docs(overhaul): bound nonempty geometry witness integration
Prepare the original unpinned witness with an explicit cathedral-shell non-vacuity gate. Write-only while the separate production bridge owns test execution; preserve the scratch and its failed diagnostics. Co-Authored-By: OpenAI Codex <noreply@openai.com>
This commit is contained in:
parent
4c3aafcd07
commit
4e4e8e8253
1 changed files with 53 additions and 0 deletions
|
|
@ -0,0 +1,53 @@
|
|||
# C1a original geometry witness — guarded integration, write-only
|
||||
|
||||
Lead bounded test-only contract,2026-09-05. The separate AP159 production
|
||||
bridge is in flight under its own contract; it owns all build/test execution.
|
||||
This work is independent file preparation only, not a fourth authored-input
|
||||
diagnostic fix. Do not change or rerun the preserved scratch diagnostics.
|
||||
|
||||
Read `c1a-geometry-activation.md` and packet§37.3–37.4/§45. The original
|
||||
scratch witness remains unpinned because the production bridge was wrong.
|
||||
Its source SHA9894EC50BF40F732237E3249E5F7815FDC931DDC66F1D5D212D0CA717373A9C5
|
||||
has been independently reviewed and read by the lead. One actual pre-pin
|
||||
gap was found: shellF4180104 presence/nonzeroGeometryId is asserted, but its
|
||||
extracted mesh's nonzero vertex/subset/index content is not. An empty shell
|
||||
could become part of the initial golden. No other source blocker was found;
|
||||
mutation and full execution evidence remain owed.
|
||||
|
||||
## Allowed change
|
||||
|
||||
Work only in the canonical campaign worktree. Add exactly
|
||||
`tests/AcDream.App.Tests/Streaming/LandblockGeometryMembershipInstalledDatTests.cs`
|
||||
by copying the preserved original file from s5-c1a-geometry with apply_patch.
|
||||
Do not copy its AuthoredDiagnostic partial, any artifacts, package or old
|
||||
golden; do not alter the scratch. Retain the original partial declaration,
|
||||
all existing methods/serialization, original three-cell/21-entry semantics,
|
||||
and `REPLACE_AFTER_FIRST_VERIFIED_CAPTURE` placeholder.
|
||||
|
||||
The only source addition is at the existing EnvCell mesh extraction before
|
||||
serialization: for the exact ShellCellIdF4180104, assert nonempty actual
|
||||
mesh.Vertices, nonempty ObjectMeshManager.OrderedUploadBatches(mesh), and
|
||||
positive total emitted index count from those real batches. A short comment
|
||||
may explain why the non-vacuity gate precedes first golden capture. Aggregate
|
||||
landblock totals, hard-coded mock geometry or a second extractor are forbidden.
|
||||
|
||||
No production, existing tests, shaders, scripts, fixtures, register or docs
|
||||
edits by the implementer. No new agent, commit, build/test or client actions.
|
||||
Return code-ready with source hash and exact diff; lead releases execution
|
||||
only after the bridge implementer returns ownership and source is reviewed.
|
||||
|
||||
## Later execution boundary (not authorized by this write-only contract)
|
||||
|
||||
After the bridge passes focused review/tests, lead will activate the original
|
||||
semantic/deterministic capture on the integrated production code. Placeholder
|
||||
failure is expected until two fresh owner captures agree, all semantic facts
|
||||
including the now-explicit nonempty shell pass, and lead authorizes the exact
|
||||
combined hash. No four-cell update. Then run the required product mutations,
|
||||
two fresh processes and official full/canonical gates under§45 with exact
|
||||
hashes and nonpassing classifications. Keep C1c's existing25-frame consumer;
|
||||
do not copy its replay. Those steps need their explicit execution release.
|
||||
|
||||
The fresh initialized0..6 part-order assertion is a production initialization
|
||||
witness, not a claim of universal equality with later live-retail retained
|
||||
part-list order0,1,6,5,2,4,3. AD117's source approximation and AP159's boundary
|
||||
ties remain explicit. C1a/G4 unpassed, FPS deferred, known retail leak unchanged.
|
||||
Loading…
Add table
Add a link
Reference in a new issue