| TEXT | HIS | MINE | Δ |
|---|---|---|---|
| JaneEyre · I | 62.36 | 62.36 | 0.0000 |
| Wuthering · I | 54.22 | 54.22 | 0.0000 |
| Walden · I | 27.57 | 27.57 | 0.0000 |
| AgnesGrey · I | 56.66 | 56.66 | 0.0000 |
| JaneEyre · β | .6442 | .6377 | −.0065 |
| Walden · β | .6918 | .6891 | −.0027 |
| JaneEyre · voiced | 9.73 | 19.35 | ×1.99 |
| AgnesGrey · voiced | 6.68 | 13.07 | ×1.96 |
I-axis LIT — reimplemented from a
one-line method note, zero drift on four texts. That is why the gap result stands.
β AMBER — consistent −0.003…−0.007 bias, same sign
every time. A fit-detail difference, usable with the caveat.
voiced AMBER — clean ×1.97 on all four. Solved in W5.
20 texts added from Project Gutenberg, chosen to aim at the empty bands. All measurements are mine; none are the author's.
| WHO | WHAT |
|---|---|
| Fiddler | Built THE VOLUME and froze its bins. Then said put something there and see what happens — proposing the one experiment that could falsify his own published claim. |
| Claude | Reimplemented the pipeline from the method note. Calibrated against 4 published texts before measuring anything new. Added 20 texts. Brute-forced the voiced rule against 7 candidates. |
| Joint | The finding that voids track the shelf, not the language. Neither half of this gets there alone — he had the instrument, the test needed texts he hadn't picked. |
| Not verified | His β fit internals. The −0.005 bias is unexplained; I matched his window but not his sampling. |
Why calibration came first. Measuring 20 new texts and comparing them to his 14 is only valid if the pipelines agree. They agree on one axis exactly, one approximately, and one not at all — and knowing which is what makes the gap result reportable and the voiced result interesting rather than embarrassing.
What filled it. Machiavelli 5.95, Darwin 8.46, Marcus Aurelius 12.31. Every one is single-author reasoning prose — an author present as a thinker, absent as a character. "I observed," "I advise," "I remind myself" — never "I walked into the room."
The original 14 held epics, novels and manifestos. No treatise, no advice-book, no private journal. The void was the shape of the shelf. LIT
W3 → W1: the cliff claim is falsified. The instrument is not — it located a real structural region the moment it was handed texts from it.
The result nobody ordered. Twenty texts added and the largest gap grew: 8.01 → 7.98 → 8.71. Filling one void opened another at the far end, between Tenant (63.87) and Robinson Crusoe (72.58).
W4 → W1: voids are not features of the axis. They are features of whatever shelf you happen to have loaded. There will always be a largest gap, and it will always look like a cliff until someone throws a book at it.
▲ W5 DISSENTS TO W1.
W1 files voiced as AMBER-unresolved. It is resolved. I brute-forced seven candidate rules against
the four published values; exactly one reproduces all four:
(total quote chars) // 4.
And that is a double-halving. A quotation pair is two characters, so pairs = chars ÷ 2. Dividing by 4 halves it again. Every published voiced value is exactly half of what the axis label describes.
Now the reversal, and it matters. The slip is applied uniformly to all 14 texts — so it is perfectly rank-preserving. Every tercile boundary, every cell assignment, every conclusion drawn from the cube is untouched. Only the stated unit is wrong. LIT
Fix without breaking the freeze: leave the numbers, correct the label to
half-pairs per 1,000 or state the rule verbatim in method. One sentence and
the axis becomes reproducible by anyone holding only the index — which is the whole point of
freezing bins in the first place.