A decision record is a choice among alternatives at a point in time. Write
one when you rejected an alternative, chose a constraint, or made something a
future edit could silently violate. Run luria new adr; it assigns the
identity and scaffolds from _template.md.
Values that decisions cite live in
design-principles.md instead — also generated, from
the fragments in ../principles/. The split, and why these are separate files,
is LU-ADR-001.
A decision whose choice changes is superseded by adding a decision and flipping the old one’s status — not by rewriting its body. A record you can quietly rewrite can’t be trusted about what you used to think.
That is a rule about silence, not about editing. A decision whose choice stands
but whose reasoning was wrong is corrected in place, with a version bump and
a history: entry saying what the previous version claimed. Nothing here is
frozen; it is only un-silently revisable. Luria’s own record carries worked
examples of both: LU-ADR-019.
By tag
The record (68) — what the schemes hold, and the rules between them: 001 · 002 · 003 · 004 · 005 · 006 · 007 · 008 · 009 · 010 · 011 · 012 · 013 · 014 · 015 · 016 · 017 · 018 · 019 · 020 · 021 · 022 · 023 · 024 · 025 · 026 · 027 · 028 · 029 · 030 · 031 · 032 · 033 · 034 · 035 · 036 · 037 · 038 · 039 · 040 · 041 · 042 · 043 · 044 · 045 · 046 · 047 · 048 · 049 · 050 · 051 · 052 · 053 · 055 · 056 · 057 · 058 · 059 · 060 · 061 · 062 · 063 · 064 · 065 · 066 · 067 · 068 · 069
Taxonomy (0) — the topic vocabulary and what enforces it.
Mechanism (4) — identifiers, generation, the lint: 047 · 049 · 051 · 057
Migration (0) — what moves between this record and the Anthology of the SOTA, and why.
By status: Active (65) · Proposed (1) · Deferred (1) · Superseded (1) · Rejected (1)
Chronological
What the status column means in this scheme — the words are luria’s, the meanings are this project’s.
| Status | Means | |
|---|---|---|
Active | Active | In force — the current answer, and what a citation should normally point at |
Proposed | Proposed | Not in force yet — an open question, so citing it as settled is the thing the report catches |
Deferred | Deferred | Not in force and not being worked on; the question is real and the answer is waiting on something |
Superseded | Superseded | No longer in force because something replaced it; the successor is named in the field, not the prose |
Rejected | Rejected | No longer in force and nothing replaced it — kept because a rejection is worth being able to point at |
| # | Title | Summary | Status |
|---|---|---|---|
| ADR-001 v3 | Nucleation holds what the anthology’s topics cannot: LIT, NOTE and THEORY, with a seeded closed vocabulary | The Anthology of the SOTA is scoped by its topic vocabulary, and its own rule is to add a topic rather than decline a document. But some reading is not about machine-learning practice at all: quantum foundations, network science, complex systems, mathematics, neuroscience, law. Nucleation is the catchall record for it. It ports the anthology’s LIT, NOTE and THEORY schemes unchanged, so a work can move between the two without re-shaping. It leaves out the SOTA practice scheme, and starts with a closed topic vocabulary seeded from the out-of-scope clusters in the anthology’s reading-time triage. | Active |
| ADR-002 v2 | published: is the exact date of first appearance where a source gives one | A LIT’s published: records the day the work first appeared — an arXiv v1 submission date, a court filing date, a post’s date — and falls back to the first of the month only when no source gives a day. The month-only convention inherited from the anthology threw away a fact every source already states. | Active |
| ADR-003 | Ten more topics, seven of them grouped as philosophy | The topic vocabulary gains information-theory, social-science and linguistics, and seven philosophy words — metaphysics, mereology, agency, identity, ethics, consciousness, cognition — which a philosophy tag group labels as a set. The group is a label (require: any), not a rule: a work may carry any number of them, alongside any other topic. | Active |
| ADR-004 | philosophy-of-science joins the philosophy group | An eighth philosophy word, philosophy-of-science, for what science is and what its theories say about the world: realism and structural realism, explanation, causation and evidence, the interpretation of physical theories. It joins the philosophy tag group ADR-003 declared. | Active |
| ADR-005 v2 | anthology-candidate: a curation flag in the tag vocabulary | A work that may belong in the Anthology of the SOTA carries the tag anthology-candidate, after its topics and never first. The tag is a curation flag, not a subject, and a flags tag group labels it as one. Its generated tag page is the report of pending transfers. The work stays here until somebody decides to move it. | Active |
| ADR-006 | epistemology and logic join the philosophy group | Two philosophy words join the topics and the philosophy tag group. epistemology covers knowledge, belief and evidence. logic covers valid inference and its formal theories. The 2026-09-26 sweep of the papers-feed tracker for philosophy found works that neither word’s nearest neighbour describes. | Active |
| ADR-007 | philosophy-of-mathematics and religion join the topics | Two words join the topic vocabulary. philosophy-of-mathematics covers what mathematics is about and how we know it, and joins the philosophy group. religion covers religion as a subject, and stands outside the group. The pass over the metaphysics tag found works that mathematics and the existing philosophy words describe only approximately. | Active |
| ADR-008 | No reader sends an email address to any service | An agent reading for this record never supplies an email address to an outside service: not the owner’s, not anyone’s. Where an API asks for one, the field is left out, or given noreply@anthropic.com. The rule is in the reading brief every reader follows, because on 2026-09-26 a reader sent the owner’s address to a lookup service. | Active |
| ADR-009 | philosophy-of-language joins the topics | philosophy-of-language joins the topic vocabulary and the philosophy group. It covers meaning, content and reference as philosophy poses them. The pass over the philosophy-of-science tag found a paper on meaning holism tagged philosophy of science for want of the right word, and four more held works share the gap. | Active |
