses a setup in which ordinary movement
could legitimately earn a new attack. Movement, actor tuning, deadlines and native
queries remain enabled and unchanged. Attack counters independently guard the actual
sight/damage-only interval. This last delta is confined to the development peer command
and its controlled acceptance helper/tests.

Final frozen source is
`2778462D7B47B44DA0E1BFEA8DC103E4DB0A14B800043E72789B0AD0B1CD6398`.
`integrated-h-owned-source.json` and `integrated-h-owned-delta.diff` capture23owned
source/meta/tool inputs against the preserved dirty baseline. Independent audit
rehashes every other original inventory entry and finds no missing/changed file,
including assets, metas, packages and settings. Review finds no additional omitted
native boundary: remaining lifecycle/capability/search/retrospective cases use
unaffected counter, unrooted, nontransit or no-arrival-facing branches.

Final full EditMode `h` is1652/1652 GREEN, no skips, matched terminal; compiled
`A7833F9164A99BD276DE2C2B7C959EB04A67E477650FB3BB6B3CEF2004797761`.
The exact ten-fixture affected native selection discovers **311leaves** (29cooperation,
17investigation,68navigation,28retirement,40shooter,10aim,23melee,64pressure,
25replacement,7melee-review), request `d8ed07dac9ab48a8ad26462a4ed03323`, currently
running. Compiled `4CFF527792DD97259ADDB19554A4E907ED99DD65D24BD724F80EEF2C5E0C7D6E`.
Earlier rough300case estimate was arithmetic only, not a selection/ceiling. The exact
full-fixture selection is unchanged. Estimated runtime~20–21minutes from actual
baseline case durations and the added physical/root probes.

`h` terminated **310/311**, zero skips/inconclusive, matched terminal and clean
cleanup. The sole failure was cooperation's
`EliteUsesRealGraphThenOwnSightToFireAndRetainsSingleContribution`, waiting for
>.3m actual report/coarse graph movement within its unchanged4s bound. Full output
and identities are retained in `integrated-h-summary.json`; the four-millisecond
assertion polling overshoot is not a root-cause explanation.

Read-only owner/native diagnostics were added to that leaf without changing
geometry, readiness, assertions or bounds. Isolated `i` passed1/1 and ordered full
cooperation `j` passed29/29, zero skips, matched terminals, source
`0B4F7C547D0CC26A9F2866A390E76D08B6D2927C1FCE3626E8D56D4F7FBD63DD`, compiled
`B11745BB145983E90DCFA0B603F8D8BE204F827BF6DB8EAC575D1EE93C9082EC`.
Both showed actual supported Regroup movement, no native violations/rejections.
H's elite spawn24 differed from isolated3/ordered25; spawn%5 affects assessment
scheduling, not candidate geometry. Neither later pass explains H's missing
movement. Preserve it as an intermittent unresolved qualification finding, not
a waived or fixed failure. No further unchanged rerolls are authorized by these
results. `coop-diagnostic-i-summary.json` and `coop-order-j-summary.json` retain
complete output, selection and clean Editor handbacks. Build/cohort gates may
proceed independently; final automated GREEN still requires honest reconciliation.

Prebuild correctly stopped before any build/move/save on dirty loaded
QualitySettings and EditorBuildSettings (no dirty ProjectI assets/scenes). Lead
loaded independent saved objects through the Editor serialization API and compared
their complete Editor JSON with the live objects: both exactly equal. Immutable
`*-live-prebuild.json`/`*-disk-prebuild.json` snapshots retain that check. After a
second equality guard, only those redundant dirty flags were cleared; no save,
reload, authored-value change or filesystem edit. Build ownership then returned
to the operator for the same three-build packet.

The native evidence union is materialized per exact current leaf in
`native-coverage-union-k.json`:656current leaves,314superseding executions and342
compatible original passes. It explicitly retains H as unresolved and the
original full run as RED. This is not a new656-case suite execution.

Normal development build succeeded in77.666456s,0errors/704warnings, public build
GUID `716ddf69e0d744148bb504473fc46abe`. Unique output is
`Builds/EngagementTask4-20261023-normal-development-357fcab432304aa0a1e7e2c0d7f7f5f2`.
The build packet `builds-357fcab432304aa0a1e7e2c0d7f7f5f2/` retains the public
summary,8479packed entries, steps, raw report, binary/data identity and unchanged
2503input hashes. Previous Prototype and DebugArena outputs were restored with
zero hash mismatches. Before C3, the normal build's dirty PlayerSettings flag
caused another preservation stop. Complete independent disk/live serialization
again proved equal; `PlayerSettings-{live,disk}-postnormal.json` retains proof.
Only that redundant flag was cleared. For remaining builds, identical dirty
project-setting flags may be cleared only after the same complete equality guard
and unique snapshots; any actual authored difference still stops execution.

