nucleation

CLAIM-466: Lock-in from selectively admitted evidence and lock-in from self-scoring need different remedies: weighting admitted evidence by known, positive admission probabilities undoes the first and leaves the second, so correcting for how evidence was selected is not correcting how it is scored

2 min read

Properties8
StatusActive
Version2
Rolethesis
Warrantdemonstrated
Tagsself-governance, probabilistic-modeling, epistemology
Linedistributed-agency
GroundsCASE-162 — Capture and selectively admitted evidence in two lab realizations: in a four-variable model of authority, record and competence the equilibrium fold under self-scoring is at c ≈ 0.4247 while the declared history is trapped from c ≈ 0.4469, and in a 32-member event-level population, admission of cues that depends on the record reverses the sign of the admitted evidence and leaves the worse alternative in charge in 41 of 64 worlds, which exact inverse-propensity weighting brings to 0 of 64 below full selectivity
ComplementsCLAIM-404 — An organization can be governed from outside without any of its decisions being made for it, through the evidence its own procedures receive or through the criteria by which they weigh evidence; the two routes are distinct and each is possible without the other, so informational and evaluative independence do not imply each other; and where the organization's role is to evaluate another that affects its evidence or resources, the interpretation–authority loop can close between the two institutions, so that formal independence of the evaluator's procedure does not make its evaluation independent, CLAIM-297 — When an operator's authority is updated by how well it performs on the interpretation it helped make dominant, the coupled meaning and influence of an organization can amplify an initial interpretive advantage without any gain in fidelity to independent evidence

Graph View

The claim

CASE-162’s v1.3 results, as stored, worlds with a wrong incumbent out of 64, no self-scoring: at selectivity 0.5, 0.75 and 1, naive credit leaves 31, 41 and 41 wrong; exact inverse-propensity weighting leaves 0, 0 and 5, the 5 where admission probabilities near zero break the correction’s positivity condition. At s = 0.75 selection reverses the sign of the evidence: the raw cue mean is −0.193 and the admitted mean +0.184. And: “Self-scoring alone, c = 0.5 with s = 0, leaves 39 of 64 wrong. Nothing is censored there, so the correction changes nothing.” With both, the correction removes the part due to selection (at c = 0.25, s = 0.5, 41 to 0; at c = 0.5, s = 0.5, 41 to 34).

CLAIM-404 says an organization can be governed from outside through the evidence its procedures receive or through the criteria they weigh it by, and that the two routes are distinct. The runs show the remedies are distinct too: knowing how evidence was admitted repairs the first route and is idle on the second, because a self-scored record is biased in what it counts, not in what it was shown. CLAIM-297’s amplification without fidelity is the self-scoring half; CASE-156 the selection half.

What it does not say

  • A procedure for unknown propensities. The correction was given the exact admission probabilities, as an oracle (CASE-162: “A procedure for unknown or zero propensities” is what it cannot show).
  • The remedy for self-scoring. Independent scoring is the natural one (CLAIM-342’s evidence that does not pass through the interpretation it tests); the runs do not test it.
  • Anything beyond the stipulated population of the case.

Backlinks