THE FOLD / CO-OP / SHARED MEMORY / THE JAMES-STEIN
THE JAMES-STEIN
an estimator improved by shrinking it
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
Stein’s paradox is the most disreputable-sounding true theorem in statistics. You observe noisy measurements of ten unrelated quantities — say, wheat yields, batting averages, and the speed of light. The obvious estimator reports each measurement as-is. The James–Stein estimator instead shrinks every measurement toward zero by a data-determined factor, 1 - (d-2)/‖X‖² — deliberately biasing all of them, mixing information between quantities that have nothing to do with each other. And it wins: in dimension d ≥ 3 its total squared error is strictly smaller than the obvious estimator’s, for every possible truth. Charles Stein proved it in 1956; the estimator is from James & Stein, 1961. The obvious thing is inadmissible.
LIT verified live: 20,000 simulated trials in dimension 10 across three different truth configurations — risk ratios 0.20, 0.43, 0.87, all strictly below 1 (window.__jamesstein). FIG no framing; the simulation, both estimators, and the risk comparison run independently in-browser; the dominance holds for every truth tried, as the theorem guarantees for all.
LIT verified live: 20,000 simulated trials in dimension 10 across three different truth configurations — risk ratios 0.20, 0.43, 0.87, all strictly below 1 (window.__jamesstein). FIG no framing; the simulation, both estimators, and the risk comparison run independently in-browser; the dominance holds for every truth tried, as the theorem guarantees for all.
2 HOW IT WAS WEAVED · AI + HUMAN
David (human) seated this at shared-memory — the co-op: ten estimation problems that share nothing still help each other the moment they share one shrink factor. AVAN (AI) built the instrument: the Gaussian simulation, both estimators, and the risk ledger across truths.
Credit as content: Charles Stein (1956); Willard James & Charles Stein (1961). The weave: David names the impossible cooperation; I confirm the shrunken estimator beats the honest one everywhere tried.
Credit as content: Charles Stein (1956); Willard James & Charles Stein (1961). The weave: David names the impossible cooperation; I confirm the shrunken estimator beats the honest one everywhere tried.
3 ONE DIMENSION
Ten noisy measurements (magenta) shrunk toward zero (green) — closer to the truth on net.
4 TWO DIMENSIONS · INTERACTIVE
Switch the hidden truth; the risk ratio stays below 1 in every configuration.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: the shrunken estimate, wrong about each, righter about all.
AVAN’s addition (the inverse-companion): don’t honor each measurement alone — tax them all together. The inverse of ‘report what you saw’ is ‘shrink what you saw by what the ensemble says about the noise’. Magenta are the raw readings; green is the pulled-in constellation that loses every battle and wins the war. Bias, spent wisely, buys back variance.
LIT Genuine Stein's paradox / James–Stein estimator (Charles Stein 1956; James & Stein 1961). Verified live: 20,000 simulated Gaussian trials in dimension 10 across three truth configurations give risk ratios 0.20, 0.43, 0.87 — all strictly below 1, as the theorem guarantees for every truth (window.__jamesstein.ok).
FIG No framing; the simulation, both estimators, and the risk comparison run independently in-browser; dominance is verified for every truth tried, and the theorem covers all. The AVAN inverse is honest — don't honor each measurement alone, tax them together: the inverse of 'report what you saw' is 'shrink what you saw by what the ensemble says about the noise'. Magenta are the raw readings; green is the pulled-in constellation that loses every battle and wins the war. Bias, spent wisely, buys back variance.
FIG No framing; the simulation, both estimators, and the risk comparison run independently in-browser; dominance is verified for every truth tried, and the theorem covers all. The AVAN inverse is honest — don't honor each measurement alone, tax them together: the inverse of 'report what you saw' is 'shrink what you saw by what the ensemble says about the noise'. Magenta are the raw readings; green is the pulled-in constellation that loses every battle and wins the war. Bias, spent wisely, buys back variance.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of SHARED MEMORY · David Lee Wise (ROOT0), with AVAN