Topics — What a work is about. Nucleation holds whatever reading the owner files here, whether or not an Anthology of the SOTA topic could also hold it; instruction for ML practice belongs to the anthology (ADR-048). The words are seeded from the clusters the reading-time triage of 2026-09-25 found, and grow by decision.
Epistemology — knowledge, belief and evidence — justification and rational credence, the value of knowledge, testimony, trust and expertise, and social and inductive-risk questions about evidence (group: philosophy).
14 of 191 CASE documents. Back to the full index.
| # | Title | Summary | Status |
|---|---|---|---|
| CASE-082 | The sufficient posterior without its history: a Bayesian agent that keeps only its current posterior updates as well as one that keeps everything, and cannot say how it got there | A6’s fourth case, which A6 called “probably the most important counterexample in the set”. It grounds the redundancy half of CLAIM-300 and one of CLAIM-299’s separations. A6’s reserve case, revising the model class, is where it stops holding, and CASE-087 is that case. | Active |
| CASE-084 | The perfectly accountable false theory: a community whose record traces every commitment and whose vocabulary leaves out the right explanation | A6’s third case. Accountability without correctness, one of CLAIM-299’s three separations, and A6’s “direct companion to CLAIM-181 and CLAIM-190”. | Active |
| CASE-086 | The retracted finding and its descendants: an experiment fails to replicate years after three conclusions were built on it | A7’s sixth case, filed at the owner’s U7 as an illustration and not a test (“one or two of those might be handy to keep on hand too”). Tracing descendants of a defeated premise, which CLAIM-295 says a record of dependencies supports. | Active |
| CASE-089 | The compensating evidence problem: two records with the same support graph in which discounting one support leaves odds of 20:1 or of 1.1:1 | A6’s second case. It grounds CLAIM-293: the same edges E₁ → H ← E₂, different Bayes factors for E₂, different revisions when E₁ fails. | Active |
| CASE-090 | The rehabilitated hypothesis: an objection that displaced a hypothesis turns out to rest on a faulty measurement, and the hypothesis can be reopened because its defeat was recorded | A7’s seventh case, the companion to CASE-086 and filed with it. Reopening an alternative when its defeater is defeated. A12’s “defeat of a defeater” is the same case and is held here, as A12 itself proposed. | Active |
| CASE-091 | A memory ablation for revision: synthetic inquiries with known ground truth, run under six kinds of persistent memory, with information and token budget controlled separately | A6’s fifth case, refined from A3, A4 §4 and A5. The design QUESTION-108 asks for. Not run. | Active |
| CASE-093 | Same marginal, different updates: two joint distributions over a hypothesis and a condition that agree on the hypothesis and disagree on how to revise it | A6’s first case, made exact at A8. It grounds CLAIM-294. | Active |
| CASE-131 | The graph that asserted an inference nobody made: filing one claim as supporting another made the record’s link fixer write that the other rested on it, and the record answered by adding a relation | The record’s own history, as A18 of the exchange of 2026-10-22 read it at the owner’s prompting (U18): “the pre-existing representation didn’t merely omit information; it positively misrepresented an inference”. It is a move the grammar could not hold, caught and answered by amending the grammar. Not shown: that such moves are always caught. | Active |
| CASE-138 v2 | Challenger to Columbia: NASA corrected its launch decisions and its safety organization after 1986, and the Columbia board found in 2003 that related organizational failure mechanisms had recurred, while its lessons-learned system was little used in between | The owner’s case at U25 of the exchange of 2026-10-22: NASA’s mechanisms for documenting and learning from mistakes, with Challenger as the classic example. The record already reads the Columbia board’s account (THEORY-140, from LIT-734). This case files the history around it: a warning whose standing changed as it moved, a correction made in the rules, and a capacity not maintained. Not shown: why the correction decayed, which THEORY-140 explains and this case only displays. | Active |
