A methodology here is somebody else’s named, external work that this repository has adopted — a decision method, a responsibility matrix, an evidence framework. That is the line methodology-adoption draws: a method with no external origin is a house process, and belongs in skills/ written as one rather than dressed as an adoption.

So every node below carries an origin. Whether this checkout also holds the source that origin names is a separate question, and the one the third column answers.

31adopted methodologies

28with the source held here

3cited, not ingested

4instance(s) declaring the graph

Choosing one

applies-when is what a selection question matches against, so it is the column to read first. A methodology whose applicability is unstated is one an agent picks by resemblance, which is why the schema requires the field.

Scroll horizontally to see all columns →

methodology	applies when	origin held?	declared by
Adequacy-for-purpose modelling — a model is judged against what it is FOR, not against reality adequacy-for-purpose-modelling	A formal model is being built, and somebody will later ask whether it is right. Use it when the artefact is a mathematical model of a process — agents, action and reward spaces, update rules — and the question is what the model is allowed to claim. It answers what would count as this model succeeding, and nothing else. Reach for it BEFORE the model is written, because its whole force is that the purpose is declared first and the evaluation follows from it. Reaching for it afterwards turns it into a defence of whatever the model happens to do. Do NOT reach for it to choose between options (kepner-tregoe), to grade evidence, to settle a recurring rule (dmn), or to pick an algorithm (probabilistic-decision-analysis, mcdm-aggregation). It is not a decision method at all: it is a rule for judging a model, and it is filed here because methodology-adoption routes “how do we evaluate X” to a methodology node.	source held	cat-harness
Tool-Integrated Agentic Pre-Merge Review — Multi-Agent Verification Loops, Supervised Consistency, and Gate Boundaries agentic-pre-merge-review	A pull request or merge train is evaluated for merge admission. Use when designing or executing pre-merge quality gates, adversarial agentic code reviews, and multi-agent verification loops. It governs the strict boundary between deterministic compile/test checks (which must block) and LLM-mediated agentic evaluations (which must warn and record), specifies how to measure empirical false-positive rates counterfactually without destroying evidence, and pre-registers promotion criteria before data collection. Not applicable to deterministic compile checks (Lean, SUSHI, JSON-LD schema) which have zero false-positive rate by construction.	source held	cat-harness
Agentic Software Engineering Literature Synthesis — Cross-Paper Synthesis and Bidirectional Adversarial Analysis agentic-se-literature-synthesis	Architecting, auditing, or optimizing agentic software engineering workflows, swarm management, cost-aware model routing, pre-merge verification gates, and adversarial code reviews. Use this methodology when determining: (1) whether to route tasks statically or dynamically based on partial execution trajectories; (2) when to deploy competitive multi-agent debate versus serial or single-agent analysis; (3) how to enforce the boundary between deterministic compilation gates and stochastic agentic reviews; and (4) how to prevent runaway token costs and zombie agent executions using explicit Andon-cord predicates. Not applicable for single-prompt code completion without tool integration or multi-turn execution loops.	source held	cat-harness
Bidirectional agentic autoformalization — extract, compile-fix, check faithfulness, then informalize back without the source bidirectional-agentic-autoformalization	A whole paper, not a single theorem, is being formalised with an agent doing the Lean, and the question is how to organise the run: what to extract first, how the compile-fix loop is bounded, what happens to a statement the agent cannot prove, and how a mathematician who does not read Lean reviews what came out. Do NOT use it to decide that a formalisation is faithful because the pipeline’s own faithfulness step said so (see §3 and §”Refusals”). Not for choosing an axiom policy — this platform already has one, and it is stricter.	source held	folio-assistant-sci
Blueprint-driven formalization — Lean as the single source of dependency and status, the blueprint node as the unit of work blueprint-driven-formalization	A formalization is large enough that its state has to be tracked node by node — many interdependent definitions and theorems, several contributors or agents, and partial progress that someone must be able to read at a glance. Use it to decide where dependency and completion status are RECORDED and who may write them, and to decompose a target into units an automated prover can attempt one at a time. Do NOT use it to decide whether a Lean statement says what the prose says: a blueprint records that a node is sorry-free, never that it is faithful. That question belongs to the equivalence and vacuity audits, and this method makes it MORE pressing, not less (see §4). Not for a single-theorem formalization, where the graph has one node and the bookkeeping costs more than it saves.	source held	folio-assistant-sci
Consensus-grounded subject evaluation — independent indexers as the answer key, and a panel instead of one score consensus-grounded-subject-evaluation	Judging how good a set of controlled-vocabulary assignments is, when qualified people would themselves disagree about the exact answer. Use it to evaluate a subject-indexing system, compare two of them, or decide whether one is good enough to draft for a human reviewer. It is also for any labelling task whose gold standard is expert judgement with a subjective surface. It answers how to measure. It does not answer how to produce the assignment, which is skill-pipeline-subject-indexing, a parallel node. Not for certainty of evidence behind a recommendation (grade), and not for choosing among options (kepner-tregoe). Not applicable where the answer is decidable, meaning a single correct output a checker can verify, because then there is no disagreement for a consensus to absorb.	source held	cat-harness
