Dated source reference. Current release status: handoff state. Full original relative paths are in the developer archive.

Ordered native callback effects

November 9, 15:07 UTC checkpoint. RG passes all 149 Windows groups over 817 frozen inputs, ten freshly built native profiles and all 666 API tests. It reruns every runtime group; 37 fixtures rebuild and 88 reuse exact verified compiler outputs. Its identical DLL passes original EG and college EH for 30.098/45.046 seconds after verified clock movement. Both observe the original shared-quad CRT allocation, capture the full 48,672-byte index span with zero unavailable DrawUP inputs, and pass mixed-target/direct-parameter joins. Selected CRT-origin tracking remains healthy with zero faults; no native free of that process-lifetime table is observed. Both restore 99 files and release custody. Their complete 4,074-file/184,807,589-byte and 3,354-file/191,069,406-byte archives and rollback members are hash-verified/fullfsynced externally. See analysis/presentation-framework75/takeover/rg-eg-eh-closeout.json. This accepts bounded copied input and reference evidence. Complete callback/resource/source/free/context closure, semantic owners/cues, full visual/lifecycle acceptance and final author handoff remain open. Packet mask zero; full framework unfinished. No game or custody is active.

November 9, 14:10 UTC working checkpoint. RB passes 143 Windows groups over 802 exact inputs, ten fresh native profiles and all 666 API regressions; 53 fixtures rebuild and 68 reuse exact binaries, with every runtime group rerun. Original EF runs 107.308 live seconds and passes mixed-target and direct-parameter capture. The mandatory DrawUP gate still fails: its 48,672-byte index span has no unique tracked heap/image token. A bounded mapping query grants no lifetime authority. All 99 files restore and custody releases at 13:57:38 UTC; its complete 3,352-file/220,890,720-byte archive and rollback are verified/fullfsynced, and temporary payloads retire254. No unchanged retry or college run follows. Static tracing identifies the shared quad CPU table, its original MSVCR100 realloc call and separate GPU-buffer creation. Isolated255 passes 2,916 original-x86 factory cases and local sanitizer checks; CRT generation/withdrawal validation is in progress and remains unpromoted. QS/DY/DZ remains the last paired scoped native baseline. Full ownership, effects/resource/source/free/context closure, cues, visual/lifecycle acceptance and final author handoff remain unfinished; packet mask zero. No game or GPU custody is active.

November 9, 13:41 UTC working checkpoint. QZ passes all 143 Windows groups over 802 exact inputs and 666 API regressions; ten native profiles and 96 fixtures build fresh, with 25 exact fixture reuses and every runtime group rerun. Original ED runs 101.877 live seconds and passes the mixed general-target and direct-parameter match gates. One match cohort accounts for all 28 targets, including 17 independent surfaces, zero unavailable targets and 11 explicit nulls; the surface bank retains/releases 54 references and closes empty, healthy and fault-free. ED fails the mandatory DrawUP gate because its indexed request remains unavailable; no unchanged retry or college run is started. All 99 files restore and custody releases at 13:29:55 UTC. Its 3,119-file/217,659,077-byte archive and rollback are verified/fullfsynced; temporary payloads retire 249. An isolated timing correction passes 9,647 input, 12,972 batch and 60,446 production-body Windows checks: exact validation remains the default, explicitly declared consumer indices validate after strict original-token recopy. Bounded preparation diagnostics distinguish shape/code/storage/index refusals; the native cause remains unconfirmed. This eight-file correction is promoted and RA startup checks are running before a fresh RB full gate. The separate buffer-factory prototype passes 366,355 x86 checks/82,944 scalar cases without resource-ownership acceptance. QS/DY/DZ remains the last paired scoped native baseline. Full effects, source/free/context closure, semantic owners/cues, visual/lifecycle acceptance and author handoff remain unfinished; packet mask zero. No game or GPU custody is active.

