li-deference-collapse-scoped
and · drafted
Statement
For two deductive processes with the novice's extending the expert's, Total Trust of the novice over the composite bets, the expert's concentration on near-maximal options, its asymptotic self-knowledge and conditional-stability of the menu sequence imply that the novice asymptotically weakly prefers following the expert's argmax choice to each fixed option.
Source
projects/deference/note-dump-2026-08-11/wiki/total-trust-implies-value.md, the Theorem and Hypotheses (H1)–(H3).
projects/deference/faf-cleanroom-2026-10-02/packages/def-argmax-value/def-argmax-value-report.md, Target 8 (scoped_value) and Target 4 (CondStableOn); projects/deference/faf-cleanroom-2026-10-02/DIGEST.md §2 item 14.
Notes
The Lean-checked refinement of the toy collapse's second arrow, with the scope condition named. It is the evidence parent of the deck's first arrow. Every known non-vacuity witness of its full package lies in a decisive or dominance stratum (report, State), which is the question scoped-value-content-witness.
The hypotheses are the conditions the theorem uses, judged literally (do they hold of the novice-expert pairs the programme is about); li-deference-collapse-scoped@valid reads the theorem as an implication.
Bridge card
Connected to no informal claim yet: held by no frame.
Declaration Cleanroom.Deference.DefArgmaxValue.scoped_value (record li-deference-collapse-scoped.record)
The Lean hypotheses hTT (Total Trust on the composite), hfold (the concentration fold), hramp with comp (ramp quotes on a composite the novice's ledger prices) and h3 (CondStableOn) are the wiki theorem's (H1), (H2) and (H3), and the conclusion is its argmax Value inequality.
Hypotheses, read:
- scoped-h1-total-trust-composite open: The novice totally trusts the expert over the composite bets the scoped
theorem builds from the menu options and the expert's quotes.
The deference hypothesis itself. For the self-expert its only known inhabitant is the expert reading itself (the docstring's N− witness); for a human novice and a distinct AI inductor it is the trust the whole programme asks about. - introspective-concentration-lemma proved-in-model: The expert's self-prediction mass concentrates on near-maximal-quoted options:
the expert's probability of selecting an option whose quote is at least ε
below the maximum tends to zero.
Proved in the clean room for the self-expert (concentrates_self) and, from the fold clause, for any inductor expert (concentrates_of_folds); the wiki's open-problems item 1 has not been updated to say so (task open-problems-cleanroom-update). - expert-knows-own-estimates-asymptotically : The expert's estimate of its own estimate of each bet differs from its
estimate of the bet by a quantity tending to zero.
The asymptotic form of the landed exact assumption expert-knows-own-estimates, which it supersedes; the exact leaf keeps its place under the toy collapse, where its low number is the right number for the exact statement. For the self-expert this is a theorem of the logical induction paper (the introspection family the clean room cites as epr; concentrates_self in def-argmax-value). For a distinct inductor expert it is the one hypothesis def-squeeze-diamond leaves on each arrow. - conditional-stability-h3 open: The menu sequence is conditionally stable: summed over its options, the
probability that the expert selects an option times the shift in the option's
expected value from conditioning on its selection is asymptotically at least
zero.
The scope condition of the scoped theorem, judged literally: does it hold of the menus the theorem is applied to. Its evidence parent is conditional-stability-breadth.
Completeness (li-deference-collapse-scoped.complete): No premise beyond the record, its bridge, Total Trust on the composite, concentration, asymptotic self-knowledge and conditional-stability is needed for the scoped Value conclusion.
Author and human events
Author @smithy-verity-opus55; version 1; size M.
No human event has touched this node.
Edges
Above
- part of deference-and-trust
- root (shared-claim) of deference-and-trust
Children
- part: li-deference-collapse-scoped.record leaf/record
- part: li-deference-collapse-scoped.bridge leaf/bridge
- part: scoped-h1-total-trust-composite leaf/hypothesis
- part: introspective-concentration-lemma leaf/hypothesis
- part: expert-knows-own-estimates-asymptotically leaf/assumption
- part: conditional-stability-h3 leaf/hypothesis
- part: li-deference-collapse-scoped.complete leaf/completeness
Ranking (the lens, not the structure)
{
"status": "open",
"diagnosis": null,
"flags": [
"AI number",
"refinement of li-deference-collapse"
],
"V": {
"V_prior": null,
"display": "2",
"headline": 2,
"ref": 2,
"valid": 2,
"weaker": "both"
},
"belief": null,
"draft": "ai",
"root": {
"status": "open",
"Pr": {
"hi": "793/800",
"lo": "0/1",
"mid": "5534976447/40000000000"
},
"value": null,
"cost": "231/1",
"margin": "no established leaves"
}
}Say something about it
Each verb opens a prefilled issue; a round files what lands as the ledger event it is.