escapists/EscapistsReintroduced
1
0
Fork 1
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Harden base progression and cleanup; validate finite physical resupply

Browse Source
This commit is contained in:
jakub
2026-09-19 10:39:39 +02:00
parent 40a474934c
commit 7c53b4bb81
377 changed files with 51882 additions and 92 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/validation.md
+63
View File
@@ -268,3 +268,66 @@ At this checkpoint, repository policy/resource checks, 15 synthetic performance-
[085048 local admission](evidence/native-2026-09-19-085048/report.json) passes 54 native assertions without errors. Replayed callbacks preserve the actual local prisoner, key, roster object and spent life without new spawn work. Disposable helper fault inputs latch failure, and a deliberate invalid-controller method argument makes the real production context guard end the round while preserving the original body/life. This is method injection and callback replay in a standalone native session, not a genuine network disconnect or authentication test.
[085302 discovery probe integration](evidence/native-2026-09-19-085302-map-probe-compile-failure/report.json) retains missing native map-marker serializer methods caused by modding the replicated vanilla base class for test access. The probe now uses separate owned plain/objective subclasses and preserves the native RPC type. The production discovery-marker fix is separate from this test access correction; native widget verification remains pending.
[085442 Core restart and discovery widgets](evidence/native-2026-09-19-085442/report.json) passes two complete Core-only escape/restart rounds with 586 assertions and zero engine or script/VM errors. The snapshot includes the standalone admission helper and successor discovery-marker fix. Each round's eleven marker assertions instantiate real hidden native marker layouts, reproduce the vanilla stale-label/annotation-ownership behavior, verify successor immediate label refresh and protected ownership, recreate the widget with its latest label and remove only owned probe widgets/markers. Existing physical clue, radio defence/feedback, objective UI, extraction and restart checks complete. This is local scripted evidence for the archived source; real map input, remote clients, flight and FPS remain outside scope. The separate [085048 admission](evidence/native-2026-09-19-085048/report.json) result remains 54 clean assertions with method-fault injection and callback replay, not new network authentication. Neither pass resolves the earlier companion-only restart catalog error.
[090203 powered-flight pad refusal](evidence/native-2026-09-19-090203-flight-pad-refusal/report.json) compiles and passes 51 preliminary assertions, then fails to find a separate clear flat pad among its twelve initial candidates. Both fixture failure messages and zero completions are retained. No fixture aircraft/pilot, engine start, flight-input hook observation or powered flight is established; the extraction vehicle remains untouched. The refusal motivates a bounded candidate-search revision without reducing physical-clearance requirements.
[090443 HQ proof compile failure](evidence/native-2026-09-19-090443-hq-proof-compile-failure/report.json) stops before runtime, first reporting `Formula too complex` at `ERR_NativeHQResponseProof.c:236`. [090705 pad-search compile failure](evidence/native-2026-09-19-090705-compile-failure/report.json) stops first at `ERR_NativePoweredFlightProbe.c:36` with `Unknown operator '%'`. Each report preserves all subsequent compiler errors and has zero assertions/completions. Corrections require a fresh archived native run; earlier clean Core and ordinary startup snapshots do not qualify this later source.
For these five 085048–090705 archives, console SHA-256 values were checked against runner metadata, and launch/stopped source-manifest bytes and their recorded hashes match. All seventeen release cases and current-source clean runtime remain pending. No actual multiplayer, powered flight, matched-workload FPS improvement or released package is established.
[091306 mortar expiry](evidence/native-2026-09-19-091306/report.json) passes 48 native assertions with zero captured errors. The corrected HQ proof and pad helper compile in this source. A single unchanged synthetic observation traverses the actual radio/HQ/native firing path; two physical shells are committed while its original age is below sixty seconds. A subsequent firing attempt is explicitly withheld for report expiry at 61.004 seconds. HQ continues operating beyond the deadline and through an additional twelve-second stable interval without a new order, new projectile or further ammunition consumption. Commitment is observed before native inventory transfer and linked to the actual emitted projectile, preserving legitimate completion of already authorized operations. This is expiry/deduplication evidence, not real perception, a complete salvo, measured impacts, multiplayer or the later station-cleanup source.