November 9, 13:14 UTC working checkpoint. QY passes 27 startup groups over 215 inputs. QZ is running 143 Windows groups over 802 frozen inputs with the general target profile enabled; no QZ game run has started. The promoted typed 2D/cube/direct-surface path passes 43,335 controlled production-body checks through publication, capture, surface retention, copied predictions, cancellation and retirement. Forty-eight receipt checks require exact metadata/physical/effect copy-retire joins and a mixed match-bound cohort with zero unavailable targets. The older default-only physical gate is superseded only when that stronger general gate and the same healthy closed surface bank pass. Eight oversized target metadata cases refuse before reads in the 28,121-check sanitizer run. QX remains the latest passed full CPU gate (137 groups, 666 API regressions); QS/DY/DZ remains the paired scoped native baseline. Whole effects/resource/source/free/context closure, semantic owners/cues, full visual/lifecycle acceptance and author handoff remain open. Packet mask zero; full framework unfinished. No game or custody is active.

November 9, 12:57 UTC working checkpoint. QX passes all 137 Windows groups over 783 frozen inputs, ten fresh native profiles and 666 API regressions. The QW fixed-window fixture failure is corrected by reserving its complete two-phase private windows before portable heap work; an intentional occupied-window probe exits before any original instruction. Forty-four receipt tests pass. Isolated texture-target inputs pass 15,111 Windows/sanitizer checks; typed 2D/cube/direct surface acquisition passes 19,315 controlled-COM checks; the general target batch is still under verification. None of this isolated target work is game-installed. QU EA/EC native refusals remain preserved; QS/DY/DZ remains the last paired scoped baseline. Complete targets, effects/resource/source/free/context closure, semantic owners/cues, full visual/lifecycle acceptance and author handoff remain open. Packet mask zero; full framework unfinished. No game or custody is active.

November 9, 12:33 UTC working checkpoint. QW is building 137 Windows groups over 782 frozen inputs after QV passes all 27 startup groups and 44 receipt-parser tests pass. The promoted DrawUP path has 9,216 original VM calls, 9,170 input checks, 11,384 batch checks, 34,457 mixed-plan checks, 7,076 candidate-window checks and 57,800 production-body checks. It copies original heap/image spans, validates vertex/index extents and rechecks the complete cohort. Direct parameter preparation now uses the existing candidate contract; consumption retains strict original-token and byte checks. QU passed 133 CPU groups, but original EA and unchanged retry EC failed required surface coverage after 102.777/110.767 live seconds; EA alone passed the new parameter match gate. Both restore 99 files and release, with full archives and raw retirement 235/236. No third unchanged retry or college run was started. QS/DY/DZ remains the last paired scoped baseline. General texture-backed targets, complete effects/resource/source/free/context closure, semantic owners/cues, full visual/lifecycle acceptance and author handoff remain open. Packet mask zero; full framework unfinished. No game or custody is active.

November 9, 11:50 UTC checkpoint. QS passes 131 Windows groups, 762 frozen inputs and 666 API tests. All ten native profiles and 33 fixtures compile fresh; 76 fixtures reuse verified exact source graphs, commands and artifact hashes, while every runtime group reruns. The same DLL passes original DY and college DZ for 30.142 and 30.127 live seconds. Device requests and consumer wrapper serials have positive match-bound copy/retire pairs with zero unavailable request inputs. Serial observations preserve the original allocation/source tokens and whole-wrapper rechecks; the two producer write prefixes are qualified separately and do not establish full producer lifetime. Both runs restore 99 files and release custody. Full 3,135-file/183,807,005-byte and 4,971-file/187,365,017-byte archives and rollback verify with fullfsync; temporary payloads retire 231/232. DS, DU and college DX failures remain preserved. Effect conversion stops at direct parameter opcode 0x0a; isolated original-VM, mixed-plan and consumer-batch work is not in QS. Full effects, resource/source/free/context closure, semantic owners/cues, visual/lifecycle coverage and author handoff remain open. Packet mask zero; full framework unfinished. No game or custody is active. See qs-dy-dz-closeout.json.

