Validate byte-preserving record fields as base64 strings

docs/integrations/publication-records.md

@@ -42,7 +42,9 @@ endpoint bytes uses version 2 with the same `evidence` object and an additional
 `evidence_bytes` object. Each entry in `evidence_bytes` uses an evidence field
 name and a standard base64 string containing its exact bytes. That entry
 supplies the field's value when loading; the writer omits its lossy string form
-from `evidence`. Unknown field names are ignored. This representation applies
+from `evidence`. Invalid known-field types or base64 encodings produce a
+record-format error with the available decoding cause. Unknown field names are
+ignored. This representation applies
 to every evidence field and preserves the same identity as typed Go evidence.
 Older CLIs that support only version 1 report an unsupported version for these
 byte-containing records; use a CLI supporting version 2 to load them.

docs/roadmap/implementation.md

@@ -140,7 +140,11 @@ mismatch tests confirm that no remote tag query occurs.
 without relying on diagnostic output.
 
 1. Add record encoding and storage helpers under `internal/publish`. Use a JSON
-   envelope with integer `format_version: 1` and an `evidence` object. Its fields
+   envelope with integer `format_version` and an `evidence` object. Ordinary
+   UTF-8 evidence uses version 1; non-UTF-8 fields use version 2 and a separate
+   `evidence_bytes` object of base64 strings, preserving exact supported bytes.
+   See the [canonical format](../integrations/publication-records.md#private-attempt-record).
+   The evidence fields
    are `repository_dir`, `module_dir`, `module_path`, `version`, `tag`,
    `candidate_commit`, `tag_object`, `endpoint`, `remote`, and `recorded_at`.
    Store the exact endpoint. Fields other than the envelope version/object can
@@ -150,7 +154,10 @@ without relying on diagnostic output.
    malformed JSON, and structural type errors are `record_format` failures.
    Accept additional fields within a supported version. Semantic evidence
    completeness is evaluated by reconciliation, so a readable partial record
-   remains usable for discovery and any classification it can support.
+   remains usable for discovery and any classification it can support. Decode
+   known version-2 byte fields as strings, then decode standard base64. Invalid
+   structural types and malformed base64 return `record_format` with their
+   available causes. Additional unknown fields retain their existing handling.
 3. Resolve the default directory with
    `git rev-parse --path-format=absolute --git-common-dir`, then append
    `releasekit/publication-attempts`. Use Git metadata discovery rather than
@@ -192,6 +199,11 @@ ignored, and unignored override. Use injected failures at the small file/Git
 boundary when platform permissions cannot reliably induce a failure. Verify
 that private records retain synthetic credentials while diagnostics redact them.
 
+**R17 follow-up completed:** Known byte fields validate their string representation
+before base64 decoding. Tests cover valid/empty strings, numeric arrays, other
+structural errors, and retained JSON/base64 causes. Existing all-field byte
+round trips and original-versus-decoy Git endpoint observations remain covered.
+
 ## Stage 3: Add the facade-owned recovery interfaces
 
 **Completed:** 2026-11-02
@@ -522,6 +534,8 @@ contracts and a short consumer recovery workflow.
    explicit push readiness and fresh later invocations.
    R16 covers setup failure classifications and human/JSON guidance without
    querying an invalid or substituted endpoint.
+   R17 adds byte-field structural and encoding failures at the codec boundary,
+   reusing the existing partial-record and lossless-observation fixtures.
 2. Establish read-only behavior through resulting refs, worktree/branch state,
    saved records, and controlled observations of candidate/hosted boundaries.
    Reuse lower-level classification and codec tests. Tests use local remotes,

docs/roadmap/remediation.md

@@ -1,8 +1,8 @@
 # Publication reconciliation remediation
 
-Status: open. R01–R13 are addressed and individually verified. R14–R17 record
-remaining findings from the subsequent stage-by-stage review of `363f7cd`.
-Address the two handoff defects first, then the classification and format gaps.
+Status: all corrections are implemented and individually verified; final
+assembled verification is pending. R01–R13 retain their completion evidence.
+R14–R17 describe the subsequent stage-by-stage review of `363f7cd` and its fixes.
 
 R01–R13 describe commits `4dff808` through `599ad81`, against planning baseline
 `447b42a`. Their audit evidence and defect descriptions refer to that reviewed
@@ -734,6 +734,12 @@ reaching a substituted endpoint when a rewrite mismatch is detected.
 
 ## R17: Validate the documented version-2 byte-field representation — P3
 
+**Completed:** Known byte fields load through explicit JSON string validation and
+standard base64 decoding. Tests passed for valid/empty strings, numeric arrays,
+other structural types, malformed base64, and retained decoding causes, alongside
+existing partial/unknown-field, version-1, and lossless endpoint regressions.
+Stage 2/8 instructions and the canonical format guide are updated.
+
 **Relevant code:** `internal/publish/record.go:48–59`;
 [private record format](../integrations/publication-records.md#private-attempt-record).
 Affects Stage 2 and its Stage 8 format coverage.