[091641 medical-kit teardown](evidence/native-2026-09-19-091641/report.json) passes 76 native assertions with zero captured errors. Two fixture-owned stock kits initialize in separate owned preview worlds, then only each station's cached catalog reference is deliberately cleared. Actual native resource callbacks stop reaching the station after the scoped production teardown, duplicate teardown does not repeat work, and queued delayed initialization remains cancelled after 500 ms. The global catalog remains unchanged. Both stock kits report zero installed damage-callback sources, so runtime damage-listener removal is not established. This is explicit catalog-expiry injection, not spontaneous expiry or reproduction of the earlier companion restart error; a later complete companion restart is still required.
[091819 expanded flight-pad survey](evidence/native-2026-09-19-091819-flight-pad-refusal/report.json) compiles but refuses all 128 candidates before aircraft creation. Three have low terrain and 125 exceed the unchanged 0.8 m cardinal height-difference limit; the best is 0.803917 m. Cylinder and swept-box checks are not reached. A later bounded local refinement is intended to find nearby suitable geometry without relaxing this limit; no powered flight is established.
[092026 prison breach registration](evidence/native-2026-09-19-092026-prison-registration-failure/report.json) compiles but fails its first gate-registration assertion before any gate damage or boundary crossing. The registration's actual scenario ownership/hierarchy is being investigated; no native destroyed-gate progression pass is claimed.
[092237 genuine HQ response](evidence/native-2026-09-19-092237/report.json) passes 35 native assertions with zero captured errors. A native visual sample from an original source-group soldier becomes report M1, travels through its original physical RTO/uplink to HQ, and is delivered through downlink M5 to a different original squad. The original recipient physically moves toward the current native Search waypoint; the original force/group/actor sets are unchanged. No target, report, radio event or order is seeded by the observer. The real prisoner is deliberately positioned near the source and temporarily protected for this arranged local test. The established death/spectator/defeat ending also passes. This does not establish multiplayer, unrestricted perception quality or later logistics changes.
[092512 companion restart](evidence/native-2026-09-19-092512/report.json) passes two complete companion-loaded escape/restart rounds with 96 assertions and zero captured errors. Teardown diagnostics are disabled. This snapshot includes the scoped medical-kit cleanup, report deduplication and strengthened native proof sources. The prior 084026 failure remains archived; a clean pair is regression evidence, not proof that intermittent catalog expiry was reproduced or cannot recur. The subsequent prison-registration correction and flight-footprint changes need separate native runs.
[092923 prison breach](evidence/native-2026-09-19-092923/report.json) passes 62 native assertions with zero captured errors. The owned prison root resolves normally; its gate child's raw prefab name is empty while the installed resource-name resolver identifies the exact successor door. The corrected owner/action lookup registers the actual gate with health 1000. Native `HandleDamage` deletes that gate; the weak reference becomes null, while a fresh unsubscribed helper refuses zero candidates and cannot grant escape. Destruction alone, a distant jump and outside-only arrival do not progress. A deliberately repositioned bounded inside-to-outside crossing then enters RADIO_SEARCH once, reveals both supply sites and stays free of duplicate completion. This is actual native destruction plus scripted movement, not walking through damaged geometry, held input or multiplayer.
[093115 ordinary Chat regression](evidence/game-2026-09-19-093115/game-startup-report.json.txt) runs the regular game with all five successor addons, no WORKBENCH define or smoke fixture. The matching Chat header initializes MissionSystems, normal preparation reaches PRISON, and the real sole local controller receives a living canonical prisoner. Ready-state observation and owned cleanup complete with zero captured errors and matching launch/stopped addon bytes. The diagnostic noSplash path and standalone round-local identity are explicit. This production-source startup includes the prison correction, but excludes WORKBENCH flight probes; it does not qualify full ordinary escape, normal splash/input, remote identities, multiplayer or FPS.
