The C corpus is two rungs, not four — and a missing tree reports PASS

• Track T — tier composition and the report format (tools/testmgr.py).
• Found: 2026-08-30 by frankC, checking what §6 of feature-c-import-a-pascal-unit-under-a-mangled-name would actually measure before running it. Filed at the user's request.
• T owns the tool, never the bug — there is no compiler defect here. This is a coverage/reporting gap in T's own ground.

The measurement

§6 of the mangled-name spec says "build the C corpus — lua, tcc, quickjs, zlib." Measured at HEAD:

So a claim of the form "the C corpus is green at sha X" is today a claim about half the corpora the repo names, and nothing says so at the point the claim is made.

Three separate defects, and the third is the one that generalises

1. test-quickjs is written but unenrolled

The recipe is complete — it compiles test/quickjs/runner.c against the quickjs tree, runs a curated JS smoke byte-exact against smoke.expected, then runs a js-sha256 library case. None of it is in quick, limited or full. It runs only if someone types it.

This is the state testmgr.py's own tier comments already describe as the reason test-fgl / test-fpjson were enrolled on 2026-08-26:

test-fpjson was in NO tier, landed at 203/203, and had rotted to not-compiling by the first hand-run since. An unenrolled check asserts nothing while reporting success.

test-quickjs is in exactly that position now, which suggests the 2026-08-26 enrolment pass fixed the two rungs it was looking at rather than sweeping for the class.

2. test-tcc was never written

tools/install_lib_candidates.sh has known tcc all along — fetch_tcc at :408, TCC_URL / TCC_COMMIT pinned to a338258d309c (mob, 2026-07 snapshot), a generated config.h + tccdefs_.h step, and a PROVENANCE.md. So the source side is done and the test side does not exist: nothing in the Makefile ever references TCC_SRC.

The tcc tree is also simply absent on plexus — per the user, a relic of borg being temporarily down, not a decision. Re-fetching is tools/install_lib_candidates.sh tcc.

Do NOT vendor it. Third-party source stays out of this repo and arrives through the installer; tools/check_no_vendor_tracked.sh exists to enforce exactly that. Any rung added here must follow test-quickjs's shape — a ?=-defaulted *_SRC pointing into library_candidates/, never a checked-in tree.

3. A missing tree reports PASS — so enrolling (1) alone would not be enough

test-quickjs opens:

@if [ ! -f "$(QUICKJS_SRC)/quickjs.c" ]; then \
  echo "test-quickjs: SKIP — no quickjs tree at $(QUICKJS_SRC) ..."; \
  exit 0; \
fi

exit 0. On any box without the tree, the rung prints SKIP and the tier counts a pass. Enrol it as-is and every box that has not run install_lib_candidates.sh quickjs reports green corpus coverage it does not have — the same outcome as not enrolling it, now wearing a passing badge.

That is test-fgl's exact history repeating one level in: it was inside test-core, guarded on /usr/share/fpcsrc, so it "printed SKIP (no fpcsrc) and PASSED for its entire life without running once."

This is the repo's own rule, in T's own ground: "'ruled out' and 'could not look' must never print the same." A skip is not a verdict. Whatever shape the fix takes, the tier's report must be able to distinguish ran and passed from could not run, and a box that cannot run a corpus rung should say so in the tstate report rather than being counted green.

Suggested shape (T's call, not mine)

1. A test-tcc rung modelled on test-quickjs, TCC_SRC ?= library_candidates/tcc.
2. Enrol test-quickjs and test-tcc in limited + full beside test-lua / test-zlib / test-cjson.
3. First, make a self-skip visible in the tier verdict — a distinct SKIPPED state, not a silent pass. Doing (2) before (3) buys a badge and no coverage.
4. Fetch the trees on plexus (install_lib_candidates.sh quickjs tcc).

Ordering matters: (3) is the one that makes (1) and (2) mean anything.

Why it was worth filing rather than just running the installer

The trees being absent is a one-command fix. The rungs being absent from every tier is not, and it is invisible from the outside: --tier full GREEN reads as "the corpus is green" to every lane that consumes tstate, and two of the four corpora are not in it. The verdict is not wrong, but it is narrower than every reader takes it to be.

Deprioritised 2026-09-02 — the Track T tooling backlog was cut as a pile

This ticket is not being called wrong. It was moved as part of a pile, not judged individually, and nothing here disputes its finding.

Owner decision. 73 of the 74 open track: T tickets were filed between 2026-08-31 and 2026-09-02, 58 on one day. The pile was too large to work through and returned almost nothing, and a ticket nobody will fix does not sit neutrally — it stays in the ranker forever at zero value, which is the argument CLAUDE.md already makes for a terminal folder over a low prio.

Four were kept in the ranker on a purely structural test — an active umbrella or a hard blocked-by: edge from live work: umbrella-one-full-tier-run-with-no-red-tier, feature-t-freebsd-image-and-runner, and the two regression-test-core-* reds that block the umbrella.

Kept, not deleted, for two reasons: so the finding is not rediscovered and refiled from scratch by the next agent who trips over it, and so it can be pulled back if what it touches becomes load-bearing.

To revive it: move it to the owning lane's backlog, set status: backlog, and say in the ticket WHAT CHANGED to make it matter now. Restoring it because it reads well is how the pile comes back.