C3 development succeeded in 23.3246573 s with zero errors / 640 warnings, GUID
`175edcf64af24ba5ac25f1dc747c515a`; release succeeded in 53.5490333 s with zero
errors / 1026 warnings, GUID `811cc62ea9ef41f5a710907a6ac030b8`. Their unique
outputs use the same build suffix and `c3-development` / `release` names. All
2503 frozen input hashes remained unchanged. Temporary M5BBuildOnly assets/metas
were removed and the original prefab registry restored byte for byte. All three
previous fixed build folders were restored with zero inventory mismatches.

The operator reported model capacity failure during final bookkeeping, then
explicitly confirmed all tool cells and shell sessions terminated, completed
archives/restoration, and returned ownership. Lead independently confirmed the
correct stopped/idle Editor, successful compilation, no dirty ProjectI assets or
settings, clean SkyboxBPreview, options true/3, VSync0 and zero Console errors.
`build-exclusion-k.json` records the outside-Editor Mono.Cecil audit: all three
profiles pass; development observer/overhead/scenario/diagnostics, nested
EngagementProof/EngagementPeek and inspection deadline getters are absent in
release. Retired policy types and non-script debug fixture assets are absent.
Empty MonoScript references are reported separately.

Formal `Test-TestEvidence` initially returned 23/24: its extracted owner-launch
function lacked the newer optional `FirstPersonFeel` variable. Lead inspected the
actual function and supplied its existing default false in the controlled fixture,
preserving all original menu/simulation assertions. This is the only difference
from build inputs and cannot affect player compilation/assets/settings. The
corrected script passes 24/24; both director contract scripts pass. Original RED
output is retained in `controlled-contracts-k.json`, corrected proof in
`controlled-contracts-l.json`. Lead inspection is proportionate for this one-line
fixture initialization; no runtime change or additional player build is needed.
`builds-and-contracts-l-summary-final.json` and `cohort-freeze-l-inputs.json` bind
the completed build proof to the final controlled-tool correction.

The previous operator encountered capacity failure again before any cohort
started. Lead confirmed no player process and no new director-run directory;
exclusive execution was reassigned to a fresh Sol6.1/Medium operator with the
same negative / two-peer / four-peer packet. No model substitution, new retry
budget or gameplay acceptance is inferred.

The first C3 negative cohort,
`TestResults/enemy-director-fcfbf29189554689b02488460690ca9d`, exited1 before the
deliberate assertion: online host readiness timed out at60.1099s with Unity
ServicesInitializationException / DNS failures for the cloud configuration and
telemetry hosts. This is infrastructure failure, not expected failure proof;
checks were empty, evidence accepted=false, identity unavailable. Actual host
build GUID matched the C3 build. Cleanup passed with no errors, shell36008 was
terminal, no ProjectI process remained and all2503input hashes were unchanged.
Positive cohorts correctly did not start.

Lead's auto-reviewed DNS command outside the network-restricted sandbox resolved
`config.uca.cloud.unity3d.com` to34.111.113.40,
`cdp.cloud.unity3d.com` to34.107.172.168 and
`services.api.unity.com` to34.160.10.162. This establishes a changed external
execution condition for one retry of the same negative / positive2 / positive4
packet. Player launchers now request network-accessible execution through the
normal approval tool; no proxy workaround, spending, code or deadline change.
The failed original run remains immutable and no Natural campaign budget is reset.

### Packaged report-source blocker and bounded correction

Network-accessible negative cohort
`enemy-director-46f6d674b0fa44ea88f7b29578127ded` reached the intended structured
S4 assertion: child exit1, all7checks passed, verified C3 GUID/data/helper identity,
75/15/1 readback and clean owned-process exit. Positive2 cohort
`enemy-director-516314318f7f4bb9b80961e5f5c61820` then failed its original20s
report-movement / personal-release wait after12preceding checks. Four-peer execution
correctly stopped. Both shell sessions96012/77490 terminated; elevated process
inspection found no remaining ProjectI player; cleanup had no secondary errors;
all2503 frozen inputs were unchanged. The failed positive remains accepted=false.