For the six newly checked 091306/091641/092237/092512/092923/game093115 archives, console hashes match runner metadata and launch/stopped source-manifest bytes and recorded digests match. Existing mortar-expiry, scoped-station, genuine-perception HQ and companion-restart entries above are retained rather than duplicated. All seventeen release cases and current-source clean runtime remain pending. The active 208-candidate/whole-aircraft-footprint flight experiment and private courier draft have no runtime qualification in this checkpoint. The deterministic source handoff tool produces labelled editable development ZIPs only; it is not native packaging or publication.
[093319 enlarged flight-clearance refusal](evidence/native-2026-09-19-093319-flight-clearance-refusal/report.json) compiles the 208-point refinement with enlarged clearance, but stops before creating a fixture aircraft: three candidates fail minimum terrain, 182 fail slope and 23 fail the entity-clear cylinder. The best terrain delta is 0.0999985 m; no candidate reaches the swept-box or measured aircraft-footprint check. The archive retains 51 preceding assertions, two explicit fixture failures and zero completions. Console and launch/stopped source-manifest hashes match runner metadata, and both source manifests have identical bytes. A subsequent draft adds 169 airport-grid candidates after the local 208, for 377 total, and adds native-input diagnostics; its runtime result remains pending. This updates the preceding active-draft description without relabelling that earlier checkpoint. No powered flight, native control callback, real fuel use, multiplayer or FPS result is established.
[093749 flight initialization](evidence/native-2026-09-19-093749-flight-initialization-failure/report.json) finds airport candidate 213 in the bounded 377-point survey, with all terrain/clearance requirements retained, then creates one fixture aircraft and its empty group. It fails the immediate compound native simulation/seat/fuel check after 53 preceding assertions. The later bounded native-component polling correction is separate; no engine start or flight is established.
[094021 courier compile failure](evidence/native-2026-09-19-094021-courier-compile-failure/report.json) identifies the reserved Enforce parameter name `owned`; [094317 courier compile failure](evidence/native-2026-09-19-094317-courier-compile-failure/report.json) identifies incompatible reference-array specialization in `Batteries.Copy`. The parameter is renamed and enumeration uses explicit clear/insert. Cascading errors remain in both original archives rather than being filtered. Neither run reaches gameplay.
[094436 courier setup](evidence/native-2026-09-19-094436-courier-storage-refusal/report.json) compiles the finite courier production implementation and verifies 27 original crew shells plus twelve distinct physical cache shells. It creates three empty fixture stores, then refuses the fourth obstructed fixed point before moving any original ammunition. The 51 preceding assertions are retained with both fixture failures and zero completions. A later bounded search for five clear stores is a fixture correction; this archive proves neither collection nor delivery.
[094843 flight seating](evidence/native-2026-09-19-094843-flight-settling-failure/report.json) again finds the safe airport candidate, creates one aircraft/group, and observes its native controller, valid simulation, fuel manager, 1938.75 fuel and seventeen compartment slots after initialization. It selects the empty primary `PilotCompartment` and accepts one pilot allocation. The subsequent combined ownership/seating/ground-settling gate does not complete within thirty seconds, retaining 55 preceding assertions and two failures. Individual predicate diagnostics are required before changing behavior. Measured footprint, input callbacks, engine start and powered flight remain unproved.