Correspondence analysis for retrieval — decompose the departure from independence, not the counts correspondence-analysis	The same question as lsi — which units of a prose graph are close in what they are about — when the answer must not be dominated by how LONG a unit is or how COMMON a term is. CA removes those margins by construction, so it is the method to reach for when LSI’s first dimensions are margins (a first dimension with no negative pole) or when the question is which units are UNUSUAL — outlier pages, specimen text, a list among prose. It answers which units have similar term profiles, relative to independence. Like lsi, every output is a PROPOSAL. Not for choosing between the two methods by blending them (methodology-adoption §”Parallel, not composable”), not for a controlled vocabulary (skill-pipeline-subject-indexing), and not for any decision.	source held	cat-harness
DIIG — Digital Implementation Investment Guide diig	Planning, costing and monitoring a DIGITAL HEALTH IMPLEMENTATION inside a health programme — from forming the team through to the budget and the monitoring plan. It is a programme-investment method, not a judgement method: it does not grade evidence (that is grade), does not choose between design options (kepner-tregoe), and does not constrain how a decision is recorded (madr). Reach for it when the question is what shall we build, with whom, at what cost, and how will we know it worked — and specifically when the answer has to survive a funder.	source held	smart-base
DMN — Decision Model and Notation dmn	The criteria RECUR and the inputs are data. A gateway that must branch the same way on the same facts every time. Not for a one-off judgement — that is kepner-tregoe, recorded per madr.	source held	cat-harness
Doc-Researcher — parse for multiple granularities, then research iteratively against a sufficiency threshold doc-researcher	A question must be answered from documents this folio has ingested, and one retrieval pass will not do it. Use it when the answer is spread across several documents, or across text and figures within one, or when the asker will follow up. It answers how to search a corpus you already hold, never what to hold — library-ingestion and the L1 completeness gate answer that, and this method assumes their output. Do NOT reach for it for a single lookup in a known document. The loop below costs iterations, and a method whose cheapest path is more expensive than reading the page is the wrong method.	source held	folio-assistant-core
Hybrid LLM/deterministic — the model proposes a RULE, machinery validates and runs it hybrid-llm-deterministic	An agent must produce an artefact that something downstream will act on. Use it when a language model is in the loop and a wrong output would be acted on rather than merely read — a schema, a mapping, a classification, a branch, a judgement. It answers how to get the output safely, never which option to choose: a one-off choice among options is kepner-tregoe, a recurring rule is dmn, certainty of evidence is grade, the record of a decision is madr, who is involved is raci, and situation analysis is swot. Those pick an answer; this one constrains how an answer is produced. Not applicable where no model is involved, and unnecessary where the output is only ever read by a person who will notice it is wrong.	source held	cat-harness
JSON-LD 1.1 — the knowledge graph serialised as Linked Data in plain JSON json-ld-serialisation	Choosing how a node of the knowledge graph is written to disk or published, so that it is ordinary JSON to a reader with no RDF tooling and RDF to one with it. Use it when adding a generated .jsonld sibling, a term to a published @context, a prefix, or a new exported document. It answers how a node is identified, typed and linked in the serialisation. It does NOT answer what a node means (that is the vocabulary and the schemas), whether a relation is true (uses-editorial-review), or how a table is modelled (CSVW, bound as a vocabulary, not a serialisation).	source held	cat-harness
Kepner-Tregoe Decision Analysis kepner-tregoe	A decision with several candidate options and no recurring rule — a platform choice, an architecture question, which of three fixes to take. Contextual, not default: if the criteria recur, use dmn; if the question is certainty of evidence for a recommendation, use grade.	cited, not ingested	cat-harness
Latent Semantic Indexing — retrieve and relate by co-occurrence structure, not by shared words lsi	Finding or relating units of text that discuss the same thing in different words, across a corpus too large to read whole, where no controlled vocabulary has been assigned. Use it to ask “what else in this graph is about this?”, to propose a home for an unfiled item among existing groups, to find near-duplicates, and to surface clusters nobody named. It answers which units are close in co-occurrence structure. It does NOT answer whether a unit is relevant, correct, or a dependency: every output is a PROPOSAL a person or an agent confirms. Not for assigning terms from a controlled vocabulary (skill-pipeline-subject-indexing), not for judging an assignment (consensus-grounded-subject-evaluation), and not for any decision (kepner-tregoe, dmn).	source held	cat-harness
MADR — Markdown Architectural Decision Records madr	Bean context — the owner’s binding, 2026-09-20. When a bean records a decision, this is the form. Not for the decision METHOD (see kepner-tregoe) and not for a recurring rule (see dmn).	cited, not ingested	cat-harness
Aggregation-type MCDM — an alternatives-criteria matrix collapsed to one score per alternative mcdm-aggregation	A FIXED, FINITE set of alternatives is to be ranked against several explicit criteria, all known up front. The input is an alternatives-criteria matrix: every alternative scored on every criterion, with weights obtainable. The output is a ranking. Use it when the decision is made ONCE, from data already in hand. Choose within the family by what you can supply. SAW and MEW need only weights and normalised scores. AHP needs pairwise comparisons, which grow as the square of the criteria count, and gives a consistency ratio in return. ANP needs those plus the interdependence structure, and is the only member that admits feedback between criteria. COPRAS, MOORA, FUCA and WASPAS sit between SAW and AHP in what they ask for. Do NOT reach for it when the alternatives are explored REPEATEDLY and information accrues as you go — that is probabilistic-decision-analysis. Do not use it for a recurring rule (dmn), for a decision whose criteria are MUSTs and WANTs rather than weighted scores (kepner-tregoe), or for grading evidence. It also assumes the criteria set is complete: a criterion nobody wrote down is weighted zero, silently.	source held	cat-harness
