Closing the loop: the board's reach measured, the fleet's counts current, and no red row that is not a finding
Founder: the UI cannot go stale, and studies do not fail. Both were true and both are fixed here.
NOTHING FAILED. No study on this board failed. 90 programs compile and run, all 22 Studies 48/49
pins print and sit on their pages, and the two rows that were red were not study results:
shear-ledger-v1 is the payload SCHEMA for Study 44 - seven type declarations and no program.
Measured: print( 0, out( 0, emit( 0, FileHandle 0, the only file of 97 in reproduce/ with no
emit mechanism of any kind. It compiles, runs, exits 0 and writes 0 bytes, which is correct
behaviour for a file that declares types. The harness globs every *.swift and required non-empty
stdout, so it graded a declaration against a study's contract. That was the instrument's scope,
not a finding. It now discriminates structurally: a file that emits by none of the three
mechanisms carries no program, and is graded on what a declaration actually owes the tree -
some other file must compile against its types, or it is named an orphan. Both arms fire: a real
study with no output is still reported as one.
domain_count 50 vs 49 was never a failure. It is the instrument comparing the live court against
the pages and finding the pages stale, which is the job. Re-measured per cell with --resolve on
all nine A records, unanimous: 51 tools and 50 domains, against 49 and 49 measured 2026-09-02.
The 50th domain is cs, the Coding Court - roles compiler, consensus, ml_engineer - answering
AFFINE_MATH_COURT on 9 of 9 cells and correctly refusing an unnamed role. The fleet grew; the
pages were fifteen days behind.
THE UI NO LONGER CARRIES A STALE COUNT. Eight pages corrected to 51 tools / 50 domains, each with
a dated note naming the figure it replaced, per the standing rule that when a doc and a measurement
disagree the measurement wins and the doc gets a dated correction. The harness constant follows the
pages and now records why it moved, so the next reader knows 48 -> 49 was biosphere and 49 -> 50 was
the Coding Court rather than finding a bare number.
THE REACH IS THE BOARD'S, NOT A PAIR'S. Home.md gains a "Closing the loop" section in the voice the
founder asked for, and every figure in it is a measurement: one law holding across twenty-four
orders of magnitude, from Study 27's 3.27 fm lithium matter radius to Study 05's L1 point 1.5
million km sunward of Earth, with eleven studies populating the stretch between 3.840 Angstrom and
thousands of kilometres. What it does not claim is stated as plainly: the span is a range at its
bottom and a bracket at its top, five studies are OPEN because the data is not here yet, and one
study reports a measurement of our own that proved nothing and is kept for that reason.
AND THE NEW DISCRIMINATOR WAS GREEN FOR THE WRONG REASON ON ITS FIRST RUN. Its consumer check was
written `grep -l "$ty"`, so shear-ledger-v1 reported as consumed by aso-offtarget-exact-vs-float
because that file contains "Bases" and one declared type is `Base`. 719 passed, 0 failed, on a
substring. Repaired to require the type delimited by a non-identifier character on both sides;
re-tested, the schema is now correctly named an ORPHAN (zero whole-word references to any of its
seven types anywhere in reproduce/) and the Study 48 program, which declares three types and emits
67 times, is still graded as a program. A green row earned by accident is worse than the red row it
replaced.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>