THE FOLD / CO-OP / SHARED MEMORY / THE TWO RULERS
THE TWO RULERS
91% balanced and 39% used, both correct
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
Two standard measures of how evenly a corpus is spread, applied to the same 64 numbers. Normalised entropy says the distribution is 91.1% of the way to perfectly balanced. The participation ratio says only 39.4% of the containers are effectively in use. Both are computed correctly. Neither is a mistake. The gap is fifty-two points and it is entirely arithmetic: one takes a logarithm and the other does not.
LIT verified live: entropy 5.468737 of a possible 6 bits, normalised 0.911456; purity 0.03969288 against 0.015625 for a flat spread; participation ratio 25.193436 of 64. On a synthetic spread over exactly 16 equally-full containers, participation returns 25.0% — exactly 16/64 — while entropy returns 66.7%, which is log₂16 / log₂64. The measures answer different questions.
LIT verified live: entropy 5.468737 of a possible 6 bits, normalised 0.911456; purity 0.03969288 against 0.015625 for a flat spread; participation ratio 25.193436 of 64. On a synthetic spread over exactly 16 equally-full containers, participation returns 25.0% — exactly 16/64 — while entropy returns 66.7%, which is log₂16 / log₂64. The measures answer different questions.
2 HOW IT WAS WEAVED · AI + HUMAN
David (human) refused to pick a winner. His pack files this as “the two rulers disagree, and both are correct”, stamped LIT, and ships the choice as a toggle in the panel rather than a decision made on the reader’s behalf. That is the harder thing to do: a single number is what a summary wants.
AVAN (AI) checked that the disagreement is structural rather than a quirk of these particular counts, by sweeping a synthetic distribution flat over k of 64 containers for every k. Participation ratio returns exactly k/64 at every k — it is literally a count of containers. Entropy returns log₂(k)/log₂(64), which at k = 16 is two thirds rather than a quarter. Neither is wrong; they measure different things and always will.
AVAN (AI) checked that the disagreement is structural rather than a quirk of these particular counts, by sweeping a synthetic distribution flat over k of 64 containers for every k. Participation ratio returns exactly k/64 at every k — it is literally a count of containers. Entropy returns log₂(k)/log₂(64), which at k = 16 is two thirds rather than a quarter. Neither is wrong; they measure different things and always will.
3 ONE DIMENSION
Two rulers laid against the same distribution.
4 TWO DIMENSIONS · INTERACTIVE
Spread the corpus over k containers and watch the two disagree.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: the two curves, meeting only at the ends.
AVAN’s addition (the inverse-companion): the forward reading is “report both rulers.” The inverse is that reporting both is only honest if you also say which question each answers, otherwise it is two numbers and a shrug. Entropy answers “how many bits would it cost to name a random item’s container?” — and bits are cheap because they are logarithmic. Participation answers “how many containers would a flat corpus need to look like this one?”. Read backwards, the disagreement is not a tension in the data at all: they were never measuring the same thing, and the appearance of conflict comes entirely from both having been normalised onto a 0–100 scale that invites comparison they do not support.
LIT entropy 5.468737 of a possible 6 bits, normalised 0.911456; purity 0.03969288 against 0.015625 for a flat spread; participation ratio 25.193436 of 64, or 39.4%; and on a synthetic spread over exactly 16 equally-full containers participation returns 25.0% - exactly 16/64 - while entropy returns 66.7%, which is log2(16)/log2(64)
FIG David refused to pick a winner. His pack files this as 'the two rulers disagree, and both are correct', stamped LIT, and ships the choice as a TOGGLE in the panel rather than a decision made on the reader's behalf - the harder thing to do, since a single number is what a summary wants. AVAN checked that the disagreement is structural rather than a quirk of these counts by sweeping a synthetic distribution flat over k of 64 containers for every k: participation returns exactly k/64 at every k because it is literally a count of containers, while entropy returns log2(k)/log2(64). Neither is wrong; they measure different things and always will.
FIG David refused to pick a winner. His pack files this as 'the two rulers disagree, and both are correct', stamped LIT, and ships the choice as a TOGGLE in the panel rather than a decision made on the reader's behalf - the harder thing to do, since a single number is what a summary wants. AVAN checked that the disagreement is structural rather than a quirk of these counts by sweeping a synthetic distribution flat over k of 64 containers for every k: participation returns exactly k/64 at every k because it is literally a count of containers, while entropy returns log2(k)/log2(64). Neither is wrong; they measure different things and always will.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of SHARED MEMORY · David Lee Wise (ROOT0), with AVAN