ODRL 2.2 — what a party may do, as permission and prohibition rules over actions odrl-policies	What an actor may DO — may this party perform this action, here? Use it when a grant or a refusal has to be written down so a machine can evaluate it: who holds an action, scoped to a process, a task or a role, and which broader action it falls under. It answers permit, deny or nobody has said. It does NOT answer who is involved in an activity (raci), which lane an actor may act in (role-model), or how a decision is reached (kepner-tregoe, dmn). It is not used here for source licences, which are recorded as SPDX ids in each library entry’s licence.json.	source held	cat-harness
Probabilistic decision-making algorithms — and what their regret bounds are claims ABOUT probabilistic-decision-analysis	The alternatives can be TRIED, repeatedly, and what you learn from one try changes what you should try next. Bandits, Bayesian optimisation and tree search all live here. The defining features are that the alternative set is sampled rather than scored, that information is gathered adaptively, and that each trial costs something — the monograph’s motivating setting is scientific discovery, “where experiments are costly”. Reach for it also when the question is not which algorithm to run but what an existing bound entitles anyone to say: the monograph is an analysis text first, so it is the right source for reading a regret guarantee rather than quoting one. Do NOT reach for it for a one-shot choice from a fixed matrix — that is mcdm-aggregation — nor for a recurring rule (dmn), a MUST/WANT decision (kepner-tregoe), or grading evidence. And do not reach for it when you cannot actually run the alternatives: adaptivity is the whole premise, and without it every guarantee in the book is vacuous.	source held	cat-harness
Process-driven autoformalization — judge a formalised statement by compiling it WITH a proof, and read the first error as the step signal process-driven-autoformalization	A natural-language statement is being turned into a Lean statement and the question is how to test the candidate, or a Lean statement is being turned into prose (a blueprint, a docstring, a narrative) and the question is how to keep that prose independent of the Lean it came from. Use it for the two checks it names — compile the statement together with a proof, and decompose informalization so the result is not a paraphrase of the syntax. Do NOT use it as evidence that a statement is FAITHFUL. The paper says itself that the compiler “can only validate the formal proof’s correctness, not its semantic correspondence to the original natural language” (§5.1.2). Not for choosing, training or ranking models: the numbers in it are about the authors’ models on the authors’ dataset.	source held	folio-assistant-sci
PROV-O — the record of who did what, in which role, under which plan prov-o-provenance	Recording, after or as it happens, what was done — which activity ran, which agent was responsible for it, in which role, following which plan, and what it read and wrote — in a form another system can read without knowing this one. Here that is the log of every BPMN task run, which the QA/QC after-check grades against policy, and the derivation and alternate-of links on the published knowledge graph. It answers what happened and who answers for it. It does NOT answer whether the action was permitted (that is the ODRL policy the record points at), how to decide anything (kepner-tregoe, dmn), or who should be involved in a task before it runs (raci).	source held	cat-harness
RACI — who is involved in an activity, and in which of four ways raci	Who is involved in an activity, and how. Use it when a process or a breakdown exists and the question is participation — who answers for this, who must be asked first, who is told afterwards. Not a decision method: a one-off choice among options is kepner-tregoe, a recurring rule is dmn, certainty of evidence is grade, the record of a decision is madr, and situation analysis before any of them is swot. RACI answers who, never what or whether.	source held	cat-harness
RASCI — RACI plus Supportive, for when doing the work and owning it come apart rasci	Who is involved, when a role does the work without owning the deliverable. Use it where a separate Supportive party is real — someone who contributes effort or resources to an activity that another role answers for. If no such party exists, use raci instead: a fifth letter nobody fills is a column that makes the chart look more considered than it is. Like raci it answers who, never what or whether — a one-off choice among options is kepner-tregoe, a recurring rule is dmn, certainty of evidence is grade, the record of a decision is madr, situation analysis is swot.	cited, not ingested	cat-harness
Assessment of open standards for requirements and methodologies fitting the bootstrap Requirement requirements-standards-assessment	Evaluating how requirements authored or verified in the harness map to or from open industry standards, interchange formats, statement grammars, and quality models — and deciding which transformations belong in higher harness layers (cat-harness, and content instances above it) while keeping the bootstrap Requirement schema strictly self-contained and free of external dependencies. It answers how bootstrap requirements translate to ReqIF, OSLC-RM, EARS, ISO 29148, and ISO 25010, what each standard costs in serialization and semantic complexity, and where mapping adapters must live. It does NOT govern runtime execution logs (prov-o-provenance), decision table structures (dmn), software package bills of materials (spdx-3), or access policy enforcement (odrl-policies).	source held	cat-harness
SDLC agentic landscape — mapping LLM multi-agent systems to the software development lifecycle sdlc-agentic-landscape	Assessing, designing, or auditing multi-agent systems and automated workflows across the software development lifecycle (SDLC). Use this framework to evaluate which SDLC phases an agentic platform covers, identify architectural gaps, and benchmark agent collaboration models (e.g., Waterfall sequential vs. Agile sprint cycles vs. dynamic workflows). Not applicable for selecting single-prompt generation techniques or isolated bug-fix heuristics.	source held	cat-harness