| ADR-010 | representation-learning and learning-theory join the topics | Two words join the topic vocabulary. representation-learning covers how learned systems come to represent their data. learning-theory covers why learning generalizes. The owner asked the record to file a set of papers on self-supervised learning, information bottlenecks, description length and generalization bounds. Nothing in the vocabulary could say what they are about. | Active |
| ADR-011 | information-retrieval joins the topics | information-retrieval joins the topic vocabulary. It covers finding what is relevant to a need: representation, relevance, ranking, feedback, and the models of retrieval. The owner asked the record to file van Rijsbergen’s The Geometry of Information Retrieval. Its seed found no word for the book’s subject. | Active |
| ADR-012 | contextuality joins the topics | contextuality joins the topic vocabulary. It covers whether outcomes can be explained without reference to the context of measurement, in physics and outside it. The owner asked for the tag, and for the record’s two threads on it to be brought together in THEORY documents. Until now the word lived only inside the quantum-foundations blurb. | Active |
| ADR-013 v2 | A work may be held in both records when each reads it for its own question | ADR-001 said nothing is filed in both nucleation and the anthology. That rule is narrowed: a work may hold a LIT in each record when each record reads it for a question of its own. The two entries name each other, and neither copies the other’s reading. The owner asked for it so that the Lattice Representation Hypothesis, held in the anthology for concept geometry, could be re-read here against the geometry of information retrieval. | Active |
| ADR-014 | emotion-and-affect joins the topics | emotion-and-affect joins the topic vocabulary, in the philosophy group beside consciousness and cognition. It covers emotions, feelings and affect: what they are, their bodily and neural basis, their role in decision and behaviour, and whether they ground consciousness. The owner asked for it when Antonio Damasio’s work came into the record, whose subject no existing word could say. Rejected: a separate word for homeostasis and interoception, which only the Damasio works would hold yet. | Active |
| ADR-015 | self-and-personhood joins the topics, and identity becomes individuation | self-and-personhood joins the topic vocabulary in the philosophy group, and identity narrows to the identity of things of any kind and is renamed individuation to say so. The new word holds what a self or a person is and what makes one persist: personal identity, the narrative, minimal and bodily self, self-models, agency over one’s own life, and whether machines or groups can be selves. Fifteen existing works take it, six as their first tag. Rejected: keeping personal identity inside identity, where it was mixed with quasi-set theory, dualities and biological individuality. | Active |
| ADR-016 | phenomenology joins the topics | phenomenology joins the topic vocabulary in the philosophy group, for the first-person description of experience as a method: the philosophical tradition, its use in the sciences of mind, and the theories that take their data from it. Six works take it, none as its first tag. Rejected: leaving method unsaid because the subjects are held, and stretching consciousness or self-and-personhood to cover it. | Active |
| ADR-017 | social-ontology joins the topics | social-ontology joins the topic vocabulary in the philosophy group, for what social facts, groups, institutions and social kinds are and what makes them exist. Six existing works take it, List & Pettit’s Group Agency first. Rejected: holding the subject under society-and-governance or social-science, and a narrower word for group agency alone. | Active |
| ADR-018 | moral-psychology joins the topics | moral-psychology joins the topic vocabulary in the philosophy group, for morality as a natural phenomenon: moral judgement, emotion and norms, and their evolution, development and variation across cultures. Four existing works take it, three first. Rejected: widening ethics, which is normative, and a narrower word for the evolution of cooperation. | Active |
| ADR-019 | game-theory joins the topics | game-theory joins the topic vocabulary, outside the philosophy group, for strategic interaction and its equilibria in biology, economics and philosophy. Eight works take it, none first. Proposed independently by two filing agents in the moral-psychology batch. Rejected: stretching probabilistic-modeling or leaning on mathematics alone. | Active |
| ADR-020 | thermodynamics joins the topics | thermodynamics joins the topic vocabulary, beside natural-sciences, for the second law and its extensions read for themselves: entropy production, nonequilibrium steady states, dissipative structures, fluctuation theorems, and the thermodynamics of computation and of life. All three agents filing the dissipative-structures batch proposed it independently. Twenty-seven works and two THEORYs take it. Rejected: leaving the subject under natural-sciences, and a narrower word for nonequilibrium thermodynamics alone. | Active |
| ADR-021 | psychopathology-and-treatment joins the topics | psychopathology-and-treatment joins the topic vocabulary for what a mental disorder is and how psychological treatment changes it: medical models of illness, the philosophy of psychiatry, transdiagnostic accounts such as experiential avoidance and psychological flexibility, and psychotherapies and their mechanisms of change. The owner’s batch of fifteen works had no word to file under. Fifteen notes take it. Rejected: leaving it to social-science and natural-sciences, and two narrower words for psychiatry and for psychotherapy. | Active |
| ADR-022 | rivals joins the LIT and THEORY relations | A symmetric rivals relation joins both the LIT and the THEORY scheme: two accounts of the same phenomenon that cannot both be right, where neither has tested or corrected the other. It is the anthology’s relation, carried over unchanged in meaning. First use: Barrett’s constructed emotion against appraisal theory. Rejected: holding rivalries in prose only, and stretching compared_against or corrects to cover them. | Active |
| ADR-023 | motivation joins the topics | motivation joins the topic vocabulary for why organisms act and persist: intrinsic and extrinsic motivation, needs, goals and values, the internalization of regulation, and autonomous against controlled motives. It is the empirical counterpart of the philosophy of autonomy held under agency. Self-determination theory was the occasion. Rejected: filing it under agency and social-science alone. | Active |
