The brief's surface prose has drifted from the shipped binary in twenty-one

The brief's surface prose has drifted from the shipped binary in twenty-one places the machine-checked layer cannot see. These are the crosscheck findings that predate v0.7.0, kept apart from the fifteen this release introduced ([[iss-2609011424116881]]). All are recorded with evidence in the v0.7.0 receipt at .abcd/work/reviews/4b4076a10f89d4d02da359274dad8994e30cae0e/iss35-brief-surface-crosscheck.json, dispositioned deferred-pre-existing.

The shape of it

The machine-checked layer is clean. .abcd/development/release/surface.json against the live binary is byte-exact across 82 command nodes and 109 flags with zero drift, and every ## Sub-verbs table in every chapter verified correct. Every finding here is in prose that surface_coverage does not scan, which is exactly the gap 04-surfaces/README.md:33 names when it says prose is not machine-scanned and review must cover it.

That is the finding behind the findings: the parts of the brief a rule can check are right, and the parts only a reader can check have rotted. Counting them is less useful than noticing they cluster in flag enumerations, criteria lists, templates and cross-surface counts, none of which any gate reads.

Three that actively mislead

Following the brief as written produces something the binary refuses.

• 04-surfaces/06-capture.md:33,35,115,116 documents capture promote and capture resolve without --grounds, which is REQUIRED on both. The chapter's own acceptance criteria give invocations that exit non-zero.
• 04-surfaces/05-intent.md:216-269 gives a canonical intent template with no ## Grounds section. The shipped intent ready gate requires one, so an intent authored to the brief's own template cannot pass it.
• 04-surfaces/05-intent.md:3,200 names six readiness criteria; seven ship, and the missing one is grounds, which is the check that actually fails in practice.

Note the pattern: all three are the same --grounds and ## Grounds work landing in the binary without landing in the chapters that describe it.

The rest

• 05-intent.md:3,200 calls intent ready read-only; --grounds records the conjecture, so it writes.
• 06-capture.md:28 omits --production-mode (ships on create, promote, resolve and wontfix); :35 omits --shipped-in; :51-81 asserts its frontmatter block mirrors the schema exactly while omitting origin and production_mode.
• 01-ahoy.md:288-296 omits --attribution and the prepare-commit-msg hook it installs; :266-275 omits that same shipped default template; :36-39,433-440 omits the provenance record uninstall removes and its --bin-dir flag.
• 02-disembark.md:28-34 presents sub-verb arguments as the shipped surface and names no flags; --include-ignored ships on pack, probe and plan.
• 04-launch.md:22 says the plain-text --dry-run render prints only five things; it also prints a receipts: line.
• 19-identity.md:35 quotes a Tagline that does not match abcd's own canonical identity block. Title and Pitch match byte-for-byte; the Tagline does not. This is the drift class iss-143 and the identity-positioning rule exist to catch, and the brief chapter is not a registered surface in .abcd/positioning.json, so nothing gates it.
• 22-site.md:6-7,54,97 states the bare form is strictly read-only and build is the render; site check renders when the output directory holds no index.html, a write path the chapter never records. The plugin page documents it; the chapter does not.
• 05-internals/03-configuration.md:361 names a fictional ahoy destroy sub-verb.
• 06-delivery/01-build-sequence.md:29 names skills/ as a shipped markdown surface directory; it does not exist, as three other locations correctly say.
• 04-surfaces/README.md:12,16,17 index rows omit capture disposition, the docs cite sub-tree, and history drain/history staged. All three are correct in their own chapters; only the index is stale.
• 17-guard.md:141-143 cites iss-315 as tracking an unimplemented design target; iss-315 is resolved and explicitly scopes that work out, so no open issue tracks it. internal/core/guard/payload.go:191 makes the same citation.
• 14-ingest.md:10-11 says no Go verb backs ingest. True at top level, but reading ingest and memory ingest now exist as binary sub-verbs, so the phrase is falsifiable as written.

Grounds

• deferred: not fixed in the v0.7.0 cut, because none of it is caused by or blocking this release and the cut was already two rounds deep in gate corrections; deferring was the maintainer's recorded decision at PROMOTE

• declined: duplicate of iss-2609011424033603: the same twenty-one pre-existing findings captured twice from one receipt; this record carries the fuller body, and the resolution with fixing-commit provenance is recorded on the earlier id

Worth deciding, not just fixing

Three of these are index or enumeration rows that are correct in one place and stale in another, and two chapters maintain the same bare-status-board list independently. Fixing the instances leaves the duplication that produced them. Whether any of these enumerations should be derived rather than restated is the question underneath the list, and it is the same question [[iss-2609011423385217]] raises about the release-gate manifest's pinned content list.