Skill-pipeline subject indexing — one policy-grounded stage per cognitive step, each output inspectable skill-pipeline-subject-indexing	Assigning terms from a closed, rule-governed controlled vocabulary to a work: deciding what it is about and saying so in the vocabulary’s own authorised form. Use it when the vocabulary comes with a written policy manual (LCSH has the Subject Headings Manual; MeSH, AAT and FAST are the source’s own named candidates) and an authority file that can be queried, and a wrong but plausible term would be acted on in cataloguing, retrieval or linking. It answers how to produce the assignment. It does not answer how to tell whether an assignment is good: that is consensus-grounded-subject-evaluation, which is a parallel node and is not folded into this one. Not for choosing between options (kepner-tregoe), a recurring decision rule (dmn), or making a model emit a rule instead of a result (hybrid-llm-deterministic); §”Against hybrid-llm-deterministic” explains why the last is a separate method and not this one’s parent. Not for free keywording with no vocabulary to validate against, because the authority step is what the method rests on.	source held	cat-harness
SPDX 3 — a bill of materials as a graph of elements, for what crosses a trust boundary spdx-3	Describing an artefact that LEAVES this repository so that a party who does not run our tooling can check what is in it, what it depends on, under which licences, and that it arrived intact — a release, a signed package, a published instance’s dependency set, the licences of what we redistribute. It answers what is this, what is it made of, who supplied it, under what licence, with what known vulnerabilities. It does NOT answer what happened inside the repository or who answers for it (prov-o-provenance), what an actor may do (odrl-policies), how a node is serialised internally (json-ld-serialisation), or whether a block passed its QA criteria (the house QA schemas; see §”What it refuses”).	source held	cat-harness
Specification-compiled agents — the control flow comes from the diagram, not from the model’s plan specification-compiled-agents	A process is already written down as a diagram, and something must now EXECUTE it. Use it when the control flow is external and authored — a BPMN process, a DMN table, a pipeline someone drew — and the question is how an LLM should act inside it. It answers where the plan comes from, never what the plan should be: authoring the diagram is bpmn-authoring, deciding a branch from a table is dmn, and choosing between options is kepner-tregoe. Its companion is hybrid-llm-deterministic, and the two are the same shape at different grains: that one has the model emit a RULE that machinery validates and runs; this one has the model act inside a CONTROL GRAPH that machinery derived from a specification. Reach for this one when the artefact that constrains the model already exists as a diagram. Do NOT reach for it where no specification exists, or where the process is genuinely open-ended. The paper’s own evaluation is scoped to deterministic workflows and says so; a setting with ambiguity, stochasticity or open-ended human decision-making is outside what it measured, and outside what this node claims.	source held	cat-harness
SWOT — situation analysis over internal and external factors swot	Situation analysis, before a decision — never instead of one. Use it to assemble what is true about a subject’s internal attributes and its external environment, when the point is to see the field whole rather than to choose between candidate options. Not for choosing: a one-off choice among options is kepner-tregoe, a recurring rule is dmn, certainty of evidence behind a health recommendation is grade, and the record of whatever is decided is madr. SWOT produces the INPUT to those; it is not a substitute for any of them.	source held	cat-harness
Verifiable AI guideline evaluation — multi-measurement debiased inference, peer councils, and hybrid no-gold-standard bounds for clinical guidelines verifiable-ai-guideline-evaluation	Evaluating language models, knowledge-graph pipelines, and multi-agent systems on clinical guidelines (such as WHO SMART Guidelines L1 -> L2/L3 authoring, editing, and national adaptation processes, and L5 clinical workflow adherence) where gold-standard ground truth is absent, scarce, or historically conflicting, and human SME capacity is a critical bottleneck. Use it to: (1) benchmark model adherence to clinical workflows without relying on naive LLM-as-a-judge; (2) detect and classify discrepancies in clinical recommendations (extraneous advice, omissions of contraindications, timing deviations, and cross-guideline contradictions); (3) measure the fidelity and reproducibility of automated L2 (DAK) and L3 (FHIR/CQL) authoring; (4) insulate authoring from the MCQA vs free-form generative performance cliff; (5) separate intentional national/local guideline adaptation from unintentional hallucination; (6) isolate the marginal causal contribution of Knowledge Graphs (Route B vs Route A); (7) structure layered validation (mechanical -> agentic -> human SME) to protect clinical expert time; (8) operationalize automated discrepancy gating across core BPMN workflows. Not for simple single-assertion syntax checks (use mechanical compiler gates like cql-translation or sushi), and not for general software unit tests.	source held	cat-harness
WireGen: wireframing from a written design intent wiregen	Designing a USER INTERFACE, for example a page layout, a navigation scheme or a visualiser, where the choice between candidate designs has to be reviewed and, when reviewers disagree, adjudicated. It is a design-generation and design-evaluation method. It does not choose between non-UI options (use the decision-analysis methodology) or grade evidence. It does not by itself settle a disagreement: that is adjudication.	source held	cat-harness
Zero trust architecture (NIST SP 800-207) — trust is never granted implicitly, it is evaluated per request at a decision point zero-trust-architecture	A participant is about to act on something another participant gave it, and the question is whether to trust it. That covers an access decision at a boundary, a hand-over between agents, a tool or a dependency graph reaching this platform, and the design of a check that decides any of these. It answers how trust is decided: never by location, per request, by policy, at a named decision and enforcement point. It is NOT a method for choosing among options (kepner-tregoe), for recording a decision (madr), for a recurring business rule (dmn), or for who is involved (raci). It is also not a threat-modelling or risk-scoring method: it says where trust is decided, not how likely an attack is.	source held	cat-harness