| ADR-024 | Agent, individual, self and person; four unities kept apart | The owner asked that agent, individual, self and person not be synonyms, and that behavioural integration, phenomenal unity, narrative unity and normative self-governance be represented separately. self-and-personhood is retired into self and personhood, beside agency and individuation. Four new words name the unities: behavioral-integration, phenomenal-unity, narrative-unity and self-governance. Forty-two documents were retagged, and twenty-one more took a unity word. Rejected: subtopics under one word, and a separate vocabulary for unities. | Active |
| ADR-025 | Six topics for gaps the agency and self batches found | well-being, learning-and-conditioning, psychometrics, philosophy-of-biology, embodied-cognition and free-will join the topics. Each was a gap that filing agents reported from more than one batch, and each now has several works that had been placed under the nearest wrong word. Rejected: waiting for each gap to be named twice in separate decisions, and broadening cognition, learning-theory or natural-sciences to cover them. | Active |
| ADR-026 | model-comparison and information-geometry join the topics | model-comparison (evidence, Bayes factors, Savage–Dickey, Occam factors, Bayesian model reduction, structure learning, singular free energy) and information-geometry (Fisher metric, natural gradient, curvature and its block structure, Fisher and Hessian spectra) join the topics. The owner’s batch of readings on model comparison and Fisher geometry is held here, under ADR-013, as one line of inquiry. Its machine-learning works carry the anthology-candidate flag. Rejected: leaving both subjects to probabilistic-modeling and learning-theory. | Active |
| ADR-027 | loss-landscapes joins the topics | loss-landscapes (mode connectivity, barriers and permutation symmetry, re-basin alignment, star domains and connecting simplexes, local versus global landscape structure, and model merging) joins the topics. It holds the owner’s batch of 33 readings on mode connectivity and model merging, filed here under ADR-013 with the anthology-candidate flag. Rejected: stretching information-geometry or learning-theory to cover it. | Active |
| ADR-028 v2 | ecology joins the topics | ecology joins the topics. It covers interactions between organisms and with their environment: symbiosis and host–microbe partnerships, herbivory and trophic interactions, community structure, and ecosystem resilience and regime shifts. It enters with the owner’s coral batch. Rejected: holding these works under natural-sciences plus complex-systems, and adding chronobiology on the strength of one paper. | Active |
| ADR-029 | presupposes joins the THEORY relations | A presupposes relation, with its converse presupposed_by, joins the THEORY scheme. It means this account takes the other’s conclusion as a premise for a claim about something else. It is not in the chain of lines of explanation, so it needs no shared topic. First uses: Pettit’s corporate responsibility (THEORY-139), re-filed from extends, on List and Pettit’s group judgment (THEORY-127); and Scheffer and Carpenter’s evidence standard (THEORY-136) on Holling’s resilience (THEORY-132). | Active |
| ADR-030 | design-rationale joins the topics | design-rationale joins the topics. It covers recording the reasoning behind a design or a decision as it is made (issues, positions, arguments, options and criteria, as in IBIS, gIBIS and QOC) and the literature on why such capture systems are or are not used: who pays the cost of capture and who gets the benefit, premature formalization, and cognitive overhead. It enters with the owner’s batch of nine works on that literature. Rejected: filing them under society-and-governance, epistemology or cognition, and a broader human-computer-interaction word. | Active |
| ADR-031 | compositionality and mathematical-statistics join the topics | Two words join the topics. compositionality covers how the structure and meaning of a whole come from its parts and the way they are combined, in language, in logic and in learned models. mathematical-statistics covers the theory of statistical inference read for itself: comparison of experiments, sufficiency, decision theory and the bounds on estimation. Readers of the manuscript batch reported both gaps independently, and several works already held had been placed under the nearest wrong word. The owner authorised the extension. Rejected: a generative-modeling word, which is the anthology’s subject, and a translation word, which has two works so far and neither has been read. | Active |
| ADR-032 v4 | Argument development: QUESTION, CLAIM, CASE, TERM and ARG | The record gains five schemes for developing the owner’s own arguments rather than only documenting others’: QUESTION (what is being asked), CLAIM (what the record itself holds, with a voice — thesis, counter, granted, disclaimed, supposed — beside its credence), CASE (examples and thought experiments, with their standing and variants), TERM (concepts, versioned by supersession) and ARG (an inference, filed only when something must point at one), plus a workbench journal and two vocabularies, lines and works. Objections are claims. The relations a chain walks hold one scheme each, so the argument, its objection threads and its revisions render as lines; claims are indexed by line of inquiry. Rejected: a separate objection scheme, ARG for every claim, claims-only, a strikes: field, multi-scheme spines, lines as topic tags, and borrowing a full argumentation ontology. | Active |
| ADR-033 | translation joins the topics | translation joins the topics. ADR-031 deferred it until a second instance arrived. The owner has now asked for the bibliography of What Survives Translation? to be registered, and it brings five works of translation theory (Jakobson, Nida, House, Hatim and Mason, Nord). With Gutt and Hofstadter that makes seven, and linguistics is a loose fit for every one of them. Rejected: filing them under linguistics or philosophy-of-language. | Active |
| ADR-034 | pragmatics and myth-and-folklore join the topics | Two words join the topics. pragmatics covers language in use: speech acts, assertion and common ground, dynamic semantics, footing and stance, and rational-speech-act models. myth-and-folklore covers myths and folktales as objects of study, structurally or computationally. Readers registering the What Survives Translation? bibliography reported both gaps. The owner authorised extending the vocabulary as the record sees fit. Rejected: leaving pragmatics under linguistics and philosophy-of-language, and leaving myth under social-science. | Active |