November 9, 09:54 UTC checkpoint. QK passes126 Windows groups/747 frozen inputs/666 API tests (source 8d8e25b0046ede242062451f69e5ec84d258b1304e0b2973e20d8285d0d4af2f, DLL dcd9fafd09d5351c1f956f6a0d536a1e9d3747c8dd211f5850a64ac470d4a95b). All ten native profiles and27 fixtures compile fresh;77 fixture binaries reuse only exact local include graphs, commands and artifact hashes, and every runtime group reruns. Identical original DQ/college DR pass54.244/70.983 live seconds: each has three bound/copied/retired environment sets; eleven/ten clear requests have zero unavailable inputs; four/six independent surface references retain/release with healthy empty closed banks. Both restore99 files and release. Complete2,985-file/197,531,719-byte and3,785-file/204,330,795-byte archives/rollback verify and fullfsync; temporary copies retire211/212. QG's Windows macro failure, DM's tag-byte refusal and DO's sole changed match-surface sample remain preserved. The correction proves all256 tag values/both bit-seven variants in3,264 original Clear cases; only observation bounds grow175→235/190→250 seconds inside the unchanged six-minute lease, with30 required live seconds and all gates retained. Conversion now stops at opcode0x23. Isolated twelve-method request/input/batch work passes3,456 original calls and1,213,376 Windows/sanitizer batch checks; it is not in QK. Complete driver effects, source/free/context closure, semantic owners/cues, full visual/lifecycle acceptance and author handoff remain open. Packet mask zero; full framework unfinished. No game or custody is active.

Current evidence - November 9, 07:39 UTC

PP2 remains the paired game baseline. Original DI and college DJ pass the default color/depth surface-reference gates: six/seven captured sets, twelve/fourteen balanced independent references, empty healthy closed banks, 99 restored files each and custody release. Both complete archives and rollback files are externally verified and fullfsynced. These are observed default-target ownership results; whole callback effects and drawing authority remain open.

The fresh PW candidate freezes 728 inputs and is building 121 Windows groups. Its new game qualification is pending. The dated sections below preserve earlier verification scopes.

packet-dependency-effects.hpp can now convert opcode 0x2f when every target occurrence has supported, matching default-target metadata. Missing entries, unsupported sources, reordered occurrences, changed flags, wrong identities and foreign getter outputs refuse the complete conversion. This extends the copied predictor; all original callbacks continue to execute.

packet-initial-effect-plan.hpp joins copied image-context, shadow/cache, configuration and command inputs. An uncopied prior constant source remains opaque only while unused: a remap before the first setup refuses prediction. The helper supplies no native reads, driver calls, semantic owner, physical lifetime certification or callback permission. Nondefault configuration storage, complete effect/source/free/context closure, remaining semantic owners/cues and visual/lifecycle acceptance still need qualification. Accepted packet mask is zero.

The private copied-state predictor now carries one shadow, initialized cache, shader cache, constant source and declaration bank through mixed callback occurrences. native-callback-sequence-plan.hpp produces owned driver arguments and the predicted state after each occurrence. It never executes a callback, reads a native pointer, retains a COM object or grants packet authority.

The separate declaration ownership path passes its physical-reference closure and queued snapshot handoff in MU original CO and college CP. MY includes the newer compound state-input diagnostic and passes 107 combined Windows groups. Later NH freshly passes the same 107-group, ten-profile gate over 678 inputs. Original CU copies/retires two pre-match compound sets but fails the match-bound gate after 49.849 seconds of advancing clock. The one-window change removes the earlier constant control-copy failures; 13 sampled first-constant byte mismatches remain alongside 20 earlier scope withdrawals. Declaration/forwarding/closure gates pass, all 99 files restore and custody releases. Its complete 4,443-file archive and rollback verify/fullfsync externally. NI and MW8 add exact span, offset and changed-word diagnostics while retaining byte-for-byte refusal. New source is not the frozen NH candidate. MZ's effect-input converter remains separately CPU-tested and has no native integration.