The actual elite route was admitted at17.347287 from original report source2,
around the carved cover, and expired at17.370339 after0.075275m. Its evidence-bound
route deadline was23.321631 and pursuit deadline35.792337; native safety recorded
no rejection. Final actor position showed later movement but53planning calls and
two admissions / zero releases. C3's first exact accepted-plan receipt remains
strictly unqualified; later movement cannot be substituted into that receipt.

Lead traced movement validity to the single latest-report slot: a newer valid
report from another observer revoked an already admitted route even when its
original witness, scope and expiry were still valid. All three independent
technical/gameplay/design perspectives approved a deterministic investigation
and the preserved contract. New actual native regression
`AcceptedReportRouteSurvivesOtherLiveSourceWithoutRenewalAndStopsWithOriginalSource`
reproduced the cancellation after genuine newer B evidence, despite A remaining
valid, original pursuit time and physical support: RED0/1, request
`6de1ee4680ed4b7284a76c74833d6035`. This establishes causality beyond the packaged
trace. Its complete artifact is retained under `foundation-runs/engagement-task4-report-source-red-20261023m-…`.

Correction is limited to EnemyPerception and EnemyMovementContext. Contexts capture
the original trusted source ActorContext and original sender/recipient-capped
expiry at issuance, retain them through route/context copies and validate that
exact authority independently of newer pending reports. Central validation keeps
target eligibility, supported original witness, identity/context, network,
encounter/run/stage, ledger/group or live cooperation, and original TTL. Every
existing action, cycle generation, assignment, route, deadline and physical gate
remains. B cannot substitute for A or renew A. Alert orientation retains its
previous latest-cue behavior and reports still cannot authorize firing. Independent
post-source reviews found no blocker and requested additional captured-authority
expiry/restart proof. No C3 observer acceptance predicate was weakened.

The same native regression is now GREEN1/1, request
`c1afd5a2b33f4d49a3a057473a5ed35e`: actual additional supported travel with unchanged
route/evidence/deadlines/cycle, then A despawn revokes the route while B remains
valid. This focused result does not yet qualify the full runtime correction.

H's separate fixture ambiguity was measured with temporary controlled
report-before-first-plan execution on the exact original physical wall and owned
NavMesh. It returned actual NoPath, captured genuine SquadVisualReport,
six candidates / five path queries and zero travel: diagnostic RED0/1, request
`7f33042638784f87a371f3b58a5567d0`. Unlike the earlier coarse-before-report passes,
this demonstrates the missing navigation precondition. Technical and gameplay
reviews approved adding the existing Box NavMeshObstacle/carving configuration
used by the other C3 report fixtures. The physical wall, positions, dimensions,
4s movement and6s personal-fire assertions remain unchanged. A bounded native
detour precondition is now observed before that movement check. No authored scene,
terrain or navigation data was edited or rebaked. The temporary diagnostic was
removed after retaining its RED artifact.

Three further variants of the genuine A/B route regression now target shorter
sender expiry, shorter recipient expiry and same-lifetime host restart. Each must
preserve actual post-replacement travel and original limits, then reject A while
B remains valid. Current proof selection is full EditMode, full cooperation33,
pressure64, investigation17, retirement28 and the exact navigation case
`ReportsAndHeardRegions_RetainTheirPermissionsAndOriginalPrecision`. Technical
caller audit found no additional changed lease path in the omitted perception,
shooter or search import/snapshot/personal-fire cases; compatible earlier proof
remains reuse. Matching player builds and packaged cohorts are invalidated by
this runtime correction and must be rebuilt/re-executed after native qualification.

### Current native proof after the report correction

O full EditMode passed 1652/1652, zero skips, matching terminal, 37.903 s;
source `6339F24467046DBE3A5759B5F3253C178661450511E47AFCB390D07FC84E7B58`,
compiled `D3689364EEA61A9FD5F87147BF98BDD6A08A73D14B4DCB19187C9C552486402F`.
The four-fixture native run finished 140/142 in 612.676 s; the exact navigation
caller passed 1/1. Native compiled identity was
`7843254A779E6B4B446F29F4ABEE15C387152401A521A0687ED318F40E297E23`.
Both failures were the new shorter-expiry fixtures ending about .36 s before the
captured cap, so they had not proved expiry. Cooperation's other 31 cases passed,
including the corrected original graph case, A/B continuation/despawn and
same-lifetime restart. Investigation17, retirement28 and pressure64 all passed.
`integrated-o-summary.json` retains exact per-leaf results and all failures.