| ADR-035 | causality joins the topics | causality joins the topics. It covers causation and the structure of causal relations, in one word: - causal order, definite and indefinite; - sheaf- and process-theoretic accounts of causality; - causal models and inference; - interventionism; - the metaphysics of causation. It enters with the successors to Gogioso and Pinzani 2021, which had no word. The owner chose one word for both clusters, and asked that tags be added without removing others. So every work whose subject is causation takes it beside its existing tags. Rejected: a narrower causal-structure that left the metaphysics of causation out. | Active |
| ADR-036 | learning-theory widens to cover training dynamics | The learning-theory topic covered only why learning generalizes. Readers of the 2026-10-09 batches on trained networks reported, independently and repeatedly, that the vocabulary had no word for training or learning dynamics, and used learning-theory loosely for them. The owner ruled that its description be widened rather than a new word added. It now covers how learning proceeds as well as why it generalizes: training dynamics, solvable models of learning trajectories, stages and phases of training, and the implicit bias of optimisation. | Active |
| ADR-037 v2 | Mathematics of Meaning is a remote, MOM, and what this record needs from it is dual-filed as its own reading | The owner’s Mathematics of Meaning project keeps its own Luria record, whose claims (CLM), positions (POS), arguments, concepts and literature bear on questions this record is working: system boundaries, individuation, the unity of a self. It is registered as the remote MOM, so a document here can cite MOM-CLM-160. What this record needs from it is dual-filed here: each needed entry gets an entry of this record’s own, which names the MOM code, carries this record’s own status from its own reading or re-check, and reports disagreement rather than fixing across the boundary. ADR-013 is extended from works to claims. | Active |
| ADR-038 v2 | A concept invoked as the record defines it is cited by its TERM code where it is used; a word used in another’s sense is left unlinked | The owner’s rule of 2026-10-11: “where we are invoking a concept specifically as its defined on the record, we can make that explicit by linking to the TERM”. The code goes inline at the first use in an entry’s body, and uses: lists the TERMs a claim is stated in. A word used in someone else’s sense, such as the will paper’s supposed “agent”, is not linked, and the absence says so. Quotations are never altered. Applied the same day across the record’s argument schemes: 25 inline citations and 13 uses: entries. Version 2 makes the uses: half a mechanism the record already had: explain: cited on CLAIM uses, so a listed TERM must be cited in the body, written at the point of use as [[uses::TERM-…]]. Phrase detection, the half no field can carry, is filed upstream as dmarx/luria#370. | Active |
| ADR-039 | Two lines of inquiry, evidence-for-experience and learned-structure, so that readings no existing line can hold get questions of their own | The third synthesis pass left about a hundred THEORYs with no edge to any claim or question, and the owner asked for questions to place them. Two clusters fit none of the six lines without strain: consciousness science on what could show experience, and learning dynamics on why trained networks have the structure they do. Two lines were added for them, approved by the owner in their words: “add both lines, they’re approved”. Rejected alternatives: forcing the clusters onto organizational-affect and pragmatic-transport, which would have misdescribed both lines, and leaving the readings unplaced. The learned-structure line stays on understanding, because instruction for machine-learning practice belongs to the anthology. | Active |
| ADR-040 | A formal-apparatus line, held as a second line by the general results that a substantive line borrows, so that setting that line aside does not take them with it | The lines-of-argument page showed pragmatic-transport holding 188 of 262 claims and 113 of 153 rests_on edges, and the other lines resting on it for general results, not for its account of translation. CLAIM-198 and CLAIM-196 rest on CLAIM-197’s gluing lemma, and nucleating-argument rests on CLAIM-056, 081, 142 and 149. So setting pragmatic-transport aside would take theorems that stand whatever it concludes, and with them the foundations of three other lines. Proposed: allow a claim more than one line, add formal-apparatus, and give it as a second value to the 24 general results that rest on nothing. A claim stays standing while any line it holds stands. Seven general results that are filed resting on translation theses are to be split, not tagged. Seven more are left to the owner. Raised by the owner on 2026-10-12. The owner proposed adding a value instead of moving claims, and approved the decision: “approved, go ahead with the yaml change and group A”. | Active |
| ADR-041 v3 | A LIT whose lawful full text was sought and not reached is Unavailable, not Deferred | Deferred on a LIT held two different situations: a work nobody has yet tried to read, and a work whose lawful full text was looked for and not reached. The owner asked for the two to be told apart. The lit-statuses vocabulary gains Unavailable (“No copy reached”), and Deferred keeps “no attempt recorded”. 177 of the 345 Deferred LITs had a status note recording a failed search, and they move. Among them are 8 where a lawful copy exists but this client could not fetch it. Three were left Deferred because their notes record no search. The other 168 stay Deferred. | Active |
| ADR-042 | A kind field on CASE, from a closed vocabulary of six, so that a proposed study, a thought experiment, a formal construction and a real specimen are told apart by something other than how much of each is fact | The owner asked whether the record could type its cases: “proposed empirical experiments vs thought experiments vs. steel-man vs. etc”. A survey of all 81 found that standing (documented, stipulated, fictional) has been doing a second job. It says how much of a case is fact, and it has been read as what kind of case it is. stipulated alone holds proposed studies, thought experiments, formal constructions, real companies and runs made in session. Proposed: a required kind field with six closed terms (specimen, thought-experiment, construction, reported-study, proposed-study, executed-study), orthogonal to standing, with every case’s value given below. Steel-man is not a kind. It is a voice, and the record holds it as a CLAIM with role counter. Approved by the owner, who renamed the last term: “approved, but change “run-here” to “executed-study” or maybe “performed-study” or something like that”. Applied: luria.yaml, the template and all 81 cases. | Active |
