0.5.0 — 2026-08-08

Recipe format 4.3 — schema restructure, creator simplification, and metadata expansion. Breaking changes from 4.2 are limited to the recipe format itself; the patcher, validator, and annotator pipelines are unaffected.

Changed — Recipe Format 4.3

• openremap envelope dropped. type and schema_version are now top-level fields instead of being nested inside an openremap object.
• source and application added. source distinguishes "full_cook" (binary diff) from "tune_export" (exported from a saved tune). application identifies the tool that produced the file ("openremap-core" or "openremap-studio").
• creator flattened. The nested author sub-object is removed; name, handle, and id are now top-level fields in creator. The tool and tool_version fields are removed — tool provenance is carried by application and schema_version.
• metadata restructured. Dropped: context_size, max_context_size, format_version. Added: name, tags, instruction_count, tune_id.
• statistics.context_size renamed to min_context_size.
• ecu block extended. Added: oem_part_number, platform, calibration_version, serial_number, dataset_number.
• flags.confidence normalized to float (0.0–1.0). Was string "HIGH"/"MEDIUM"/"LOW" in 4.2. InstructionFlag.confidence type changed from str to float (0.9 = high, 0.5 = medium, 0.3 = low).
• Extensibility rule. Parsers MUST ignore unknown fields, allowing forward-compatible additions in minor versions.
• Schema version gate. All recipe consumers (patcher, validators) now validate schema_version at construction time. Recipes older than 4.3 (pre-0.5.0) are rejected with a descriptive error. Future minor versions (4.4+) pass through — the extensibility rule guarantees forward compatibility.

Added

• Endian auto-detection (identifier.py). _detect_endian() uses the high-byte-zero heuristic (ported 1:1 from the Rust editor's analysis/endian.rs) to detect the ECU's byte order from calibration data. identify_ecu() now returns ecu_endian and ecu_cell_bytes fields for all results, including unknown/unrecognised binaries.

• Low-entropy context scanner (annotator.py). LowEntropyScanner detects instructions whose context anchor is weak — non-unique patterns (WEAK_ANCHOR, confidence 0.9) or low-entropy regions (LOW_ENTROPY_CTX, confidence 0.3). Registered by default in RecipeAnnotator alongside the existing VINScanner.

• .orst 2.0 saved tune format. New build_orst() method on ECUDiffAnalyzer produces a minimal .orst workspace file carrying only instructions, source binary identity, and history metadata — just enough for the editor to reopen, display history, and export as a portable recipe.

  • New Pydantic schema: openremap/core/schemas/orst.py (OrstRootSchema, OrstSourceBinarySchema)
  • New format spec: docs/orst-format.md
  • server.py cook endpoint returns .orst payload when tune_id is provided

• Force Save via require_unique parameter. When require_unique=False, non-unique context anchors become cook_warnings instead of a hard error, allowing the recipe to be saved for same-binary use. Exposed through the server.py cook endpoint and ECUDiffAnalyzer constructor.

Improved

• context_after anchor in patcher. The patcher now searches for ctx + ob + context_after (not just ctx + ob), doubling the effective anchor length at zero cost. Failure diagnostics now report the full anchor description, check whether ob bytes exist elsewhere in the binary, and give a clear hint about SW revision mismatch.

Changed — Internal

• Pydantic schema split. openremap/core/schemas/analyzer.py (161 lines, format 4.0–4.2) deleted. Replaced by:
  • remap.py (168 lines) — recipe 4.3 schemas with defaults for forward compatibility (all fields optional/defaulted, no hard-required fields)
  • orst.py (89 lines) — .orst 2.0 schemas

Removed

• Trust level display. trust_level is no longer displayed in CLI or TUI output until cryptographic signing is implemented and enforced by the pipeline.