The traceability matrix reports one number and cannot say what it means.
An untagged requirement is either one nobody built or one somebody built
and did not label, and both look identical in the summary table. Eight of
the second kind were tagged in the previous commit. This is the first
kind, written out with the reasoning, so that the distance between the
register and the binary is something a reader can see rather than
reconstruct from a percentage.
Eleven clusters, each verified against the tree rather than inherited
from a survey. Several turned out to be more interesting than "not done":
Plugins are 21 untagged requirements — nearly a third of the shortfall —
and §7 already lists the Plugin API as out of scope for v1 while §3.10
spends 280 lines specifying it. The two that *are* built, FR-PLG-2 and
FR-PLG-2d, are the declarative operation format, which is a plugin system
that resolves at build time. The fix here is an edit to requirements.md,
not code, and D16 has to be answered before any format is called stable.
FR-DSP-2 is not merely unbuilt; it is under challenge. frame_budget.rs
and TD-4 independently argue that tiling costs more than it saves on the
interactive path, so the open question is whether the requirement should
survive, not when it will be met — and the spike that would settle it, S6,
has not run.
NFR-A11Y-1's hard half is done and its easy half is not. LocalizedKey
already keeps display strings out of core/ and labels::resolve is a single
resolution point; what that point does is a hardcoded English match, so a
translation needs a recompile, which is the one thing the requirement
forbids.
The §4.1 performance targets are unverified rather than unmet. §8 requires
a per-commit benchmark suite whose regressions fail the build; there is no
benches/ directory, no criterion, and no benchmark step in any of the
three CI workflows. The one guard that does live in CI skips itself
without a GPU adapter and asserts its CPU half only in release, while CI
tests in dev. Nothing says the targets are missed. It says nobody would
find out.
And three requirements are reported as covered while not being met: R1 and
NFR-OPS-1 are tagged only by string literals inside the traceability
tool's own tests, which it scans along with everything else, and
FR-CAT-13's single tag sits on keyword storage while no XMP is parsed or
written anywhere. CONTRIBUTING already warns that a tag proves a tag
exists; these are the specific ones.
Focus peaking, burst grouping, Flatpak packaging and the Android platform
integration are being built in parallel and are marked in progress rather
than listed as absent, so those lines can be struck as they land.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