| ADR-043 v2 | A warrant field on CLAIM, saying how a claim is supported (demonstrated, empirical, interpretive) apart from whether the record accepts it and in whose voice, piloted before it is required | The owner, at U9 of the exchange of 2026-10-14, on the assistant’s line “what we’ve demonstrated by construction, what we’ve made plausible through examples, and what still requires empirical evidence”: “do we have anything like this already segregating CLAIMs? if not, we should consider adding it.” The record does not. status says whether the record accepts a claim, role whose voice it is in, and CASE has standing and kind, but nothing on a CLAIM says how it is warranted: an Active claim may be a theorem, an empirical generalization or an interpretation. Proposed: an optional warrant field from a closed vocabulary of three or four, attached to the claim as stated, never inherited along rests_on, piloted on the claims filed from that exchange before any decision to require it. Accepted by the owner on 2026-10-15 and applied: the field is in the schema and the template, every claim in the record is classified, one compound claim was split in the pilot, and at the owner’s direction the field is required. | Active |
| ADR-044 | A corroborates relation between the record’s own claims, for evidence one claim gives another without the second depending on it, so that a supporting claim’s retirement lowers the other’s credence without breaking its argument | While porting the LEGO material (CLAIM-310), a claim was filed with supports: CLAIM-296. The link fixer wrote the converse, rests_on, onto CLAIM-296. That says CLAIM-296 could not stand without the new claim, which is false: the new claim adds evidence for CLAIM-296 and nothing in CLAIM-296’s argument passes through it. The relation was dropped and the link kept in prose. The owner asked for a relation that says what was meant. Proposed: corroborates on CLAIM, scheme CLAIM, converse corroborated_by, not walked by the argument chain and not followed when a claim is retired. It is the claim-to-claim counterpart of a case’s illustrates beside its supports. Rejected: overloading complements, which is symmetric and means a different part of the same question; and a narrower instance_of, which fits this pilot but not the general case. Accepted by the owner on 2026-10-15 and applied to the claim that prompted it. | Active |
| ADR-045 | databases joins the topics | databases (Databases and data structures) joins the topics. It covers how stored data is organized so that it can be found and kept current: physical design and index selection, materialized views and precomputed answers, adaptive and self-organizing structures, lower bounds on lookup, caching and paging, and the cost of organizing ahead of use against organizing in use. It enters with five works read for the question whether sorting a collection is amortized search (QUESTION-114), which had been tagged information-retrieval as the nearest word, and that tag is replaced, not kept. The owner: “don’t hesitate to extend the vocabulary as needed”. Rejected: keeping information-retrieval, whose blurb is relevance and ranking; and a narrower data-structures, which leaves out physical design and views. | Active |
| ADR-046 v2 | A TERM may be Proposed, and rival, complement or narrow another | A TERM’s status is the standing of its definition, not a credence: a Proposed term is a definition recorded and not adopted. TERM gains the relations the record already uses elsewhere: rivals (symmetric, THEORY’s from ADR-022), for definitions that cannot both be adopted for one use; complements (symmetric, CLAIM’s from ADR-032), for definitions in force side by side; and refines/refined_by (QUESTION’s), for a special case that stays in force. supersedes is kept for an adoption that retires the earlier term. Applied to TERM-050/TERM-061 (complements) and TERM-010/TERM-043 (refines, in place of a supersedes that retired nothing). Rejected: new relation names (proposes_for, alternatives_to), a separate CONCEPT scheme, a TERM-only status vocabulary, and keeping candidate definitions in prose. | Active |
| ADR-047 v4 | Promotion is audited in small persisted batches, ranked by expected yield, with decisions recorded apart from status changes | The owner’s procedure of 2026-10-19 for finding entries whose evidence has outgrown their status. Candidates are triaged from metadata by a rubric that sets review order and never standing. They are reviewed five at a time, and each review ends in one of six dispositions with a specific blocker. Every step is persisted in record/promotion-audit/: a versioned queue.yaml with a hand-off block, one decision file per candidate, and one file per batch. A check keeps the queue and the decision files in step. Status changes go in separate pull requests of at most three entries. Rejected: one large audit held in a conversation, which is lost when the session is; a queue derived only from the decision files, which cannot say what is in progress; and promoting in the same step as reviewing. | Superseded — by ADR-057; Superseded on 2026-10-27 by the owner’s acceptance of ADR-057, which derives the worklist from the record on each pass instead of keeping a queue. The three evidence rulings of versions 2 to 4 stand under it. |
| ADR-048 | Nucleation files any work the owner wants here; whether a work is also held in the anthology is the anthology’s question | The boundary ADR-001 set is dropped on the nucleation side. A work goes here when the owner wants it here, whether or not an anthology topic could hold it, and nothing is checked against the anthology before filing. Instructions for machine-learning practice still belong to the anthology, which keeps that job. An ANTH- code is cited when it is known and is no longer required. Whether a work here is also held there is for the anthology to find out. The anthology-candidate flag is still applied. | Active |