To isolate expiry, the two fixtures now apply native Root to the owned recipient
only after genuine additional post-B travel. This excludes physical prefix arrival
while preserving the source, evidence, original cap, cycle and all authority gates.
P reached the original time cap but failed a newly added overly specific
`Expired` outcome assertion: the director legitimately cancels an invalid lease
through its existing Cancel API. No runtime correction was needed. Those failures
and the fact that later B-validity assertions were not reached are retained in
`integrated-p-summary.json`.

Q verifies the original cap directly, continued Root through that cap, the exact
stopped A context/token, A's live authority, still-valid fresh B evidence and
unchanged cycle/deadlines. Both shorter-sender and shorter-recipient cases passed
1/1, zero skips, matching terminals; requests
`a12da93bc172481aa8f020030e5fad17` and `770bfe9e73fd45638c9159b18d6eb2d5`.
Source `AA1C510D563B1055BFB3EDA70304233A0FA11DDC91522526CAF7336FF3F43986`,
compiled `BF58E6E6277D7054970965AB9D8788AFBD93C9CD6A2D06FBAE2A896F24F866DD`.
Only the two fixture controls/oracles changed after O; runtime and EditMode inputs
remain compatible. No broad suite was repeated for those fixture-only changes.
All 2503 frozen inputs matched at handback; Editor/settings/preferences were
restored and no external player remained. See `integrated-q-summary.json`.

`native-coverage-union-r.json` verifies 660 unique current PlayMode leaves with
passing evidence: 342 compatible baseline passes, 172 unaffected H passes and 146
superseding O/Q/first-person executions. This is not a fresh full-suite pass.
Original 642/649 RED, H RED, O RED and every diagnostic/correction result remain
immutable. Independent caller review found no additional invalidated case.
`integrated-o-owned-source.json` captures 27 owned inputs; independent/lead audits
find the other 2476 original inputs unchanged. The temporary diagnostic is absent
from the final 660-case inventory.

The human checklist now includes coherent report-source changes during approaches
and Root before distant admission, at close contact and after a paid lunge. These
remain gameplay observations, not inferred acceptance from the native results.
Fresh matching builds are now owned by a separate clean Sol6.1/Medium build operator
on frozen Q input; no other Editor/test operation overlaps it.

### Matching R builds and release exclusion

The build operator completed and explicitly returned exclusive ownership with no
pending tool/session/build/player. All three frozen-Q Windows builds succeeded
with zero errors; warnings remain recorded rather than treated as zero:

| Build | Application GUID | Warnings | Build duration |
|---|---|---:|---:|
| Normal development | `948a2dd3916b4cc88ca4c419c37225de` | 692 | 79.415 s |
| C3 development | `71fc8edd774447499d620bd28b68ac98` | 640 | 23.558 s |
| Release | `a0342deca88041ccaf2c332ce61ba22b` | 829 | 51.143 s |

`builds-r-b3697e6ba27347a599c9f36f78e8b142/final-manifest.json` retains public
reports, exact output/whole-folder/data identities, actual scenes and packaged
fixture provenance. All2503 frozen inputs matched. Each original fixed output
folder was restored byte-exact; temporary C3 Resources assets/metas were removed
and the original network registry restored. Editor returned clean, stopped and
idle with the original scene, play options, preferences and empty Console.

Lead's outside-Editor Mono.Cecil/packed-asset audit passed in
`build-exclusion-r.json`: executable development observer/overhead/inspection types
and deadline markers are present in development and absent in release; obsolete
policy types are absent throughout; release has no non-script development fixture
assets or test scenes. Empty source MonoScript references are reported separately.
The unchanged controlled contracts retain their compatible L passing evidence.
These builds supersede pre-report-correction builds for packaged execution.

Exclusive ownership is now assigned to the peer operator for serial R C3 negative,
positive two-peer and positive four-peer cohorts. Network-accessible execution is
required by the retained DNS diagnosis; source, limits and acceptance stay frozen.

