audit · retractions

Retractions

Every claim this programme has made and then withdrawn, with what killed it. Nothing here is deleted from where it originally stood — each is struck through in place, so a reader arriving at the original text sees the correction rather than a clean document that never made the mistake.

This file exists because the site quotes a count, and a count nobody can audit is worse than no count.

Standing tally: 10 claims withdrawn, 3 experiments void, 0 findings established.


Axioms and definitions

#ClaimKilled byWhere
1A3 — no bounded message closes an arbitrary prior gapA 113-token lookup table reaching F* = 1. Restated for held-out probes under bounded cost.PRINCIPIA §4
2D_floor is "the irreducible divergence caused by the parties' own stochasticity", and belongs inside the definition of F*Two perfectly aligned stochastic agents have identical true distributions, so their true JSD is exactly zero. The floor is estimator bias, and a fidelity that changes when you sample more is not well-posed.definitions §3
3η = F*/C is "the quantity engineering should optimize"A ratio with a signed numerator is not an ordering. At F* = −1, a 100-token antinoophor scores −10.00 and an 800-token one −1.25, so the message that spends eight times as much to do the same damage ranks higher. Replaced by V_λ = F* − λC.definitions §4.3
4Falsification criterion 2 — "Φ ≈ 0 means the pathology does not exist"Bias and resolution are independent. A party predicting 0.70 on every probe and averaging 0.70 has Φ = 0 and no ability to say which probes it got wrong; it is maximally pathological and the criterion scored it as our refutation.PRINCIPIA §7

Laws

#ClaimKilled byWhere
5L2 headline formTautological as stated.laws.md
6L4 — fidelity is multiplicative along a chainIll-typed: it multiplied fractions of different prior gaps, and two antinoophors composed to a positive product. Measured: hops of −0.629 and −1.000 multiply to +0.629. Restated as L4a/L4b/L4c.laws.md
7L6 is "the field's first engineering prescription"I-PASS, deployed and outcome-measured since 2014 across nine programmes and 10 740 admissions.laws.md · prior-art §5
8L5 is "our sharpest conjecture"The human half is Carpenter et al. (2013) and Deslauriers et al. (2019). Status stays conjectured — a prior is not a test — but the framing was ours to lose.laws.md · prior-art §4

Novelty claims

#ClaimKilled byWhere
9Φ is "the part we have not found elsewhere", and is what "everyone had felt, nobody had weighed"Keysar & Henly (2002), Newton (1990), Chang et al. (2010), Endsley (2020). It was weighed in 1990, and their instruments are in places better than ours.PRINCIPIA §1, §5 · prior-art §1
10Knowledge distillation "measures success as task accuracy"Stanton et al. (2021) define fidelity separately from generalization and show accuracy does not imply it — E-001's construct failure, from a NeurIPS abstract, five years early.PRINCIPIA §2 · prior-art §3

Void experiments

idwhat it was going to measurewhy it is void
E-001fluency vs contrastiveness on fidelity and ΦSender refused to compose. Reopened as a construct critique: the headline quantity rewards mimicry, and 62% of its effect sat on four probes where the sender was wrong.
E-001bthe same, factorially, with a cost-parity gateThe gate failed on the composed messages: fluent briefs cost 1.5× terse ones under an identical budget instruction. Style and length are entangled in the generator. Also DEFECT-001 — the analysis path had never been executed and would have crashed after 30 hours.
E-002Φ for the first time, elicited per probe330 of 330 elicitations returned "yes, we will agree" — the instrument has a default answer, because it ported Keysar & Henly's granularity and not their forced-choice structure. And the transfer was perfect (0 of 33 probes diverged), so there was nothing for a belief to be wrong about.

What this list is for

Two things, and neither is penance.

A refuted claim is a measurement. Knowing that η inverts on antinoophors, or that a cost-parity gate cannot be met by instructing a budget, is knowledge the programme did not have before, and it was purchased at a price. A file containing only survivors would tell a flattering lie about how the field got here, and would make the same mistakes available to the next person.

And it is an audit surface. The count on the front page is checkable against this table, the table is checkable against the struck-through text, and the struck-through text is checkable against git. A programme whose subject is the gap between confidence and evidence should be the easiest one in the world to catch overstating itself.


This document is licensed CC BY 4.0.

All journal entries · Noophorics · Repository