| ADR-049 v2 | A QUESTION is Answered by an Active CLAIM or THEORY, and a partly answered question is split into an Answered child and a narrowed Open parent | An Active THEORY is the record’s current account, so it can answer a QUESTION as an Active CLAIM can. A CLAIM is not filed only to restate what a reading already establishes. A question is Answered when its Active answerers, alone or together, answer it as posed and in its full scope. When they answer only part of it, the answered part is filed as a child QUESTION that refines the parent and is marked Answered. The parent is reworded to what is left and stays Open. Rejected: requiring an Active CLAIM, which makes the record assert in its own voice what it relies on others to justify; marking a partly answered question Answered; and reframing the parent into new questions. | Active |
| ADR-050 v2 | A claim that calls something organized, a perspective, an agent, effective, exhausted, constituted or grounded says which sense, target or criterion it means | The perspective line’s conclusions turn on distinctions the prose keeps locally and the vocabulary does not enforce: structure against efficacy, efficacy against competence and legitimacy, explanatory sufficiency against constitution, a perspective against an agent. A proposal the owner relayed on 2026-10-22 asked for a small grammar of these words, not a term for each. The decision, Proposed, is a usage protocol: each flagged word in a claim, question, term or reading must name its sense, target or criterion where it carries a substantive inference, though not in unambiguous ordinary uses. Two words get terms of their own (organization, TERM-074; perspective, TERM-072); the rest are handled here. It is prose for now, and becomes a lint pattern only if the owner adopts it. | Deferred |
| ADR-051 | A CASE names where the record got it: reported_in links it to the work, reading or account that reports it | A documented CASE usually comes from somewhere: a paper that reports a study, a reading that describes an episode, an account whose result the case records. The record had no field for that. A CASE could support a CLAIM but not point to a LIT, NOTE or THEORY, so its provenance lived in prose. At the owner’s direction (2026-10-23), CASE gains reported_in, naming LITs, NOTEs and THEORYs, with the converse reports on each. It records provenance, not support. | Active |
| ADR-052 | The record’s own argument is stated ordinally first: orderings, partial orders and classes before scales, a numerical measure only with what it is cardinal in stated, and incomparability or partial identification reported as a result, not a gap | Three lines reached the same method separately: freedom of will as a partial order (CLAIM-354), frustration as an ordinal complex (CLAIM-347), and valence as an ordinal potential fixed up to a class (TERM-077). Each came from the owner’s correction of 2026-10-17, “We can do quite a lot with access to an ordering, even if it doesn’t give us a true measure”, and each was paid for when a measure was used before its cardinality was stated: two reviews of the valence line found Hodge measures reversing ordinal facts. At the owner’s direction (2026-10-23) the method is recorded as a rule for the record’s own argument. Rejected: leaving it to each claim, and a scale-first default with ordinal statements as a fallback. | Active |
| ADR-053 v3 | A LIT whose lawful free copy was located but could not be fetched by this client is Unfetched, not Unavailable | ADR-041 gave one word, Unavailable, to two cases: no lawful copy was found, and one was found that this client could not fetch. It weighed a separate word for the second case and declined it because “eight entries do not justify a sixth status”. On 2026-10-25 the owner supplied two such works by hand and asked for a status for “content that would be easy for me to access where you encountered obstacles”. A triage of the 140 Unavailable notes that record an access obstacle found 43 in the second case. This decision proposes a new LIT status, Unfetched (“Copy found, not fetched”), whose status note must give the copy’s address, so that the generated status page is the owner’s queue. 37 entries would move. The paywalled and the unlocated would stay Unavailable. | Active |
| ADR-054 v2 | Work is pushed to its pull request as it is made, and the pull request’s CI repairs and lints it; contributors do not gate a push on running the record’s checks locally | CLAUDE.md told contributors to run luria repair, luria link --fix, luria index and luria lint before every push. On a pull request the docs-check job already runs the repair, pushes it onto the branch, and lints the repaired tree, so the local run did the same work twice. At the record’s present size it was also slow: on 2026-10-26 one local repair–link–lint pass ran past ten minutes, and work sat unpushed while it ran. At the owner’s direction (2026-10-26), work is committed and pushed to its pull request as it is made. CI is the check, and a red check is read and fixed on the branch. luria new stays local, and a narrow local run is still allowed for diagnosing a red check. Rejected: keeping the local gate, a local lint only, and pushing straight to the default branch. | Active |
| ADR-055 | endosymbiosis joins the topics | endosymbiosis joins the topics, labelled “Endosymbiosis and reductive evolution”. It covers how a partner comes to live inside another and depend on it, and how that ends: reductive genome evolution and the dependencies it creates, including in free-living microbes; complementary loss across partners; gene transfer to the host and protein import; organellogenesis and organelle loss; eukaryogenesis; and host control of symbiont transmission. It enters with the thirty-five works registered from the exchange of 2026-10-27 on the record’s coverage. The owner asked for it (2026-10-27). ecology keeps symbiosis as an ecological interaction. Rejected: holding the works under ecology and natural-sciences, and separate reductive-evolution and organelle-evolution words. | Active |
| ADR-056 v2 | A TERM says what job its definition does (disambiguation, construct, criterion or position), piloted on a sample before it becomes a field, and a criterion TERM carries an application contract that keeps its criterion apart from its indicators | The owner, U12 of the exchange of 2026-10-26: “I feel like we need a lot of our TERMs to be like this, maybe this is a particular ‘kind’ within the TERM schema we should distinguish.” A12 proposed four kinds, a TERM that separates senses, one that introduces a theoretical object, one that states when a concept applies and one that names a position, with an application contract for the third. Proposed, and not yet a schema change: the kinds are piloted on a sample first, as CASE’s were under ADR-042, and the field says what a definition does, never whether it is true or adopted. | Rejected |