The mixed sequence includes original cache/VM/material leaves, shader bindings, constant setup/remapping, declaration create/bind and the equal-context-state branch of DDB8A0. DDB8A0's other branch calls E0D3D0 and remains refused. Context creation through the DA3620 fallback is unqualified. The constant setup fixture relocates the captured device-selector body; both rewritten calls authenticate against their original bytes before comparison.

The root model retains the native aliases and less obvious effects. A zero-count cached vertex write still invalidates the vertex shader cache. Direct vertex writes leave that cache intact. Render-cache word 103 also stores mip bias. Declaration creation uses the actual output word even on failed HRESULTs, and the cached identity can suppress a later bind after an earlier failed create. The model preserves these behaviors rather than predicting driver success.

Input bindings keep unused pointer words authenticated without following them: shadow reset ignores ECX, a null viewport reads no payload, and zero-count constant/state writers read no data array. Nonempty arrays require exact typed bounds and current storage generations. Consumer capture recopies every byte. The batch re-enumerates the complete explicit path, including repeated calls, inside one bounded local metadata window. That window permits no native calls, COM operations or waits. A later diagnostic copy never clears execution refusal.

All predictors and bindings have callback_authority=false. Complete native source/free/context closure, omitted callback effects, queued native declaration snapshot acceptance, direct parameter and ordinary VM state effects, whole-packet composition and actual visual/fallback acceptance remain separate gates. The full framework is unfinished and its accepted packet mask remains zero.

Frozen evidence is under analysis/presentation-framework75/packet75ll, lm, ln4, lo and lq, with matching source ZIPs/manifests in their parent. LN/LN2's warnings-as-errors failures and LN3's inconsistent aliased cache fixture are preserved. LN4 corrects the fixture's cache/shadow word together; it does not weaken the consistency check. LP's combined fixture refused an occupied image window before any hook was installed. The corrected fixed-base fixture setup passes in LS (95 groups) and LZ (100 groups). LZ includes the root/declaration inputs and owner/observer tests with all nine DLL profiles and 666 API tests. MA subsequently passes that receipt-inclusive set. MG's later 101-group gate passes paired original/college forwarding. MR's 105-group gate preserves a failed stale-header fixture expectation; MS/MT correct the test while retaining full-header change detection, and MU subsequently passes all 105 groups, ten DLL profiles and its paired native gates. MY's newer 107-group gate preserves CQ's failed compound match witness; NC adds diagnostics without changing capture rules. None of these copied-input tests grants a packet capability.

Joint dependency capture candidate

packet-dependency-batch.hpp prepares one identity/context-bound set containing materials, declarations and root/constant inputs. Its consumer copies all three families inside one Scope metadata window. Every child retains its existing byte, generation, context and physical-owner checks. A failed child clears all partial output and permanently invalidates that execution; the complete set cannot be rearmed from a later copy. No native or driver call runs in the window.

NL2 passes 14,349 Windows/local sanitizer checks. NN2 passes 15,206 Windows checks through the actual production queue, append check, consumer, retirement, cancellation, context refusal and active-close guard. NO passes all 27 startup groups. Five receipt tests round-trip the actual Windows formatter and refuse incomplete, changed, duplicate or foreign joins. NP freshly builds 109 groups and ten profiles over 683 inputs; no earlier runtime pass is inherited.

This is a joint copied-input contract. Initial renderer state, opaque callback classification, source/free/context closure, complete ordered effects and native visual acceptance remain required. In particular, material and declaration snapshots do not make unclassified commands executable. The accepted packet mask remains zero and the full framework remains unfinished.

Consumer-time payload candidate

Original DAFC60/DADE10 consume pointed-to values when invoked, rather than a historical copy made at publication. NS executes the authenticated original instructions in six controlled cases over three device branches (89 checks), keeping addresses, descriptors and arguments unchanged while payloads change. Its observed calls match the consumer values and differ from the publication prediction. This is CPU instruction evidence, not native-game acceptance.

packet-consumer-inputs.hpp supplies a separate, opt-in snapshot contract. It preserves original allocation/image generations, command and descriptor bytes, source pointers, object pins, packet identity and context. It samples only declared parameter/root/constant payload spans, then strictly captures the entire cohort against that sample in the same metadata window. Inline material payloads are identified by decoding the original program; only those value regions are projected into the consumer basis. Instructions, header fields and indirect pointer words remain exact. Changed texture-pointer dependencies, partial program overlaps, wrong span tokens, changed shapes and active writers refuse. Every partial output clears. The path cannot retry a failed strict capture, rearm its execution or refresh an epoch or allocation identity.