Where each one came from

The three states are different facts and are kept apart deliberately. Source held means an evidence: reference resolves to a document in a declared library — you can open it from this checkout. Cited, not ingested means the node names an origin and nobody has fetched it; that is an open question, reported by check:methodology-evidence and gated by nothing. Citation does not resolve is neither: the node claims a source and the slug names nothing, which reads as evidence in every listing and is strictly worse than declaring none.

Adequacy-for-purpose modelling — a model is judged against what it is FOR, not against reality

adequacy-for-purpose-modelling — declared by cat-harness — source held

Applies when. A formal model is being built, and somebody will later ask whether it is right. Use it when the artefact is a mathematical model of a process — agents, action and reward spaces, update rules — and the question is what the model is allowed to claim. It answers what would count as this model succeeding, and nothing else. Reach for it BEFORE the model is written, because its whole force is that the purpose is declared first and the evaluation follows from it. Reaching for it afterwards turns it into a defence of whatever the model happens to do. Do NOT reach for it to choose between options (kepner-tregoe), to grade evidence, to settle a recurring rule (dmn), or to pick an algorithm (probabilistic-decision-analysis, mcdm-aggregation). It is not a decision method at all: it is a rule for judging a model, and it is filed here because methodology-adoption routes “how do we evaluate X” to a methodology node.

Origin. Kavya Ravichandran, “Algorithmic Approaches to Sequential Decision-Making and Social Epistemology”, PhD thesis, Toyota Technological Institute at Chicago, August 2026; arXiv:2607.20636v1 [cs.DS], 22 July 2026. Chapter 5, “Why Algorithmic Approaches” — an essay inside the thesis rather than a result of it. Open access, ingested whole. The view it builds on is Wendy Parker’s (2020) “adequacy-for-purpose”, quoted directly in that chapter; the how-possibly / how-actually distinction is from the philosophy-of-explanation literature the chapter cites, and the confirmatory / applied prediction split is Elliott-Graves’. Ravichandran’s contribution adopted here is the ORDERED three-purpose taxonomy and the evaluation rule that follows from it.

The method, as extracted: read it

Ingested sources:

• library/arxiv-2607.20636v1 · view · item page · source

Tool-Integrated Agentic Pre-Merge Review — Multi-Agent Verification Loops, Supervised Consistency, and Gate Boundaries

agentic-pre-merge-review — declared by cat-harness — source held

Applies when. A pull request or merge train is evaluated for merge admission. Use when designing or executing pre-merge quality gates, adversarial agentic code reviews, and multi-agent verification loops. It governs the strict boundary between deterministic compile/test checks (which must block) and LLM-mediated agentic evaluations (which must warn and record), specifies how to measure empirical false-positive rates counterfactually without destroying evidence, and pre-registers promotion criteria before data collection. Not applicable to deterministic compile checks (Lean, SUSHI, JSON-LD schema) which have zero false-positive rate by construction.