Final technical/Ponytail and design/UX read-only delta reviews approve Q's source,
native compatibility union and matching build/exclusion evidence. Neither finds a
new material blocker; both require actual packaged results/capture inspection and
retain human gates. Technical review confirms that the27 Task4 inputs cannot form
a coherent source commit against HEAD without the uncommitted prerequisites from
earlier slices. Preserve this dependent checkpoint; neither commit broken isolated
source nor silently accumulate earlier work. A scoped evidence/document checkpoint
can be assessed after the agreed automated gates finish.

### R packaged result and T graphical-launcher correction

`cohorts-r-summary.json` retains three serial matching C3 executions. Intended
negative `enemy-director-a06844ca47c64d90ad2b9c5125ca3b01` is valid: exact structured
assertion, child exit1,7checks and clean exit. Positive two-peer
`enemy-director-d841ea91a086446dbeae4b023912d5b3` passed94/94, accepted=true,
child exit0,195.535s,46 retained PNGs. The previously failing original accepted
report route now progresses to personal release. Actual engagement proof retains
20.000s, unchanged counters through both distinct peer peeks/incoming damage,
full3.000s Breakoff, zero attempt/seek violations, legal fire during Breakoff,
two-second continuous confirmation and a fresh20s attempt. Lifecycle, targeting,
disconnect, rehost and abrupt host exit also passed.

Four-peer `enemy-director-0e281a4032a04f80814a47241874da7f` failed before its first
saved stage: authoritative host receipt for peer3/spawn2/network8/lifetime2 was
observed16.969079491093943 versus original ready16.965474848366064,3.605ms late.
This is RED, not an intended failure or a timing waiver. All25preceding checks,
16PNGs and exact host receipts are retained. Disk notes have zero clock
placeholders by design; authoritative time comes from host/session.json.
All three shells terminated, cleanup had no errors, no players remained and
all2503 Q inputs matched. No retry was authorized without further evidence.

Lead inspected pixels for both two-peer initial pack arrivals, original elite
arrival, boss/reinforcement arrival and post-route elite captures. Pack brackets
and exposed post-route elite/projectiles render; the initial elite is occluded by
the intentional wall. Some later directed arrival frames show an old actor or
wall/floor rather than their named arrival. They are retained capture limitations,
not visual passes or proof of natural noticeability. Receipt/capture timing and
effective camera matrices need measurement before changing that path.

Independent remaining regression batch S stopped on its first negative setup:
`debug-arena-8a74b428cb1341ab9cf5cbb6fc6a4c76`, shared Idle host20s timeout,
child exit1. Every other readiness field matched; both peers had one world camera
and zero overlay, ownerCameraLayoutValid=false. This did not reach the deliberate
assertion. `regressions-s-summary.json` preserves primary failure, clean exit,
unchanged Q hashes and clean restored Editor; subsequent gates were not launched.

Source diagnosis: DebugArena and EncounterPolish explicitly passed-nographics
while requiring the URP owner world/overlay stack. FirstPersonViewmodelCamera
declines creation when the renderer supplies no stack; the readiness assertion
correctly rejects that state. Technical review approves removing-nographics from
those two launchers only, retaining hidden batch operation, every readiness check,
original deadlines and actual graphical layout. MainMenu does not require that
stack and is unchanged. No player rebuild is needed for this Tools-only correction.

Actual extracted Launch/Start-Instance contracts first failed for both graphics
arguments (24pass/2fail); the first test attempt also exposed/fixed an overlong
controlled profile name. After the two launcher changes, all26contracts pass.
All RED/green outputs are unique `camera-launcher-contracts-…-t*.txt` artifacts.
`integrated-t-frozen-inputs.json` confirms only those two Tools scripts and their
contract fixture changed; the other2500 Q inputs and R player bytes remain valid.
The corrected negative/positive arena and normal menu/Verdant packet will run
serially after final Tools review. C3 four-peer qualification remains RED.

### U measured camera census and bounded capture investigation

T graphical negative `debug-arena-808d69da2e654a4f9129a1cf1602f33a` also failed the
same initial Idle20s predicate: graphics were actually active (D3D11 RTX3080Ti),
but reported ownerCameraLayoutValid=false with zero counted overlays. The Tools
correction was necessary for graphical proof but insufficient; it did not resolve
the census defect. No remaining cohort followed. `regressions-t-summary.json`
retains exact identity, clean terminal/cleanup, unchanged2503 T files and restored
Editor. No timeout, assertion or failure-proof requirement was relaxed.

