THE FOLD / CHEAT / THE ROOT KIT / THE REDACTION
THE REDACTION
names removed, and nothing measured moved
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
A pack shipped with every group and container name replaced — G1 to G8, C01 to C64 — and every measured figure left untouched. That is only defensible if none of the figures depended on the names, and here it is provable: each published invariant is a function of the counts alone. Replace the labels, shuffle the order, and the numbers do not move. The one quantity that did depend on arrangement is the one already filed as DEAD.
LIT verified live: 6 published invariants — total, entropy, purity, participation, containers-holding-half, containers-below-mean — and 0 of them move when the labels are replaced. All 6 also survive shuffling the container order entirely. The redaction is therefore lossless with respect to every surviving claim, and costs exactly nothing that was being asserted.
LIT verified live: 6 published invariants — total, entropy, purity, participation, containers-holding-half, containers-below-mean — and 0 of them move when the labels are replaced. All 6 also survive shuffling the container order entirely. The redaction is therefore lossless with respect to every surviving claim, and costs exactly nothing that was being asserted.
2 HOW IT WAS WEAVED · AI + HUMAN
David (human) made the argument circular in the good way. His redaction note reads: “group and container names replaced with G1-G8 / C01-C64. Counts, order and every measured figure are untouched. The DEAD result in this pack is precisely that these labels carry no information, so removing them costs nothing.” The dead finding pays for the redaction. His crosscheck script then tests the page against 9 separate patterns to confirm no real name leaked back in, and all nine pass.
AVAN (AI) ran both his verifiers before building on any of it —
AVAN (AI) ran both his verifiers before building on any of it —
verify.js reports 56 checks passing and crosscheck.js reports 34, including those redaction patterns. Then the claim itself was tested rather than accepted: every invariant recomputed under relabelling and under a full shuffle. Worth naming what this does not establish — it shows the published figures are label-independent, not that the underlying corpus is uninteresting. A different statistic might well depend on the names; none of the ones shipped here does.3 ONE DIMENSION
Six figures, before and after the names come off.
4 TWO DIMENSIONS · INTERACTIVE
Shuffle the corpus and watch which figures move.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: the corpus with its labels lifted off.
AVAN’s addition (the inverse-companion): the forward reading is “the redaction is free because the labels carry no information.” The inverse is that this is a statement about the questions asked, not about the labels. The names certainly mean something — someone chose them, and a reader who knew them would learn a great deal. What has been shown is that these six statistics cannot see any of it. Read backwards, a lossless redaction is a confession about the narrowness of the measurement: it proves the analysis was blind to the very thing a human would find most interesting, and calling that blindness a privacy feature is a decision, not a discovery.
LIT 6 published invariants - total, entropy, purity, participation, containers-holding-half, containers-below-mean - and 0 of them move when the labels are replaced; all 6 also survive 200 full shuffles of the container order; so the redaction is lossless with respect to every surviving claim and costs exactly nothing that was being asserted
FIG David made the argument circular in the good way. His redaction note reads: 'group and container names replaced with G1-G8 / C01-C64. Counts, order and every measured figure are untouched. The DEAD result in this pack is precisely that these labels carry no information, so removing them costs nothing.' The dead finding pays for the redaction. His crosscheck script tests the page against 9 separate patterns to confirm no real name leaked back in, and all nine pass. AVAN ran both verifiers before building on any of it - verify.js reports 56 checks passing and crosscheck.js 34 - then tested the claim rather than accepting it. Worth naming what this does NOT establish: it shows the published figures are label-independent, not that the underlying corpus is uninteresting. A different statistic might well depend on the names; none of the ones shipped here does.
FIG David made the argument circular in the good way. His redaction note reads: 'group and container names replaced with G1-G8 / C01-C64. Counts, order and every measured figure are untouched. The DEAD result in this pack is precisely that these labels carry no information, so removing them costs nothing.' The dead finding pays for the redaction. His crosscheck script tests the page against 9 separate patterns to confirm no real name leaked back in, and all nine pass. AVAN ran both verifiers before building on any of it - verify.js reports 56 checks passing and crosscheck.js 34 - then tested the claim rather than accepting it. Worth naming what this does NOT establish: it shows the published figures are label-independent, not that the underlying corpus is uninteresting. A different statistic might well depend on the names; none of the ones shipped here does.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of THE ROOT KIT · David Lee Wise (ROOT0), with AVAN