Origin. Derived from empirical agentic software engineering literature — CodeAgent (Zhang et al. 2024, arXiv:2402.02172v5), SWE-Debate (arXiv:2404.04834v4), SWE-Router (arXiv:2507.23348v1), TCAndon-Router (arXiv:2601.04544v1), and He et al. (arXiv:2607.00053v1) — combined with the folio-assistant merge-gate governance (beans folio-assistant-nok9, folio-assistant-gs0u, folio-assistant-h1uq and proposal docs/proposals/merge-gate-2026-10-02.md).

The method, as extracted: read it

Ingested sources:

• library/arxiv-2402.02172v5 · view · item page · source

Agentic Software Engineering Literature Synthesis — Cross-Paper Synthesis and Bidirectional Adversarial Analysis

agentic-se-literature-synthesis — declared by cat-harness — source held

Applies when. Architecting, auditing, or optimizing agentic software engineering workflows, swarm management, cost-aware model routing, pre-merge verification gates, and adversarial code reviews. Use this methodology when determining: (1) whether to route tasks statically or dynamically based on partial execution trajectories; (2) when to deploy competitive multi-agent debate versus serial or single-agent analysis; (3) how to enforce the boundary between deterministic compilation gates and stochastic agentic reviews; and (4) how to prevent runaway token costs and zombie agent executions using explicit Andon-cord predicates. Not applicable for single-prompt code completion without tool integration or multi-turn execution loops.

Origin. Systematic cross-paper synthesis and bidirectional adversarial critique across five foundational agentic software engineering works ingested under epic folio-assistant-0ipy: (1) Junda He, Christoph Treude, and David Lo, “LLM-Based Multi-Agent Systems for Software Engineering: Literature Review, Vision and the Road Ahead” (arXiv:2404.04834v4, 2025); (2) Zefang Li et al., “SWE-Debate: Competitive Multi-Agent Debate for Software Issue Resolution” (arXiv:2507.23348v1, 2025); (3) Yang Ding et al., “SWE-Router: Cost-Aware Temporal Routing for Multi-Turn Agentic Software Engineering” (arXiv:2607.00053v1, 2026); (4) Tang et al., “TCAndon-Router: Dynamic Agent Selection and Andon-Cord Stopping for Multi-Agent Troubleshooting” (arXiv:2601.04544v1, 2026); (5) Xunzhu Tang et al., “CodeAgent: Enhancing Code Generation with Tool-Integrated Agent Systems” (arXiv:2402.02172v5, 2024).

The method, as extracted: read it

Ingested sources:

• library/arxiv-2404.04834v4 · view · item page · source
• library/arxiv-2507.23348v1 · view · item page · source
• library/arxiv-2607.00053v1 · view · item page · source
• library/arxiv-2601.04544v1 · view · item page · source
• library/arxiv-2402.02172v5 · view · item page · source

Bidirectional agentic autoformalization — extract, compile-fix, check faithfulness, then informalize back without the source

bidirectional-agentic-autoformalization — declared by folio-assistant-sci — source held

Applies when. A whole paper, not a single theorem, is being formalised with an agent doing the Lean, and the question is how to organise the run: what to extract first, how the compile-fix loop is bounded, what happens to a statement the agent cannot prove, and how a mathematician who does not read Lean reviews what came out. Do NOT use it to decide that a formalisation is faithful because the pipeline’s own faithfulness step said so (see §3 and §”Refusals”). Not for choosing an axiom policy — this platform already has one, and it is stricter.

Origin. Yuanjie Ren, Jinzheng Li and Yidi Qi, “MerLean: An Agentic Framework for Autoformalization in Quantum Computation” (arXiv:2602.16554v1 [cs.LO], Massachusetts Institute of Technology and Northeastern University, 18 February 2026). Open access. A SYSTEM paper: it reports one agent pipeline run on three quantum-computing papers. What is adopted below is the method; several of its steps are REFUSED here, and §”Refusals” says which and why.

The method, as extracted: read it

Ingested sources:

• library/arxiv-2602.16554v1 · view · item page · source

Blueprint-driven formalization — Lean as the single source of dependency and status, the blueprint node as the unit of work

blueprint-driven-formalization — declared by folio-assistant-sci — source held

Applies when. A formalization is large enough that its state has to be tracked node by node — many interdependent definitions and theorems, several contributors or agents, and partial progress that someone must be able to read at a glance. Use it to decide where dependency and completion status are RECORDED and who may write them, and to decompose a target into units an automated prover can attempt one at a time. Do NOT use it to decide whether a Lean statement says what the prose says: a blueprint records that a node is sorry-free, never that it is faithful. That question belongs to the equivalence and vacuity audits, and this method makes it MORE pressing, not less (see §4). Not for a single-theorem formalization, where the graph has one node and the bookkeeping costs more than it saves.