Lead then measured the discovery API directly with one owned temporary active
DontSave Camera: direct active=true, Object.FindObjectsByType contains=false,
Camera.allCameras contains=true, Resources.FindObjectsOfTypeAll contains=true.
It was destroyed in finally; SkyboxBPreview remained clean before/after. The real
viewmodel camera is deliberately DontSave. DevelopmentSessionProbe's object search
therefore omitted the legitimate overlay, while the existing ownership tests
supplied manual arrays and never tested that inventory boundary.

Technical pre-review approves the responsible development-only correction: use
the actual active-camera inventory, retaining the exact owned world/overlay,
extra-camera, listener, stack and mask predicates. New
`ActiveCameraInventory_IncludesUnsavedOverlayAndExcludesDisabledCameras` uses real
enabled DontSave and disabled/unrelated cameras. Extracting the unchanged object
search first reproduced RED0/1, request `5c3ec20a71b84d0bbbb885ae5a59f40f`, matching
terminal/selection, zero skips. Its expected missing-overlay assertion failed.
Camera.allCameras then passed1/1 with matching terminal. A prior NUnit collection
overload compile error was corrected before this causal RED; it is not test proof.
No production camera, rendering policy or ownership acceptance changed.

The unexplained late receipt is investigated separately. Before altering capture
order, timing or framing, U adds bounded development-only read-only measurements:
render/readback/pixel-validation/PNG encoding/disk-write Stopwatches; runtime frame,
FPS/VSync/delta; submission and effective begin-render pose/view/projection/target
viewport and retirement-view status; note-publication and host-read UTC/frame/cost.
The temporary render callback is removed in finally. Actual capture then atomic
receipt publication, original ready deadline and every acceptance assertion remain.
These measurements can distinguish synchronous capture cost, host polling and the
existing remote retirement render fence; none is yet assumed to be the cause.
No frame cap, early publication, extra campaign budget or timing waiver is added.

This diagnostic-only delta is medium risk and receives relevant independent
technical review; earlier three-perspective gameplay source approvals stand for
unchanged behavior. U full EditMode and full native viewmodel fixture are assigned
to the exclusive operator on `integrated-u-frozen-inputs.json`2503. Only the two
development C# files and ownership fixture differ from T; the other2500 inputs
match. Rebuild normal/C3 development after proof. R release may be explicitly
reused for unchanged excluded inputs, never described as a fresh U release build.

U full EditMode passed1653/1653 in44.610s, request
`b3915edaa8e84cd78418e4c7eb89bf99`; the full discovered native viewmodel fixture
passed4/4 in14.061s, request `ad7e495856c042ecb33569c6dcc73c14`.
Exact discovery/started/XML leaves and terminal identities match, zero skips,
conflicts or errors. All2503 frozen inputs match and Editor/settings/preferences
returned clean/stopped/idle with no external players. `integrated-u-summary.json`
retains complete identities and per-leaf results. The diagnostic census does not
invalidate the other660 native gameplay leaves; the four viewmodel leaves now
have fresh U superseding evidence. Technical post-review approves the census and
bounded instrumentation, retaining the publication-start timestamp interpretation
and all remaining packaged gates. Exclusive ownership transfers to the build
operator for two fresh development outputs only.
`native-coverage-union-u.json` retains660 current leaves:338 compatible baseline,
172 compatible H and150 superseding executions, including the four fresh U
viewmodel cases. Original full-suite and subsequent RED artifacts remain intact;
this is still a compatible per-leaf union, not a fresh full660-case pass.

U development builds completed with zero errors: normal GUID
`aca775305c4040809bf6191d759e8f4c`,704warnings,72.614s; C3 GUID
`05bc4dcfb7054f968e3e0d14de165d49`,640warnings,26.301s. Unique manifest
`builds-u-c00a87577a094ce1b1faa69eb12cd612/final-manifest.json` records exact
outputs/data identities, actual scenes, temporary fixture provenance and restored
fixed folders/registry. All2503 U inputs match and Editor returned clean/idle.
Outside-Editor `build-exclusion-u-compatible-r.json` passes executable/asset checks,
including new CaptureTiming present in U development and absent in R release.
The release row is explicitly R compatible reuse, not a fresh U release execution.
`integrated-u-owned-source.json` and delta capture31 task-owned inputs;2472 other
original inputs remain unchanged.