[095513 finite mortar courier](evidence/native-2026-09-19-095513/report.json) passes 154 native assertions with zero captured errors. The bounded fixture survey finds five native stores after fourteen candidates and physically sequesters all 27 original crew shells, retaining their identities. The production low-ammunition observation traverses actual uplink/HQ/downlink; the original western patrol collects existing cache shells, its original bearer walks more than 100 m, and the empty battery receives exact original depot entities. The same immutable status remains fresh at delivery. Completion releases the HQ reservation and resumes the original patrol; all original forces and all twelve finite cache shells retain exactly one physical owner. No artillery firing, synthetic target report, force spawning or ammunition replacement is used. This is a local arranged depletion fixture, not ordinary player play, full stock exhaustion, interrupted courier loss or multiplayer acceptance.
[095950 flight occupancy diagnostics](evidence/native-2026-09-19-095950-flight-occupancy-refusal/report.json) retains the seating failure while identifying its predicates: the owned living pilot belongs to the expected group, the active primary seat matches, and the aircraft is settled/ready, but that seat is empty and the pilot has no vehicle. The group unexpectedly contains six agents after queued native initialization. Installed source reveals the supposed empty USSR base group authors six default members and queues their creation, so the immediate zero-agent snapshot was insufficient. A memberless vanilla group and explicit owned pilot activity correction require a fresh run; no powered flight is claimed.
[100148 prison action boundary](evidence/native-2026-09-19-100148/report.json) passes 67 native assertions with zero captured errors. Direct `PerformAction` calls with the original guard and canonical prisoner cannot open the locked native gate or advance progression. Actual native gate destruction, missing initial registration refusal, far-displacement/outside-only rejection and the bounded physical crossing then pass. The same run checks rough/exact coordinate storage on one native discovery marker. It precedes the disabled-marker cleanup and later identity fault-injection/casualty maintenance changes. Body movement is scripted; no walking, held input or multiplayer claim.
[100422 one-pilot/input diagnostic](evidence/native-2026-09-19-100422-flight-input-refusal/report.json) establishes one living native AI pilot physically occupying the single owned fueled helicopter, with both native preparation hooks and later control-application observations. The group correction avoids queued default soldiers. Stage 2 refuses the assumed signed collective minimum: a written -1 reads back as 0 while Direct mode is active. The simulation's documented internal-physics transform remains identity and is unsuitable for world-space altitude/drift. The next fixture revision maps its bounded signed feedback demand into native [0,1] and reads the actual aircraft entity world transform, retaining all envelope, drift, altitude, contact and landing limits. No engine start, takeoff or powered flight is established by this failure.
[100619 integrated Core regression](evidence/native-2026-09-19-100619/report.json) passes two complete escape/restart rounds with 612 native assertions and zero captured errors. Each round proves canonical-actor fault rejection at normal unlock, held targeted-document rejection without spending/discovery, immediate successful recovery after exact restoration, coarse-to-precise native marker coordinates, actual disabled-list cleanup and duplicate deletion safety. The existing finite stock, objective presentation, unconsciousness/recovery, radio interruption, extraction/result and restart sequence remains successful. The roster mismatch is synchronous explicit fault injection, while travel/extraction are scripted; this is not ordinary manual play or remote identity/reconnect qualification. Casualty pending maintenance compiles here but its empty-server timing remains covered by the actual-method harness rather than a real disconnect session. The subsequent flight-only world-pose/collective revision requires its own compilation evidence.
[101342 native engine and rotor test](evidence/native-2026-09-19-101342-flight-drift-refusal/report.json) passes the complete measured airframe/rotor envelope (110 entities, 93 meshes, 74 zones; radius 27.6823 m) inside the unchanged cleared pad. One occupied native pilot starts the engine through one normal request; actual rotor speed reaches 196 and fuel falls from 1938.75 to 1937.33. The run then fails when horizontal displacement reaches 9.223 m, beyond its unchanged 8 m limit. Crucially, matched-sequence input diagnostics show positive collective/cyclic demands overwritten to zero before native application; rotor signals agree. The earlier negative-input finding was therefore not proof of a complete numeric-domain diagnosis. This source proves engine/rotor/fuel operation, not controlled takeoff, hover, landing or an HQ aircraft driver. A final-callback correction and independent positive rotor-signal check require fresh native evidence.