| CASE-154 | One representation, two tasks, one bounded observer: with X₁, X₂ independent fair bits and an observer that reads one coordinate and cannot compute XOR, the raw pair U = (X₁, X₂) gives error 50% on Q_A = X₁ ⊕ X₂ and 0% on Q_B = X₁, and the compressed V = X₁ ⊕ X₂ gives 0% and 50%; a legibility regime can enable one perspective’s distinction while disabling another’s, with no task-independent sense in which it is better | From the exchange of 2026-10-24 (workbench entry of 2026-10-24, U22), after the owner said “let’s work on those connections”: a deduction from CLAIM-199’s XOR example, checked when filed. It ties bounded legibility to competing concerns: a representation that serves one task can destroy another’s for the same observer, so a failure of one perspective under a legibility regime may be a resource limit and not an exclusion. | Active |
| CASE-156 v2 | A legibility filter as the ν of the meaning–authority–legibility model: in a safe world where 15% of raw cues are alarms, an admission rule P(admit | y, r) = σ(c + χry) with c = 0 and χr = 2 admits 88.1% of alarms and 11.9% of reassurances, so 56.6% of the admitted record is alarms with the world and the sensors unchanged; at symmetry the mean admitted evidence is χσ′(c)r to first order, so ν = η_m χ σ′(c); and r_t = λᵗr₀ + ε Σ λ^(t−1−j) a_j is an exponentially weighted history of authority | From the exchange of 2026-10-24 (workbench entry of 2026-10-24, U18–U19). The owner: “i think we already have a three variable system somewhere, where the third component is described as legibility”, then “flesh this out further”. A microfoundation for one of CASE-124’s abstract couplings: selective admission of evidence, checked when filed. It gives r an observational semantics and reads r as the memory of past authority. Selection-aware inference is the control that the pathology has to beat. | Active |
| CASE-183 | The anchoring measures: inbound references per code and turns per filed entry, counted in the record’s own grammar | The two measures the nucleating-argument line proposes for its claims, filed as a design so that an argument can start from them. Inbound references per code would tell an anchor from a structure that only persists (QUESTION-038). Turns per filed entry would be set against the drift the workbench records (CLAIM-186). Both are counted in codes, relations and statuses. Neither has been computed. | Active |
| CASE-186 v2 | A proposed study: an organization in a world whose hidden regime reverses and then leaves its hypothesis family, with evidence generation, admission and scoring separately lesionable, its diagnostic machinery maintained by its own reciprocal production, from outside, or not at all, and finite fuel shared among operation, maintenance and diagnosis, to find whether a second correction after injury to the diagnostic and copying machinery separates arrangements that correct the first failure alike | The CASE A22 proposed and A23 wrote out at the owner’s “flesh out the proposed CASE”, in the exchange “Analyze nucleation concepts” (2026-10-31), under A23’s title “Maintained Inquiry Under Interpretive Capture, Productive Damage, and Environmental Novelty”. Its admission × scoring arms repeat CASE-162’s, and its double-lesion prediction is CASE-161’s finding. What is new is the primary endpoint, a second correction after the machinery is injured; an environment that leaves the hypothesis family; and external restoration that is paid for. Not run. The analytic error rates hold only where admission is protected. | Active |
| CASE-188 | A proposed study: two agents keep the same long-term commitment and decide alike under ordinary conditions, one protecting it from irrelevant fluctuation while letting specified defeaters revise it, the other growing ever more efficient at enforcing it while evidence against it loses its way in; varying the environment, the legitimacy of the original commitment and what the evaluative mechanisms can see, to tell corrigible resolve from entrenchment without building the answer into the diagnostic or the score | Proposed by the owner at review of the filing of the exchange “Analyze nucleation concepts” (2026-10-31), as a successor to CASE-186 and kept separate from it so that study stays interpretable. It gives CLAIM-344’s distinction between resolve and mere persistence a counterfactual test. Not run; every parameter is left to a preregistration. | Active |