One instrumented four-peer C3 diagnostic is authorized on U's exact build to
measure the previously unexplained arrival/capture path. This bounded investigation
does not assume the late-receipt cause fixed, reset any Natural budget, alter the
420/60/25s ceilings or replace the failed R artifact. Existing R negative and
two-peer mechanical passes remain compatible for unchanged behavior; their pixel
limitations remain explicit. The operator owns this one run and must return clean
terminal/identity/frozen state before other packaged checks.

U four-peer diagnostic `enemy-director-da1f566363424ee2a4cf930ec10fc184` passed
126/126, accepted=true, child exit0,221.857s,92PNGs; all2503 inputs and matching
build identity remained exact, zero cleanup errors and no remaining players.
Its initial authoritative receipt minimum lead was+.621660s. Actual20/2/3 proof
and full target/disconnect/reset/graceful/abrupt-host-loss/rehost tail passed.
`cohorts-u-diagnostic4-summary.json` retains complete timing, matrices, source/build
and lifecycle evidence. The four-peer gate passes for this current build; the
earlier R3.605ms late receipt remains unexplained historical failure. No causal fix,
repeatability/performance guarantee, waived deadline or additional C3 run is claimed.
Independent technical evidence review accepts that bound and the compatible R
two-peer mechanical proof; no further reroll is justified.

Measured initial captures took about46–62ms, with roughly37–44ms PNG encoding.
Submitted and effective target viewports agree and retirement status was inactive;
this does not support the speculative retirement-fence explanation. Capture audit
found278 snapshot birth rows/68 distinct events with zero conflicting path reuse.
All76 operator-preserved PNGs remained byte-identical; lead's separate92-file
preservation is also hashed and makes no unsupported before-rehost attribution.
No capture naming/render-policy change was needed. Lead inspected all four
post-route elite images (actors/bolts present), initial first-member bracket images
on all four peers, and boss images. Cover/other actors occlude some arrival views;
these images establish bounded rendering, not universal target visibility or human
noticeability. The initial two-peer inspection has the same explicit limitation.

The operator now owns final V regressions on U's matching builds. Corrected arena
negative `debug-arena-110ec6018ff34be2941ede8e769b2a61` reached its exact structured
failure, child exit1, both reset cycles passed, and cleanup succeeded. Its formerly
failing camera/Idle gate passes. Positive two/four arena, then normal menu and
Verdant progression remain pending until their own accepted terminal evidence.

### V arena passes and W normal-entry launcher correction

V's corrected arena negative is valid. Positive two-peer
`debug-arena-fe95b20e040647e5ab3149991a55dae3` and four-peer
`debug-arena-6d39348371754e76bb46b1a62150513d` match the operator's authoritative
`regressions-v-summary.json`. Their actual totals were
42.499s and40.780s: baseline/reset/rehost combat, both reset cycles, intentional
failure proof and all three remote-leave/host-loss/rehost lifecycle checks passed,
accepted=true, child exits0, exact U C3 build/readback75/15/1, clean process exit.

Normal menu `main-menu-abeb338e9e4d4b92b8739e3ed70fa7bc` then failed after three
passed phases (entry, library/loadout and readiness). Its original-nographics
launch reached normal stage start and GPUResidentDrawer initialization requested a
compute-backed Raw GraphicsBuffer, rejected on the graphics-disabled platform.
The retained player stack identifies URP GetRenderer -> viewmodel Update; this is
not an enemy policy assertion. EncounterPolish was not started. All four V shells
terminated, cleanup was clean, all2503 U inputs matched and Editor was restored.

Technical pre/post review approves removing-nographics from Test-MainMenu only;
no production rendering setting, menu/session/gameplay assertion or ceiling changes.
Actual extracted host/join/offline launch tests preserve hidden batch mode, menu
entry, absence of direct session/arena bypass and nondefault83/17/2 transport.
Proper RED26pass/1fail then GREEN27/27 are retained in
`menu-graphics-contracts-{red,green}-w.txt`. W's2503 frozen inputs differ from U
only in Test-MainMenu and the controlled fixture; all2501 other inputs and U
player bytes remain unchanged. No build, native, C3 or arena repeat is justified.
The exclusive operator runs only the pending graphical menu and VerdantInitial
default two/four-player progression gates, retaining failure and cleanup separately.
Legacy MainMenu lacks player GUID readback: bind its actual executable/full built
data/managed identity before/after to the public U normal report, and state this
limit explicitly rather than inventing runtime GUID evidence.