Origin. Thomas Zhu, Pietro Monticone, Jeremy Avigad and Sean Welleck, “LeanArchitect: Automating Blueprint Generation for Humans and AI” (arXiv:2601.22554v1 [cs.LO], Carnegie Mellon University and University of Trento, 30 January 2026). Open access. The blueprint itself is older and is not this paper’s: it is Patrick Massot’s leanblueprint (2020), a plasTeX plugin whose \uses, \lean and \leanok macros this paper builds on (its §2, ref. [16]). leanblueprint is NOT ingested here, so everything this node says about it is SECOND-HAND, through the LeanArchitect paper. The LeanArchitect paper is a TOOL paper: it presents a method through one Lean package. What is adopted below is the method; §”Where this rendering stops” says which parts were left behind.

The method, as extracted: read it

Ingested sources:

• library/arxiv-2601.22554v1 · view · item page · source

Consensus-grounded subject evaluation — independent indexers as the answer key, and a panel instead of one score

consensus-grounded-subject-evaluation — declared by cat-harness — source held

Applies when. Judging how good a set of controlled-vocabulary assignments is, when qualified people would themselves disagree about the exact answer. Use it to evaluate a subject-indexing system, compare two of them, or decide whether one is good enough to draft for a human reviewer. It is also for any labelling task whose gold standard is expert judgement with a subjective surface. It answers how to measure. It does not answer how to produce the assignment, which is skill-pipeline-subject-indexing, a parallel node. Not for certainty of evidence behind a recommendation (grade), and not for choosing among options (kepner-tregoe). Not applicable where the answer is decidable, meaning a single correct output a checker can verify, because then there is no disagreement for a consensus to absorb.

Origin. Kwok Leong Tang, “LCSHBench: A Multilingual, Consensus-Grounded Benchmark for Library of Congress Subject Heading Assignment” (arXiv:2606.04382v1), 2026. Open access, ingested whole and read before this node was written. The method is rendered from that paper. The parallel design it is contrasted with is the one used by the shared task the other ingested subject-indexing papers were entered in: D’Souza, Sadruddin, Israel, Begoin & Slawig, “SemEval-2025 Task 5: LLMs4Subjects — LLM-based Automated Subject Tagging for a National Technical Library’s Open-Access Catalog” (arXiv:2504.07199v3), TIB Hannover. It was ingested and read too, so that design is described from its own text.

The method, as extracted: read it

Ingested sources:

• library/arxiv-2606.04382v1 · view · item page · source
• library/arxiv-2504.07199v3 · view · item page · source

Correspondence analysis for retrieval — decompose the departure from independence, not the counts

correspondence-analysis — declared by cat-harness — source held

Applies when. The same question as lsi — which units of a prose graph are close in what they are about — when the answer must not be dominated by how LONG a unit is or how COMMON a term is. CA removes those margins by construction, so it is the method to reach for when LSI’s first dimensions are margins (a first dimension with no negative pole) or when the question is which units are UNUSUAL — outlier pages, specimen text, a list among prose. It answers which units have similar term profiles, relative to independence. Like lsi, every output is a PROPOSAL. Not for choosing between the two methods by blending them (methodology-adoption §”Parallel, not composable”), not for a controlled vocabulary (skill-pipeline-subject-indexing), and not for any decision.

Origin. Correspondence analysis is Jean-Paul Benzécri’s (L’Analyse des Données, 1973) and is set out in Michael Greenacre, Theory and Applications of Correspondence Analysis (Academic Press, 1984) and Correspondence Analysis in Practice (3rd ed., 2017). Its application to information retrieval, and the comparison with latent semantic analysis this node rests on, is Qianqian Qi, David J. Hessen and Peter G. M. van der Heijden, “Improving information retrieval through correspondence analysis instead of latent semantic analysis”, Journal of Intelligent Information Systems (2023), doi:10.1007/s10844-023-00815-y — open access, ingested whole and read.

The method, as extracted: read it

Ingested sources:

• library/qi-hessen-vanderheijden-2023-ca-vs-lsa · view · item page · source

DIIG — Digital Implementation Investment Guide

diig — declared by smart-base — source held

Applies when. Planning, costing and monitoring a DIGITAL HEALTH IMPLEMENTATION inside a health programme — from forming the team through to the budget and the monitoring plan. It is a programme-investment method, not a judgement method: it does not grade evidence (that is grade), does not choose between design options (kepner-tregoe), and does not constrain how a decision is recorded (madr). Reach for it when the question is what shall we build, with whom, at what cost, and how will we know it worked — and specifically when the answer has to survive a funder.

Origin. World Health Organization, International Telecommunication Union and the United Nations Foundation Digital Health Initiative, Digital implementation investment guide (DIIG): integrating digital interventions into health programmes (2020), ISBN 978-92-4-001056-7. Ingested at smart-base/library/9789240010567-eng/; every citation below resolves to a section there.

The method, as extracted: read it

Ingested sources:

• library/9789240010567-eng · view · item page

DMN — Decision Model and Notation

dmn — declared by cat-harness — source held

Applies when. The criteria RECUR and the inputs are data. A gateway that must branch the same way on the same facts every time. Not for a one-off judgement — that is kepner-tregoe, recorded per madr.