[101611 integrated companion regression](evidence/native-2026-09-19-101611/report.json) passes two complete companion-loaded escape/restart rounds with 96 native assertions and zero captured errors, with station teardown diagnostics disabled. It includes finite courier production, prison direct-action/crossing corrections, canonical unlock/discovery guards, coarse discovery markers with culled-list cleanup, and casualty pending-ticket maintenance. The separate flight experiment is opt-in and unused. The subsequent OnApply flight callback revision and later ordinary-game controls need their own source-bound qualification. This is scripted local integration, not manual gameplay or multiplayer acceptance; the historical catalog failure remains retained.
[102037 casualty-service regression](evidence/native-2026-09-19-102037/report.json) passes 101 native assertions with zero captured errors. The actual incapacitated local player remains canonical; an explicitly disclosed helper-only membership adapter permits the arranged NPC rescuer. Real seat contention, a spawned line-of-sight obstacle, completion-time rejection, the native owner handshake, one physical loading operation, released reservation and idempotent Stop all pass with the new pending-maintenance callback installed. The unadapted helper is rejected both before and after entering extraction. This does not simulate a real remote rescuer or last-client disconnect. Empty-population lease expiry remains qualified by the actual-method regression harness, while the live normal loading/cleanup path is covered here.
[102235 final-callback flight refusal](evidence/native-2026-09-19-102235-flight-consumption-refusal/report.json) compiles the sole OnApply input writer and repeats the physical footprint, occupied pilot, normal engine/RPM and fuel observations. Its mandatory positive-input stage refuses to begin climb: five fresh observations over 2.5 seconds show requested collective 0.05 and 15/30/45/60/75 real apply callbacks, while the native main-rotor collective signal remains zero. Immediate context write/readback is insufficient evidence of consumption. No height/drift/tilt guard is relaxed and no flight success is claimed. Further work must use an evidenced native input interface rather than tuning gains around unconsumed commands.
[102533 vanilla-only ordinary startup](evidence/vanilla-game-2026-09-19-102533/report.json) loads exactly the two installed engine/ArmaReforger addons with the regular executable, a fresh separate disposable profile, no successor path/GUID/helper and no noSplash option. It reproduces the initial IntroSplashScreen unknown `SCR_WidgetExportRuleRoot` error, then reaches the vanilla main menu. The failure remains latched despite menu initialization; hidden/muted process cleanup is confirmed. This isolates that initial splash error from successor mission code in the installed environment. It does not reproduce the separate scenario-reload splash, prove normal visible input, or qualify any successor mission/release requirement. The retained successor manifest is explicitly unused provenance bookkeeping.
[102737 ordinary companion startup](evidence/game-2026-09-19-102737/report.json) runs the regular executable with all five successor addons and no Workbench defines or smoke fixture. The expected Chat header/MissionSystems initializes, preparation reaches PRISON, and the real sole local controller receives a living canonical prisoner. Fifteen seconds of stable readiness, zero captured errors, unchanged addon source bytes and owned cleanup are verified. The scoped diagnostic noSplash workaround and standalone round-local identity remain explicit; the separately archived vanilla control reproduces the initial splash error without this mod. No normal visible splash, input, full ordinary escape or multiplayer acceptance is inferred.
[103025 checkpoint companion regression](evidence/native-2026-09-19-103025/report.json) repeats two complete companion-loaded escape/restart rounds, 96 native assertions and zero captured errors on the exact source shared with casualty regression 102037 and ordinary Chat startup 102737. Teardown diagnostics remain disabled, and the opt-in failed flight experiment is inactive. This source qualifies the development clean-runtime checkpoint; all seventeen release acceptance cases remain pending. A subsequent experimental actor-local flight-input branch must invalidate current-source qualification until separately tested.