| ADR-057 v7 | Status review is ranked by reliance, derived from the record each pass, covers relations as well as statuses, and changes status in the pull request that decides it | The owner, 2026-10-27, after asking for status changes and relation changes on claims “that a lot of interesting downstream material depends on”: the record had changed materially since the promotion audit of ADR-047 stopped, and its process would benefit from changing. The audit never reached those claims: 15 of the 19 most-relied-on Proposed entries never entered its queue, its decision files went stale (230 of 308), and its owner questions sat unasked in a hand-off string. This decision replaces the procedure. A worklist is derived from the record’s frontmatter on each pass and kept nowhere. It ranks Proposed entries by how much rests on them, and it lists credence inversions, retired premises and Open questions with an Active answer. A pass reviews a load-bearing entry together with the premises and objections that hold it back. It changes statuses and relations in its own pull request, records its reasoning on the entries and in one workbench entry, and puts its owner rulings in the pull request as a checklist. ADR-047’s three evidence rulings stand. Rejected: resuming the old queue, which would re-rank a record a third larger by a rubric that ranked these claims last; keeping per-code decision files, which drift from the entries they judge; and separate promotion pull requests, which took about 140 pull requests over three days and left five decided promotions undone. | Active |
| ADR-058 | A THEORY can corroborate a CLAIM: evidence the record links to its own claim without the claim resting on it is corroborated_by, not grounds | The third status review (ADR-057) read the record’s 38 credence inversions. These are Active claims whose grounds or rests_on named a Proposed entry. In nearly all of them the claim did not need the entry. The edge had been added later as evidence (“evidence now linked”), as an instance, or for a quotation. The record had nowhere else to put a reading that bears on a claim without being a premise of it: corroborates (ADR-044) runs only from claim to claim, and complements likewise. So readings landed in grounds, which says the claim stands on them, and the status check then read a Proposed reading as a missing premise. This decision lets a THEORY corroborate a CLAIM. corroborated_by on a CLAIM takes a THEORY as well as a CLAIM, and a THEORY gains corroborates. A reading the claim needs stays in grounds. A reading cited only for an attribution or a quotation takes no edge, because the body’s citation is the link. Rejected: dropping the edges, which loses that the evidence was linked; and a new general cites relation, which would duplicate what a body citation already records. | Active |
| ADR-059 | A status is credence in the entry’s own content: an ARG’s is in its inference, within the limits its undercutters grant, and a conceded objection is recorded by revising its target, not by holding the objection | Two rulings of the third status review turned on one question. Do the status words, which mean credence for a CLAIM, mean the same thing for an ARG and for a conceded counter? The owner: “my mental model is that an ARG is similar to a CLAIM, but it’s clearly not so similar that we didn’t collapse the distinction, so maybe using the same CLAIM.status interpretations is setting us up for problems”. And of CLAIM-438: “it’s a concession, so we want to treat it as true at least within a particular context, but it might incidentally bear on other parts of the epistemic state incorrectly”. Proposed: every status is credence in what the entry itself asserts. A CLAIM asserts a proposition. An ARG asserts an inference: that its premises, if held, carry its conclusion, within the limits its undercut_by claims grant. That the argument is sound (an Active inference with every premise held) is derived and never filed. A conceded counter is recorded by revising its target, and the counter’s own status stays its credence. Rejected: reading an ARG’s status as “its premises are held”, which duplicates its premises’ statuses; and keeping “a conceded counter goes Active”, which lets a concession made in one exchange raise the credence of everything that cites the counter. | Active |
| ADR-060 v2 | A claim or account that bears on a question without answering it says bears_on, and answers is for an answer, held or offered | The third status review (ADR-057) found 34 Open questions with Active answerers, and under ADR-049 none of the 102 edges settled a part of its question exactly. They were evidence, premises, sufficient cases for which-questions, or results for one model class. answers was the only relation from a claim or account to a question, so filing used it for “bears on” too. The owner: “split the relation … the jobs filing onto the record clearly want one, and answers is not it.” A CLAIM or THEORY now has bears_on (QUESTION), with the converse bearing. answers means an answer to the question as posed: held if Active, offered if Proposed. A part that is settled exactly is split off under ADR-049. The 102 Active edges are moved. The 240 edges from Proposed entries on Open questions are left for review, because whether an offered entry would answer if held needs reading. Rejected: a qualifier on answers, which would make every consumer read two cases out of one relation. | Active |
| ADR-061 | An ARG is any claim about support; a claim about what a source says is a THEORY | Drafted 2026-10-26 under a number main had already given another decision; refiled 2026-10-28 against the record as it now stands. ADR-032 filed an ARG only when something had to point at an inference, and none has been filed since 2026-10-10. In that time the inference objections it was built for kept coming, filed as CLAIMs. A reading of all 453 claims found 42 whose subject is whether one entry supports another, and 17 more that bundle such a verdict with a claim about the subject matter. The boundary is redrawn by subject. CLAIM holds what the record says about its subject matter. ARG holds what it says about support, a negative verdict included: “X does not show Y” is the ARG from X to Y, Rejected, with its critical questions answered in the body. undercuts narrows to a claim about the subject matter that defeats an inference. A claim whose subject is what a source says becomes a THEORY, but most such claims turned out to correct a claim’s use of a reading the record already holds, so they split into a THEORY revision and a Rejected ARG. Two argument forms are added, evidence and source. A support claim that Active claims rest on is split, not superseded: rests_on names only claims, so its subject-matter part stays a claim. Rejected: folding ARG into CLAIM, a voice field on ARG, more CLAIM schemes split by role or warrant, and leaving the boundary where it is. | Active |