NT passes 20,853 Windows/local sanitizer checks. NU passes 15,464 Windows production-body checks including changed inline material, viewport and constant values, owned references, cancellation and close. NV passes 27 startup groups. Five consumer-origin/closure receipt tests and five legacy receipt tests pass. NR separately passes 20,325 checks joining material, declaration and root/constant inputs into native callback order; unresolved creation and opaque commands refuse the whole conversion. It makes no native reads or calls.

Consumer-mode receipts use a separate version-2 prefix with origin=1. Qualification requires this origin, all nonempty dependency families in a match-bound set, exact copy/retire identity, no unresolved creation, zero faults and live frames, and acknowledged physical-reference closure. The historical publication-exact results remain independently reported, and are mandatory for the unchanged default mode. They do not certify a consumer-time payload. Other native forwarding, code, context, resource, process closure and restoration guards remain required. NX passes all 113 groups, ten DLL profiles and 666 API tests. The identical DLL passes bounded original DA and college DB consumer-origin cohort/physical-closure gates. Both restore all 99 files and release custody; complete external archives and rollback copies verify. Their publication-exact state-input observations remain false and are reported separately. These runs accept copied inputs and physical closure, not effects or drawing.

A successful input conversion is not an ordered effect or drawing permission. Initial renderer state, alias/write footprints, unclassified callback effects, complete source/free/context coverage and native visual/fallback acceptance remain open. The packet acceptance mask remains zero.

Guarded invalidation inputs and initial context facts

NY passes 21,085 Windows x86 checks for the copied-input conversion of the literal DDB8A0 callback to Invalidate. The persistent planner still refuses unequal context-state words, whose native branch may call E0D3D0. Converting a command does not qualify that branch in a native frame.

packet-context-facts.hpp samples the cached declaration, vertex/pixel shader words and the two branch words from an exact observed context allocation. It authenticates the complete original DDB8A0 and DA3620 bodies, copies each field twice in a bounded metadata window and checks the original context pointer, packet identity and allocation generation. A readable mapping or enclosing allocation cannot substitute for the context's own observed allocation. Missing storage stays explicitly unbound without reading context fields. Identity, code, read and consistency failures invalidate the execution and clear output.

NZ passes the 330 context-fact checks and 20,854 consumer-core checks on Windows and under local sanitizers. The production integration extends the same cohort window and keeps every candidate local until that whole window succeeds. Its new source is newer than the paired NX DLL; production and native qualification of the context facts are still pending. These five fields are an initial subset, with no complete renderer state, callback authority or drawing permission.

OA subsequently passes 15,962 production checks and OC passes all 113 Windows groups, ten DLL profiles and 666 API tests over 696 inputs. OC remains a CPU checkpoint; its heap-only context facts were not installed for a new game run. Six archived DA/DB structural samples identify the native getter's literal fallback object at 0x20e42f0, rather than an observed heap allocation.

packet-image-context-facts.hpp therefore supplies a separate typed image path. It binds the exact pointer slot and complete required object span as main-image metadata during publication and carries both tokens with that cohort. It never fabricates an allocation or constructor identity. Consumer capture authenticates the original getter/reset code, checks both storage tokens and pointer, and copies the five fields twice inside the joint window. Missing storage stays unbound; stale storage, wrong shapes, overlaps, changed pointers/code/fields and failed reads refuse without retaining a partial cohort. Image and heap facts use separate receipts and exact copy/retire identity checks.

