Standard NAL revision overestimates confidence by assuming independent evidence sources. The pessimism lens research (g140-g153) proved that the Fréchet mean on the confidence manifold requires shrinking the fused weight by λ≈0.49. This operator implements that correction as a drop-in MeTTa function.
| Input 1 | Input 2 | NAL Revision | Corrected | Fréchet |
|---|---|---|---|---|
| (0.7, 0.5) | (0.6, 0.4) | (0.66, 0.625) | (0.66, 0.449) | c≈0.45 ✓ |
| (0.9, 0.8) | (0.3, 0.6) | (0.74, 0.857) | (0.74, 0.729) | Shrunk ✓ |
| (0.8, 0.9) | (0.8, 0.9) | (0.80, 0.947) | (0.80, 0.898) | Shrunk ✓ |
(= (rev-w $c) (/ $c (- 1 $c)))
(= (corrected-revise (stv $f1 $c1) (stv $f2 $c2))
(let* (($w1 (rev-w $c1))
($w2 (rev-w $c2))
($ws (+ $w1 $w2))
($fr (/ (+ (* $w1 $f1) (* $w2 $f2)) $ws))
($cr (/ $ws (+ $ws 1)))
($wc (* (/ $cr (- 1 $cr)) 0.49))
($cc (/ $wc (+ $wc 1))))
(stv $fr $cc)))
!(corrected-revise (stv 0.7 0.5) (stv 0.6 0.4))
!(corrected-revise (stv 0.9 0.8) (stv 0.3 0.6))
!(corrected-revise (stv 0.8 0.9) (stv 0.8 0.9))
Raw source: paste.rs/hSgr6