Final W technical/Ponytail, gameplay and design/UX delta reviews approve the
complete final source scope. They retain the compatible native/release labels,
unexplained R timing failure and occluded-capture limits; none requests a further
C3 reroll or unrelated batch. Automated closure remains conditional on the pending
Verdant two/four-player gate. Human feedback and a coherent future source commit
remain distinct from this documentation/evidence checkpoint.

Closure inspection also finds seven pre-existing staged GroundShape-v2 art files.
They are unrelated and must remain staged byte-exact. A local checkpoint must use
explicit documentation paths, preserving that index and the accumulated runtime/
authoring changes. Do not use a blanket commit or include earlier dirty slices to
make the32 Task4-owned source inputs appear independently buildable against HEAD.

W's corrected normal menu `main-menu-d00de658dec94f7fb48885ce8000aa59` passed
10/10phases in119.184s. VerdantInitial
`encounter-polish-e68c36c0ffc94ea9816b67ca0427621b` passed23/23phases in536.553s,
including full two/four-player Stage1->2->3 runs, rewards, builds/scaling,82 unique
team kills, detailed cached results, normal host departure and fresh rehost state
clearance. No FourPlayersOnly/timeout/acceptance waiver was used. Both shells
terminated with child exits0 and empty cleanup errors. All six Encounter player
readbacks match normal GUID `aca775305c4040809bf6191d759e8f4c` and75/15/1.
MainMenu uses the explicit legacy exact-build attribution described above.

`regressions-w-summary.json` retains complete results, phases, pre/post executable/
full-data/managed/helper identities, loadout preference name/type/byte restoration,
player readbacks and final state. All2503 W inputs match, no external player remains,
and Editor is stopped/ready/idle with the preserved clean SkyboxBPreview, play
optionsTrue3, VSync0, cached/saved preferencefalse, FOV74.3, motion strengths0,
frame120 and empty start scene/Console. Final independent source reviews have no
remaining material blocker; their last automated condition is now satisfied.
No further batch, Natural attempt, tuning or roster work is required for this
authorized automated scope. Human acceptance and dependent source commit remain
explicitly separate.

The agreed packaged selection (actual results and reuse are recorded above) was:

1. Preserve fixed existing build outputs in verified workspace paths. Build normal
   Windows development, deliberate C3 development through the existing packaging
   helper, and release exclusion once each into uniquely retained outputs. Keep
   source/compiled/build/whole-folder identities, build reports, scene/packed-asset
   lists, temporary fixture provenance, registry restoration and clean authoring.
   No terrain generation, scene regeneration, dependency upgrade or rebake.
2. Inspect executable managed types/markers with Mono.Cecil outside the Editor.
   Distinguish harmless empty MonoScript references from executable diagnostic
   types and non-script fixture assets. New development observer/nested proof and
   overhead diagnostics must be absent in release; retired policy types absent.
3. Run `Test-TestEvidence`, `Test-EnemyDirectorContracts` and
   `Test-EnemyCoordinationContracts` with immutable outputs/helper hashes. The latter
   two already passed lead-controlled runs; retain formal integrated artifacts.
4. C3 deliberate-failure two-peer invocation must exit nonzero with its intended
   failure and clean process exit. Then real C3 two/four-peer positive cohorts use
   explicit75/15/1 simulation,60s engagement acknowledgement and unchanged420s
   cohort ceiling. No retry without changed evidence/conditions. Inspect required
   captures separately from manifest/image-size success.
5. Debug-arena deliberate failure and positive two/four-peer cohorts retain
   `ResetCycles2` and normal failure-path proof, lifecycle/disconnect/rehost and owned
   process cleanup. Do not disable failure proof or cancel unrelated players.
6. Only after confirmed owned-player exits, run matching normal development
   `Test-MainMenu` and `Test-EncounterPolish -VerdantInitial` with75/15/1 once.
   These retain supported progression and menu/session coverage. Unchanged owner
   input/warp/progression branches do not require unrelated launcher campaigns;
   native full-baseline compatibility is recorded per actual passing leaf.
7. Publish per-leaf coverage/reuse and exact manifests. Keep original full PlayMode
   **642/649 RED**; a union of compatible original passes, affected reruns and new
   cases is never described as a fresh full-suite pass. End with human gameplay,
   keep/tune/revert, hardware and separate-network gates explicitly open.