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

source/scenario/wind.obj — Jonas partial recovery

Disposition

This branch recovers an admissible 7 / 9 strict-exact partial from the authoritative c335fd8e20f253f59d4d38852c671122645d25f5 baseline. The object remains NonMatching; no object-level completion claim is made. The older Claude ledger, wind_obj.md, was read as evidence and was not edited.

The recovered strict functions account for 1,324 meaningful code bytes and 1,376 padded COFF bytes:

Function	Meaningful / padded bytes	Relocs	Normalized SHA-256
_wind_dispose_from_old_map	8 / 16	1	6d756ab123ca36d54d960697fa16bc6dc86933084aa5b92457f2b58c5174722d
_wind_update	593 / 608	41	27144f49abed5dd9195b688b5628ebd5f8981c1db61e78405d8b0741e7f25266
_scenario_get_current_from_weather_palette	311 / 320	12	ba6670dbb6da6d988ddd3d9e2d3c33a8c36155e43e4bfb18cad292c72dc173db
_wind_initialize_for_new_map	75 / 80	10	190722dd697f50f998d0ed5ca1e9eb06ff28a673fc9ea9bc57876ae4950bde75
_scenario_get_current	273 / 288	7	7a0b601cbd8bdefa3c729433b85d924e871d8e169773d9a49e2add51334bd175
_scenario_get_wind	32 / 32	1	091ef8221b5166d2b830d887770e8afb22f8fc91fb8e9349f763ac38a5095df8
_scenario_get_water_current	32 / 32	1	5ed4582f15c8c59e1cc27a724e42b90c08f0023f96be453f0f0fba00025bfbfc

Ordinary reporting independently measures 7 / 9 functions and all 100 target data bytes exact: .data 4 / 4 and .rdata 96 / 96. The semantic global_environment_index = NONE definition restores the target's owned four-byte data section.

Zero-credit necessary callee

_code_0017f750 is deliberately zero credit. Its definition is needed as the natural private callee: removing it changes VC7's private calling topology and loses the exact _scenario_get_current_from_weather_palette caller. The current defined-C body is target/candidate 224 / 208 padded bytes with 8 / 8 relocations, so it is not claimed exact.

The old 224-byte exact experiment read a different inactive union member and is not retained. In this branch byte is the repository's typedef for unsigned char, and byte const *sample_key_bytes = (byte const *)&sample_key therefore reads the float's object representation through an unsigned-character type, as C permits. It is character-representation access, not an inactive union read or another type-pun. There is no union in the recovered source.

A standard fixed-size memcpy representation probe was also tested. VC7 lowered it to calls and produced 368 padded bytes / 23 relocations, so that probe was rejected. The selected unsigned-character representation is the smallest fully defined form found and preserves the exact caller.

Omitted residual initializer

_code_0017fbd0 is intentionally absent. The current COFF symbol is an undefined external function (section == 0); no residual body is emitted. _wind_initialize_for_new_map retains its natural relocation to that unresolved callee and is strict exact. The old 384-byte / 9-relocation nonexact spline initializer was not retuned, copied, or admitted, and its body-only constants were not manually materialized.

Typed reconstruction and source oracles

The behavior and layouts were reconstructed with typed ordinary C from the old golden candidate and the HCEA C-series oracle in research-cache/halocea-cseries-20260820/src/blam/scenario, especially the wind_*.c, scenario_get_current*.c, and headers/wind_state.h material. The private typed layouts in wind.c assert:

• wind_state: 0x20 bytes;
• wind_globals: 0xD0C bytes;
• structure_weather_palette_entry: 0xF0 bytes, wind tag reference at 0x80 (its datum index is consequently at 0x8C) and wind direction at 0x90;
• wind_definition: 0x40 bytes.

The weather-entry and wind-global layouts are private to wind.c. An early draft put the weather entry in the shared structure_bsp_definitions.h; the clean A/B gate caught a four-normalized-byte independent-load reorder in the nonexact _build_structure_lens_flares. The declaration was localized and structure_bsp_definitions.h was restored byte-for-byte to c335fd8e before the final build.

No assembly, volatile shaping, force-inline directive, new pragma, compiler barrier, undefined behavior, or manual byte forcing is used.

Regression and validation

The shared-header and frozen-object A/B used clean builds from the authoritative baseline and this worktree. The project regression gate over 16 direct consumers plus the frozen objects reported only .debug$S differences caused by the two distinct absolute worktree paths. It reported no final runtime-code, runtime-data, relocation, or symbol-ownership change.

A broader hardened COFF audit compared every available non-wind base object:

• 568 objects;
• 4,306 strict function sections;
• 5,611 non-debug sections and 1,902,854 logical bytes;
• all meaningful non-debug symbol ownership.

All passed. In particular, vehicles.obj, ai_debug.obj, ai_script.obj, actions.obj, and units.obj retained identical strict code, non-debug section bytes, relocations, runtime data, and meaningful ownership. No source for any of those five objects was edited.

Final gates:

• clean halobetacache_build: pass;
• ninja progress: pass, 0 unit errors;
• strict comparator: the seven functions above exact, helper nonexact and zero credit, initializer absent;
• baseline fail-closed wind check: six newly exact review leads, helper still nonexact, and only intentional additive object/data/debug/symbol changes; the previously exact disposal function independently remains strict-identical;
• post-change clean snapshot/check: stable;
• python -m unittest discover -s tools -p 'test_*.py' -v: 179 / 179 pass;
• parked-function validation: 3 active, 0 stale, 0 invalid;
• object-admission audit: 0 candidates and 0 revocations (the pre-existing shell_xbox contradiction remains unrelated).

No configuration status was changed, no Claude Markdown was edited or deleted, and nothing was pushed.