Cesarus85--mjolnir-vr/docs/object_matching_logs/hs_obj_codex_checkpoint.md

HS large-object reconstruction checkpoint

Scope and disposition

This is an honest partial checkpoint for source/hs/hs.obj and the two independently recovered functions in source/hs/hs_runtime.obj. Both units remain NonMatching; this change grants no object or data credit and makes no configuration, parked-function, symbol, or frozen-object change.

The source basis is local Claude branch commit fae4cc9a82b3ea906bddb30673a8d1b19f072f2d, itself built from the existing campaign history plus the later ordinary-C cleanup through 8a9509f1. Only source/hs/hs.c and source/hs/hs_runtime.c were transplanted into a fresh worktree based on canonical a0bad121; branch history and unrelated files were not imported. The source contains no assembly, volatile, __forceinline, optimization pragma, compiler barrier, byte forcing, or omitted return.

Strict result

Fresh XDK 3911 /O1 /Gy reconstruction gives:

• hs.obj: 442/448 functions strict-exact, 28,656 padded exact code bytes and 1,417 exact relocations. The canonical base was 435/448, 27,504 bytes and 1,359 relocations, so this checkpoint adds seven strict functions, 1,152 padded bytes, and 58 relocations without losing an exact function.
• hs_runtime.obj: 24/66 functions strict-exact, 1,776 padded exact code bytes and 117 exact relocations. The canonical base was 22/66, 1,504 bytes and 106 relocations, so this checkpoint adds two strict functions, 272 padded bytes, and 11 relocations without losing an exact function.
• Combined gain: nine strict functions, 1,424 padded code bytes, and 69 exact relocations.

The seven newly exact hs.obj functions are:

• _code_000b2f00
• _code_000b3450
• _code_000b3e80
• _hs_find_script_by_name
• _hs_find_tag_reference_by_index
• _hs_scenario_merge
• _hs_scenario_postprocess

The two newly exact hs_runtime.obj functions are _code_000ba310 and _hs_can_cast.

Non-code ownership is not complete and is not claimed. The hardened snapshot finds only 30/911 target non-code sections strict in hs.obj and 25/61 strict in hs_runtime.obj; both units therefore remain closed to object admission regardless of their code counts.

Remaining hs.obj code

All six target functions are emitted. Five have exact padded size and relocation count and differ only in normalized instruction scheduling:

• _code_000b2110: 80 bytes
• _code_000b2160: 80 bytes
• _code_000b2a40: 96 bytes
• _code_000b33d0: 128 bytes
• _code_000b3f50: 96 bytes

_hs_compile_and_evaluate is the only structural residual: target 592 bytes, candidate 608 bytes, with all 40 relocations present. January reserves EBX as the shared zero carrier and homes the selected source pointer through the dead incoming argument slot; the current compiler keeps that pointer in EBX.

The final bounded review rejected the following do-not-repeat shapes:

• HCEA's is_blank/check_blank scan grows the full function to 624 bytes and 41 relocations by emitting a second static isspace call site.
• Direct parameter reuse, a function-scope initialized source alias, and a function-scope uninitialized source assigned at the existing point all stay 608 bytes and do not change the EBX source lifetime.
• Direct grouped-BSS postfix increment for _code_000b33d0 emits a memory increment and regresses the tail. A local-short postfix and a typed short struct view normalize byte-for-byte to the existing mov eax/inc eax residual. The flat-static family can produce mov ax/inc ax, but 720 layout permutations failed to preserve January's BSS offsets.
• The three FPU wrappers already exhausted argument binding, declaration, ordering, scope, cast, pointer, struct, and union variants. The retained ordinary double locals reproduce January's x87 push path without the forbidden volatile stand-in.
• _code_000b3f50 requires its explicit readable (real) conversion; the implicit form grows to 112 bytes and adds two relocations.

hs_runtime.obj remainder

Seven emitted functions remain nonexact (_code_000b9a10, _code_000ba090, _code_000ba290, _code_000ba2c0, _code_000ba390, _hs_evaluate_wake, and _hs_wake_by_name) and 35 target functions remain unwritten. This checkpoint does not imply that hs_runtime.obj is near admission.

Validation

• Forced clean XDK rebuild of both objects: passed; the strict counts above reproduced from the clean artifacts.
• Full halobetacache_build and libcmt_build: passed across 567 build edges.
• semantic_progress/progress: 470 units, 3,611 functions evaluated, 3,529 accepted exact functions, and zero unit errors. The partial checkpoint adds code-function evidence but intentionally leaves both objects incomplete.
• Admission audit: zero candidates and zero revocations; only the pre-existing unrelated shell_xbox contradiction remains.
• Parked-function validation: zero stale and zero invalid entries.
• Full tooling suite: 179/179 tests passed.
• Clean-commit regression snapshot, forced deletion/rebuild of both objects, and check: ok: true, no changed nonexact functions, no newly exact warnings, no ownership warnings, and no failures. The manifest freezes all accepted functions plus complete non-code and symbol ownership for both units.
• House/Berth scan: passed; raw tag_get occurs only inside the two typed local retrieval macros.
• Diff and scope audit: exactly the two HS source files plus this new Codex-owned ledger; no configuration, parked manifest, Claude-authored Markdown, or frozen-object path changed.