Independent region-census acceptance: REJECT
Historical source last updated, 2026-10-27. Runtime and implementation claims below describe that dated source; see current app notes and source for subsequent changes.
Exact candidate 66436f785bbe3166a992402f95abf083ca0d6736, compared with base
1cf9788cc3ac807027206c39fcbf1d2771599845, falsely succeeds when region
declarations change between the four combined build gates. No candidate repair
or main integration was performed.
The first behavioral checkpoint ran the actual original CLI in four separate
processes and the exact candidate CLI with --build-gates against the unchanged
portable fixture. Both exited 0 with byte-identical stdout and stderr. The build
file retains the command order --gate, gen-region-map --check, then
--build-gates; this order was inspected/asserted, not executed as a full build. The four original processes execute js-copies, embedded-wat, dup-payloads,
hand-rolled, stopping on first failure.
The next checkpoint planted one change after the first read of host.js, after
the js-copies gate had read declarations and computed its layout:
(region.declare $INDEPENDENT_INVALID (size 0x1000) (align 3))
| Path | Exit | Observable outcome |
|---|---|---|
| Original separate processes | 1 | js-copies passes; embedded-wat rejects (align 3) is not a power of two |
| Exact candidate combined process | 0 | All four gates report OK; no stderr |
tools/region-census.js:694 retains the first collectDeclarations() result;
the embedded-WAT scan reuses it through declarations(options) at line 629. tools/region-layout.js:87 also returns a memoized layout before rereading its
source. Grouping previously separate processes makes that existing cache span
gates. Later hand-rolled scanning reads the changed declaration text but does not
recompile it. Thus the candidate suppresses the validation the original next
process performs. This finding concerns changes between gates, not a claim that
an initially invalid declaration passes the first gate.
Acceptance stopped at this first demonstrated defect: eight actual CLI invocations total (five clean, two failing-path original, one failing-path candidate). No broad legacy/default/record/unsupported matrix, additional omission controls, main-dependency execution, or independent read/parse/layout reduction measurements were run after rejection. The unchanged fixture/baseline was used for the clean comparison; no baseline was edited to pass.
The donor's final-delta-results.json names census source SHA-256
86de689e82f7d942a6d81d3db78b8a77dbd67aebac44b65e8acbedbc9973b5b1,
matching the tested final candidate, rather than only the intermediate revision. Its retained final cases cover same-file JS text and JS directory discovery;
they do not establish declaration freshness between gates. Its prior test
results and reported reductions are inherited evidence, not this worker's
independent acceptance or independent measurements.