OD passes 564 Windows/local sanitizer image checks; OE passes the expanded 20,855-check consumer core, and OF passes 16,703 production checks including typed image capture and whole-cohort failure cleanup. Four image receipt tests, six consumer/context receipt tests and five legacy receipt tests pass. Optional publication-pin refusal diagnostics are newer working code and will be rebuilt in the next full gate. Native image-context capture is still pending. None of these observations supplies complete renderer state or callback authority.

OJ subsequently passes all 113 groups, ten DLL profiles and 666 API tests over 698 inputs. Original DC passes the consumer-origin cohort/physical-closure gate after 30.154 live seconds and restores all 99 files before releasing custody. Its two pre-match image-context sets bind and report equal state words18/18; both match sets are unbound. The separate match-context gate is therefore false. The complete 2,969-file archive and rollback verify/fullfsync, and duplicates retire182. College DD was not started. Its 22 image-pin refusals occur at the publication metadata-window gate; they do not constitute positive context proof.

The newer metadata_candidate selects only an existing, still-current image storage identity. It copies no context values, creates no token and supplies no quiescence. Selection precedes a new publication observation; consumer capture continues to require the original code, pointer, token and Scope checks and the double field copy. Failed consumer executions remain invalid. OK passes 686 Windows/local sanitizer checks for selection during an active publisher, stale or wrong-sized storage and continued refusal after a failed consumer copy. OL/OM and ON startup pass; full OO is freshly rebuilding 113 groups. Its native match-context evidence remains pending.

Target-binding prediction component

native-target-binding-plan.hpp predicts the target cache, temporary surface references, target/depth driver calls and dependent render-state updates of the captured target helper. It consumes separately supplied query/getter results; it does not obtain resources or grant ownership. OP executes the original target, surface and release instructions and their dispatch tables, together with the ready/context resource query and original render-state helpers, against controlled driver objects. All 5,775 cases and 65,533 checks pass on Windows x86. They compare the entire shadow/context, command inputs, call order and balanced references across three device branches, null targets, tags, mip boundaries and zero, positive and failed HRESULTs. Fourteen portable golden traces also pass79 local sanitizer checks. Only an exact zero surface-getter result accepts its output; target index zero can skip a null bind while still updating its native cache.

This component is separate from frozen OO and has no native integration. Actual source/target binding, physical frame ownership, initial renderer state and mixed sequence qualification remain required. Unsupported opcode0x2f continues to refuse the whole copied effect conversion. The bounded harness capture now records candidate target words/wrappers read-only, marked structural and unqualified; a linear candidate scan is not an execution trace or lifetime proof.

Default target inputs and independent surface acquisition - November 9

The OO original DE/college DF pair proves the initial image-context subset and consumer-origin dependency closure described in the current framework record. Publication-exact state copies remain false; whole effect conversion still refuses opcode0x2f. OP2 separately compares5,775 original target/surface/release cases with65,533 checks.

packet-default-target-input.hpp binds literal0x2f command bytes and image wrapper fields to the image-context tokens. OY passes858 Windows checks. Its candidate mode permits producer overlap without claiming quiescence; the consumer must still strictly copy the original spans, fields and command. packet-default-target-batch.hpp joins every target occurrence to the complete copied trace. OZ passes578 Windows checks, including unsupported-source accounting and sticky foreign-thread refusal. This is explicitly a partial default-target subset, never complete target-effect coverage.

The working native consumer probe carries that subset in the same dependency cohort and metadata window, then logs and clears it at retirement. PB passes17,630 production checks. Exact pointer or flag redirection refuses the whole sample. Native original/college proof of this new path is pending; OO remains the paired baseline.

native-default-surface-acquire.hpp uses owned typed device-getter outputs and only compares the borrowed context word. It never calls COM through the borrowed word. OR4 passes690 controlled-driver checks, including wrong device/role/generation, positive/failed getter results with bogus outputs, reentry, late-change deferred release and close/reopen. OS2 passes85 actual D3D9 HAL checks: six acquisitions, twelve balanced independent holds/releases and one Reset/resize. Metadata is controlled, native game target ownership is not qualified, and this component is not integrated into the game DLL.