test(01-07): integrate counted fidelity regressions and exact local results

Run existing fidelity Python, Node and observer checks with positive-count guards and explicit TAP output. Record the completed 782.05-second local suite and preserve human acceptance at 0/120.
This commit is contained in:
UnixNight committed 2026-11-13 16:44:59 -05:00
1 parent b732256751
commit 472500bf4e
5 files changed
+128 -1

No files matched your search

+21
View File
@@ -1,5 +1,26 @@
# Decisions
## Final fidelity audit — phase 1 plan 01-07
Readiness verifies immutable prepared bytes and current copied materials;
strict acceptance separately requires complete original/native matrix coverage,
matching package/source/resource identities and receipt-bound human judgments.
Structural `--allow-open` success never promotes acceptance. Preserve raw collector
receipts and failed/superseded history; the integration agent performs explicitly
specified evidence-backed post-return normalization, while the tester only
collects, observes and returns local files. A1/A2, both unclassified assumptions
and P1/P2/P3 remain visible until actual review. No universal fidelity tolerance,
automatic physical/audible pass or movement/landing reset is introduced.
Current result: ready11,000 files; strict audit human_needed,0/120. Positive-count
guards cover the existing fidelity suites in one consolidated local run; Node
uses explicit TAP so installed reporter defaults cannot hide the executed count.
The completed local run exits 0 in 782.05s: 556 Python (180 fidelity), 27 Node,
23 observer and 2 separate HUD Python tests; all 51 managed test entry points,
the main build, 36 GPU rendered receipts, 17 material/readback assertions,
dummy-audio checks and 539 original hashes pass. These bounded regression facts
do not resolve physical/audible/native/original acceptance.
## Linux capture access — phase 1 plan 01-06
A common failed native display/audio preflight may explain all40 Linux open