This suite contains 34 cases and 1,331 probes per revision against the three original WAByteArray exports: intToBytes, uint8ArrayToBuffer and compareArrayBuffer. All 35 tests passed against actual archived sources at 1051710191 and 1051908825 through the local core Subject, including unchanged comparisons. The run used explicit diagnostic process mode (isolated:false) with runtime source digest 7609874cb9622bf1d7df6146b123b3a25618ae5217d4cc139bda3e2eca5fab6f; this is partial leaf behavior coverage.
Integer byte expectations use a separate Buffer/BigInt authoring oracle: truncate finite numbers, encode their unsigned 32-bit word in big-endian order, then take low bytes or add zero left padding. Widths zero through eight and selected larger widths cover boundaries, signed values, overflow and fractions. Literal tables characterize selected numeric coercions, nonfinite inputs, fractional widths and native errors followed by a valid retry. They do not establish a validated integer API or exhaustive numeric domain.
Returned-object histories retain actual intToBytes Uint8Arrays, convert those same objects twice through the original buffer conversion, distinguish the resulting ArrayBuffers by identity and compare their bytes through the original comparator. Offset views include prefix/suffix sentinels and empty slices. These observations cover bytes and distinct conversion objects; they do not claim backing-store mutation isolation, since no mutation seam is used.
Comparison cases use independently generated byte sequences, every single-bit change in bounded selected lengths, length mismatches and literal coercive inputs. The source guards only falsy comparison inputs and otherwise delegates input conversion to Uint8Array; uint8ArrayToBuffer accesses buffer/view fields without explicit validation. Unexpected coercions are characterized, rather than replaced with invented validation requirements.
Both source pins have the same import-free, 591-byte primary module and no alternate variants. The complete declared graph contains one node. Analyzer v4 consumes 2,436 units of its unchanged default two-million-unit budget, with no recognized gates or unresolved diagnostics. That narrow static inventory is not semantic proof or acceptance of unrelated WA codec, crypto or protocol behavior. Existing byte/protobuf suites do not select this root; the separate byte-buffer suite covers its own root and Latin1 behavior.
Run portable checks with bun test tests/byte-array.test.ts. A reviewed source adapter must advertise wa-internal:byte-array for implementation execution. No provider fixture, host callback, pending operation, bootstrap, browser, network, container isolation, production scheduling or named-library acceptance is supplied by this suite. Numeric/coercive domains, byte lengths and histories remain unbounded.