Case kinds — What kind of thing a CASE is, beside how much of it is fact (standing). The two are orthogonal (ADR-042).
Executed study — a study the record’s collaborators ran in session, whose results the record itself produced, with its provenance and limits stated.
10 of 191 CASE documents. Back to the full index.
| # | Title | Summary | Status |
|---|---|---|---|
| CASE-017 | E1: synthetic three-state stochastic computations (C2–C3) | Run at U23 (“run it”): hand-chosen row-stochastic matrices on three abstract states, seed 20261008, “not fitted to text, pretrained models, or human measurements”. Executed; results are exact properties of the chosen matrices. None reached a later draft. | Active |
| CASE-020 | E3: the controlled inference harness, dry run only (C5) | Built at U25: four frame conditions (stable complicit, neutral, moralizing; shifting), 12 reset chains each, 5 generations, 240 calls, gpt-4.1-mini at temperature 0.7, seed 1729. Only a dry run was executed: “No model-generated observations were recorded.” The design is the manuscript’s Case II. | Active |
| CASE-030 | E2: assistant-authored reconstruction chains with content-word overlap (C4) | Run at U24 (“do it”): three four-generation chains written by the assistant from a modernized Villon seed, scored by content-word Jaccard against the seed. The texts are authored with their intended trajectories; the numbers are exact properties of those texts. | Active |
| CASE-032 | E4: within-session chains, stable complicit against a shift to moral instruction | Run after the owner’s U26 (“lol you ARE a text generation action. just delegate subtasks”): two three-generation chains written by the assistant in the same context. “The measured lexical similarities are near zero for both final texts”; one keeps complicity, the other becomes advice. | Active |
| CASE-157 v2 | Boosting the vote: in the owner’s 2014 model, voters of fixed accuracy drawn from Beta(2, 2) vote in a weight-averaged election, and every voter’s weight is multiplied by 1 + f when their own vote was right and 1 − f when not, with a fraction replaced each round at weight 1; with full-information feedback the weighted result rises from 0.50 toward 0.98 as f grows, and churn only costs in a world that never changes; the model escapes the hive mind’s absorbing states because every voter is scored, heeded or not, and it assumes away the capture its author predicted | Filed at the owner’s direction, 2026-10-24, after the owner suggested taking inspiration from the experiments in dmarx/BoostTheVote. A thought-experiment written in R in 2014, read at commit b5bca8f and re-run here through a Python port (scratch/formal/boost_the_vote.py). It is the full-information counterpart of the replicator construction. What it shows, what it assumes, and the three-axis extension it suggests are recorded, and the extension was run on 2026-10-25. | Active |
| CASE-161 v2 | Reciprocally produced, externally reset and feedback-only governing rules in a 32-member population: with ordinary trajectories identical, reciprocal production repairs the loss of one rule and not of both and fails after producer outages of 36 updates for audit and 40 for admission, after an audit-rule pulse the next audit integrity is 1.00, 0.18 and 0.00 in the three arrangements, and a self-dependent production graph does not regenerate a zeroed rule | Versions 1.6 and 1.7 of the Perspective Population Lab, dmarx/PerspectiveLab@7e2edc2. Two governing rules, the integrity of audit and of evidence admission, are kept up in three ways in the same 32-member population. Ordinary operation cannot tell the three apart, and loss, repair and producer interventions do. It carries CASE-126’s second construction to a stochastic population with a different production law: here reciprocal production repairs a single loss, where the case’s closed pair collapses. | Active |
| CASE-162 v2 | 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 | Versions 1.1 to 1.3 of the Perspective Population Lab, now held as dmarx/PerspectiveLab@7e2edc2. v1.2 is a deterministic four-variable model whose analytic fold and whose trapping of a declared history come apart. v1.3 is a separately coded event-level population in which the filtering of CASE-156 runs on sampled cues, and a known-propensity evaluator undoes it while admission stays positive. v1.1 adds capture and selective audits to CASE-158’s design, and the router still wins. The sign reversal is CASE-156’s; what is new is two running realizations, with their code. | Active |
| CASE-163 | Identifying how governing rules are maintained from diagnostic probes: a predictor fitted under reciprocal production fails under self-dependent production with histories matched, an audit pulse with a short producer interruption identifies which of four known mechanisms applies and predicts withheld interventions, and fitting decay and threshold as well helps after the parameters shift and hurts at nominal ones, within a known catalog and with every probe run on a counterfactual reset | Versions 2.0 to 2.2 of the Perspective Population Lab, dmarx/PerspectiveLab@9f82bfd, on the maintained population of CASE-161. v2.0 shows that a response model does not transport across a change of production law that ordinary histories do not reveal. v2.1 and v2.2 infer the mechanism, and then its parameters, from probes, and predict interventions never probed. Everything is relative to a catalog of four known mechanisms whose equations the forecaster is given, and every probe is a separate branch from the same starting state. | Proposed |
| CASE-178 v3 | How the record’s grammar grows: after its scaffold, this record filed documents on all 32 days, added vocabulary terms on 18, fields or relations to schemes on 8, whole vocabularies on 3, schemes on 1 and a remote project on 1, so the owner’s ordering of levels holds by frequency; but each refinement came in the same pull request as the documents that needed it and, nine times in ten, an ADR, so the levels are rarer events of one loop of work, not slower processes of their own, and the grammar’s rates fell while filing rose | Run on 2026-10-27 at the owner’s direction, after CASE-180: “some of the most important rates are when and how schema refinement happens”, with the ordering documents, vocabularies, scheme fields, schemes, projects. scratch/formal/schema_rates.py reads every change to luria.yaml in this record, in luria’s own and in strata-g’s. The ordering holds by days with an event. How the refinement happens: on demand, by decision, and nested, the rarer level arriving with the commoner ones in one pull request. | Active |
| CASE-180 v2 | The record’s own rates: over its first fifty days, measured from git, an entry’s prose is revised about once in 22 days once drafting is over, its status once in 128, an authored relation gained or lost once in 132 and an authored relation retracted once in about 600, while no relation type has been redeclared; the layers’ event rates are ordered as the exchange guessed, but only after the first day, when all of them change together, and the grammar is slow to revise without being slow to grow | Run on 2026-10-27 at the owner’s request (“run your measurements against the record”), for the rate reading A12 and A13 gave luria’s layers and that CLAIM-476 left untested. Every non-merge commit to this record, read by scratch/formal/record_rates.py at commit a357048. The ordering (prose, then standing, then relations, then the grammar’s types) holds in aggregate rates, by one to two orders of magnitude. It does not hold in the first day after filing, or for every scheme, and the machine’s validation runs at the rate of the edits it follows, not on a scale of its own. | Active |