| ADR-062 | social-identity and altruism join the topics | Two topic words join the closed vocabulary. social-identity takes the psychology of group membership: social identity and self-categorization theory, identification, intergroup relations, rivalry and fandom. It enters with the self-categorization and fandom readings of 2026-10-28 and 29, which had only social-science. altruism takes whether action can aim at another’s good rather than one’s own. It enters with the owner’s thesis that there is no selfless action, only self-interest on a larger self, and the works filed for it. The owner asked for the vocabulary to be expanded where merited. Rejected: putting both under social-science, or the first under social-ontology and the second under moral-psychology. | Active |
| ADR-063 | future-generations joins the topics | One topic word joins the closed vocabulary. future-generations takes what the living owe, and why they care about, people who will live after them: longtermism and existential risk, intergenerational justice, the non-identity problem and person-affecting views, population ethics, discounting, and the psychology of concern for the future (legacy, generativity, symbolic immortality, continuity with one’s future self). It enters with the legacy works of 2026-10-29 and the owner’s request to extend that line to obligations toward the future. Rejected: leaving them under ethics, altruism and motivation, or naming the word longtermism. | Active |
| ADR-064 | Copies of published articles on course, faculty and scholarly third-party pages count as lawful copies to read | The first pass over the Deferred LITs on 2026-10-29 read two works from copies on university course or group pages, and turned down similar copies of ten others as being “of unknown provenance”. The owner ruled the refusals mistaken. A copy of a published article on a university course or reading-list page, a faculty or research-group page, a scholarly third-party site, or an author’s upload counts as a lawful copy to read. Shadow libraries and uploads of whole in-copyright books still do not. This narrows ADR-053’s “an upload whose uploader is unknown”. | Active |
| ADR-065 | A reading’s promote_when asks only for what its title asserts | A THEORY’s promote_when asks for exactly what its title asserts, at filing and at review. A title that reports what a source proposes, argues or finds, or what one study measured with the study’s own hedges kept, goes Active once its source is read in full and nothing in the record contradicts the report. A demand for replication, an independent test or the version of record of a copy already read moves to a QUESTION, or, where the source asserts the wider thesis, to a Proposed THEORY that extends the entry and keeps the old condition. A proof a reading re-derived counts once a second checker has gone through it. The survey of 2026-10-30 found about 58 of 597 Proposed readings whose conditions exceed their titles; general titles keep their strict conditions. Rejected: loosening conditions across the board, which would collapse the difference between reading a source and believing it; and leaving the rule where it was, in superseded ADR-047’s text, which intake did not apply. | Active |
| ADR-066 | control-theory joins the topics | One topic word joins the closed vocabulary. control-theory takes the analysis and design of feedback: stability and Lyapunov certificates, invariance, viability and barrier certificates, input-to-state stability and small-gain conditions, contraction, adaptive, dual and safe control, and controller synthesis under constraints. It enters at the owner’s word, after the exchange “Analyze nucleation concepts” leaned on these tools across six turns with nowhere to file them. Rejected: leaving them under mathematics and complex-systems, and a broader systems-theory. | Active |
| ADR-067 | line takes several values in every scheme, and CASE and ARG require it | The line: field holds several values wherever a scheme declares it: QUESTION, CASE, TERM and ARG now follow CLAIM, which already did. CASE and ARG now also require it, as QUESTION and CLAIM did; TERM does not. The nine CASEs without a line got theirs from the graph: the lines of the entries they are joined to. Both changes are the owner’s word, given 2026-11-01. Rejected: keeping one line outside CLAIM and recording the others in prose, which hid an entry from a line it bears on. | Active |
| ADR-068 | Reading and filing follow a written protocol: its invariants are held in this decision, and its steps in a skill that changes with practice | The record takes in reading through a practice of eleven steps: intake, a check against the record, identifiers, registration, a lawful copy, reading, verification, filing the reading, filing the argument, accounting for everything, and acknowledgement. Until now it lived in session memory and in the pieces ADR-002, ADR-008, ADR-041, ADR-053, ADR-054, ADR-064 and ADR-065 decided one at a time. Each session rebuilt it, and applied it unevenly (ADR-064’s case). This decision fixes the eight invariants. The steps go in .claude/skills/read-and-file/SKILL.md, edited in the pull request whose work shows a step wrong. Rejected: leaving the practice unwritten; writing every step into an ADR, which would make each refinement a supersession; a section of CLAUDE.md, which is a map and not a manual; a page under docs/, which is generated; and building it into luria before the steps have settled. | Proposed |
| ADR-069 | A Read note on a work marked Unavailable, Unfetched or Deferred fails lint | Three pattern checks join lint.fail_on: a NOTE with status Read whose paper: is a LIT marked Unavailable, Unfetched or Deferred. Read says the full text was read closely; the first two LIT statuses say no full text was reached, and Deferred that the work was not read closely enough to place. None can hold beside it. A Skimmed note on such a work is not flagged, because a précis or an introduction is a skim of an unreached work. The record has no match today, so the checks fail on the first one. The owner asked for the rule on 2026-11-01. Rejected: flagging every note on an unreached work, which would have flagged eight honest skims. | Active |