authorship-and-transparency
frame · drafted
Question
When is a verdict the principal's own program's and a channel's influence the declared reference on the declared inputs, and does the thick definition of legitimacy satisfy the thin specification?
Owners: @anson (GitHub A-M-Berns), @abram (GitHub abramdemski) · owner has not confirmed (drafted from the wiki)
Path: the-program › legitimacy › authorship-and-transparency
Roots (the frame's goals)
plan
The public repository defines legitimacy at two levels, a thin specification (correct, sufficient, value) on a finite setting with her prior, model, actual process and a baseline, and a thick definition (Integrity, authorship, Robust Openness, transparency), and proves that the thick conditions satisfy the thin specification.
shared-claim
Value for every finite decision problem under a correct credence implies that the reference is a garbling of the later record, so sufficient is necessary for value.
The value lost to defects is at most the decision problem's range times the prior mass where the actual record departs from the model's plus the same range times the mass where the actual verdict departs from her program's, for deterministic records.
Effective authority is modelled with a queue that cannot be bypassed, caused and flood-caused shortfalls, a comprehension shortfall, approval discounts and a revocation default, with sincerity conditions on which verdicts count.
The informal argument of Corrigible Normative Learning, as the thin-legitimacy round states it, is backed by that round's theorem.
The thick and thin legitimacy theory, stated on finite settings with a parameter baseline, can be improved until it reflects legitimacy between a human principal and an AI over real channels well.
In the public theory internal legitimacy is factorization through the normative reasons: authorship requires at each step a grounding selection from the reason trace's pre-state prefix whose license contains the verdict entered.
Authorship of a future evaluation is exclusive principal binding together with reason mediation: the committed payload factors through the declared reason view at a fixed principal policy.
The effective-authority layer, in which authority is exercised through a protocol that cannot be bypassed, with caused and flood-caused shortfalls, a comprehension shortfall, approval discounts and a revocation default, and sincerity conditions on which verdicts count, can be improved until it reflects a principal whose later judgment counts as her own because the agent asks through her protocol, keeps her attention budget and takes no authority back from her well.
What it holds
Nodes this frame is about that are not goals of the ranking: no status, value or rank here.
formal results (15)
- authorship.mediation-by-reexecution: Clause 2 of item 87 only, not the bill. For any frame over the ecosystem's logs, any audited class of activated continuations sharing the mandated program, and any policy, the committed payload factor
- thin.correct-of-thick: Theorem 1 of the thin specification. Inhabited by `ThinLegitimacy.Witness.thick_inhabited` (two worlds, the revealing process its own model).
- thin.sufficient-of-integrity-openness: Theorem 2. Sufficient is Blackwell's order by definition; this lemma places the earlier record in the reference.
- thin.value-of-correct-sufficient: Theorem 3, the easy direction of Blackwell's theorem. Inhabited by `Witness.thin_inhabited`.
- thin.converse-fails-on-correct: The revealing process on two worlds with the point mass at one record and the prior at the other has value against the blank reference for every decision problem and is not correct. A witness claim; i
- thin.reflection-inherited-value: Theorem 5. From `totalTrust_of_correct` and the inherited `value_witness_iff_totalTrust`; uses authorship, transparency and Integrity, not openness (`sufficient_reflection_of_integrity`, `garbling_tri
- thin.preservation: Theorem 6 in the conditional form (`condReflection_iff_coherent`). The marginal martingale is too weak: `thin.marginal-martingale-too-weak`.
- thin.preservation-pair: The pair form of theorem 6: the actual later record beside the earlier one carries the posterior of the later record alone. Inhabited by `Witness.thick_inhabited` (earlier record blank).
- thin.marginal-martingale-too-weak: A witness claim; its own inhabitant. The inherited `AntiExpert` is the same phenomenon.
- thin.value-loss-le-defects: The approximate form (A6), for deterministic records, from `TransparentChannel.abs_expect_sub_le_width` on `Act × W`; `value_approx` composes it with theorem 3 for the model. Inhabited at `a = m`, `V
- thin.covert-blocking-correct-iff: Part 5 of the follow-up, correcting PR #116's finding that covert blocking leaves her correct — true only when silence has the same likelihood in every world. `correct_silence_iff` is the posterior fo
- transparent.posterior-weight-eq: The transparent-channel round's statement, relied on by Part A (`ThinLegitimacy.post_eq_of_realizes` is its kernel reading) and Part B (`content_fact_eq`). Inhabited by `TransparentChannel.Witness.pic
- transparent.blind-payload: Relied on by Part A's reading of the landed definitions. Inhabited by `TransparentChannel.Witness.picker`'s honest class with the blind pair class of `ReasonMediatedAuthorship`'s witnesses.
- legitimacy.payload-of-view: Authorship composed with transparency, pathwise; the τ = 0 case of the approximate bound. Inhabited by `Legitimacy.Witness.segment`.
- thin-value-of-thick: Coherence, authorship and weak transparency give correct; Integrity and openness give sufficient; and correct and sufficient give value for every finite decision problem, so the thick conditions satis
items (3)
- item-109: The thin preservation theorem holds in asymptotic form for a logical-inductor principal whose self-trust is asymptotic and whose program is not a posterior: with the inductor's conditional self-trust — item 109
- item-110: The baseline, what would have reached her without anyone's interested interference, has a definition from which the eight-world model's cases are derived rather than declared. — item 110
- item-108: The thin properties have a form for valuations with no outside fact to be correct about, and the thick conditions deliver them. — item 108
Statement
Coherence, authorship and transparency give the thin property correct, Integrity and openness give sufficient, the two give value, a lie through the declared channel fails transparency while a sincere error counts, and the composition of verdict with view factors through the declared inputs under legitimacy.
Source
wiki/Legitimacy.md, Legitimacy has two levels, Scope and relationality and the classification table; projects/deference/CLAIMS.md (thin., sincerity., transparent., authorship., legitimacy.payload-of-view).
Notes
Drafted from the wiki per ruling M3 (prompts/2026-10-22-map-as-canonical); the owners have not confirmed. Held (holds, ruling G2 of 2026-10-23), not roots: the registered thin-legitimacy, sincerity, transparency and authorship results as formal results; its roots are the public two-level theory and the claims the dissolution of legitimacy-and-reasons placed here; the three work orders that dissolution placed here (items 108, 109, 110) are held since the closing pass of 2026-10-24. No wiki page narrates this programme alone; wiki/Legitimacy.md narrates the line.
Author and human events
Author @smithy-verity; version 1; size M.
No human event has touched this node.
Edges
Above
- part of legitimacy
- root (plan) of legitimacy
Children
- part: public-legitimacy-theory and
- part: thin-converse-open leaf/question
- part: thin-value-loss-bound leaf/step
- part: effective-authority-layer and
- part: cnl-post leaf/step
- part: public-legitimacy-improvable or
- part: public-internal-legitimacy-is-reasons leaf/step
- part: reason-mediated-authorship leaf/step
- part: effective-authority-layer.improvable or
Ranking (the lens, not the structure)
{
"status": "open",
"diagnosis": null,
"flags": [
"owner has not confirmed (drafted from the wiki)"
],
"V": {
"V_prior": null,
"display": "2",
"headline": 2,
"ref": 2,
"valid": 2,
"weaker": "both"
},
"belief": null,
"draft": "ai",
"root": {
"status": "open",
"Pr": {
"hi": "1/8",
"lo": "819/80000",
"mid": "32661/320000"
},
"value": {
"hi": "1/1",
"lo": "-1/1",
"mid": "32661/320000"
},
"cost": "198/1",
"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.