Origin. Object Management Group, Decision Model and Notation (DMN) Version 1.5, OMG document formal/24-01-01, January 2024 (omg.org/spec/DMN). RECORDED, not held: its licence forbids posting copies on a network, so the library entry identifies the exact PDF (sha256) and its outline and holds no text.

The method, as extracted: read it

Ingested sources:

• library/omg-2024-dmn-1-5 · view · item page

Doc-Researcher — parse for multiple granularities, then research iteratively against a sufficiency threshold

doc-researcher — declared by folio-assistant-core — source held

Applies when. A question must be answered from documents this folio has ingested, and one retrieval pass will not do it. Use it when the answer is spread across several documents, or across text and figures within one, or when the asker will follow up. It answers how to search a corpus you already hold, never what to hold — library-ingestion and the L1 completeness gate answer that, and this method assumes their output. Do NOT reach for it for a single lookup in a known document. The loop below costs iterations, and a method whose cheapest path is more expensive than reading the page is the wrong method.

Origin. Kuicai Dong, Shurui Huang, Fangda Ye, Wei Han, Zhi Zhang, Dexun Li, Wenjun Li, Qu Yang, Gang Wang, Yichao Wang, Chen Zhang and Yong Liu, “Doc-Researcher: A Unified System for Multimodal Document Parsing and Deep Research” (arXiv:2510.21603v1, Huawei Technologies, 24 October 2025). Open access. THE PRIMARY IS HELD AND PROMOTED, at library/arxiv-2510.21603v1, which is what evidence: below points at. It was staged and unpromotable for a day: the extractor reported 383 images, 335 of them on page 3, and its image-descriptions requirement could not be met. It holds fifty. See §”What this checkout holds” — the first reading of those 383 was wrong, and the correction is kept there rather than tidied away. It is a SYSTEM paper reporting one implementation against a benchmark its own authors built, so what is adopted below is the METHOD, and §”Where this rendering stops” says which parts were left behind.

The method, as extracted: read it

Ingested sources:

• library/arxiv-2510.21603v1 · view · item page · source

Hybrid LLM/deterministic — the model proposes a RULE, machinery validates and runs it

hybrid-llm-deterministic — declared by cat-harness — source held

Applies when. An agent must produce an artefact that something downstream will act on. Use it when a language model is in the loop and a wrong output would be acted on rather than merely read — a schema, a mapping, a classification, a branch, a judgement. It answers how to get the output safely, never which option to choose: a one-off choice among options is kepner-tregoe, a recurring rule is dmn, certainty of evidence is grade, the record of a decision is madr, who is involved is raci, and situation analysis is swot. Those pick an answer; this one constrains how an answer is produced. Not applicable where no model is involved, and unnecessary where the output is only ever read by a person who will notice it is wrong.

Origin. Felix Neubauer, Jürgen Pleiss and Benjamin Uekermann, “AI-assisted JSON Schema Creation and Mapping” (arXiv:2508.05192v2), University of Stuttgart. The authors’ own word for it is a “hybrid approach that combines large language models (LLMs) with deterministic techniques”. Open access, read whole and ingested here — unlike most nodes in this graph, the primary IS held. It is a tool paper reporting one implementation (MetaConfigurator), so what is adopted below is the METHOD it generalises, and §”Where this rendering stops” says which parts were left behind.

The method, as extracted: read it

Ingested sources:

• library/arxiv-2508.05192v2 · view · item page · source

JSON-LD 1.1 — the knowledge graph serialised as Linked Data in plain JSON

json-ld-serialisation — declared by cat-harness — source held

Applies when. Choosing how a node of the knowledge graph is written to disk or published, so that it is ordinary JSON to a reader with no RDF tooling and RDF to one with it. Use it when adding a generated .jsonld sibling, a term to a published @context, a prefix, or a new exported document. It answers how a node is identified, typed and linked in the serialisation. It does NOT answer what a node means (that is the vocabulary and the schemas), whether a relation is true (uses-editorial-review), or how a table is modelled (CSVW, bound as a vocabulary, not a serialisation).

Origin. W3C, “JSON-LD 1.1 — A JSON-based Serialization for Linked Data”, W3C Recommendation 16 July 2020 (https://www.w3.org/TR/2020/REC-json-ld11-20200716/), editors Gregg Kellogg, Pierre-Antoine Champin and Dave Longley; produced by the JSON-LD Working Group. Held as the Working Group’s own publication snapshot of the REC (github.com/w3c/json-ld-syntax, commit 029777cf), under the W3C Software and Document License. The companion Recommendations — JSON-LD 1.1 Processing Algorithms and API, and JSON-LD 1.1 Framing — are NOT held.

The method, as extracted: read it

Ingested sources:

• library/w3c-2020-json-ld-1-1 · view · item page

Kepner-Tregoe Decision Analysis

kepner-tregoe — declared by cat-harness — cited, not ingested

Applies when. A decision with several candidate options and no recurring rule — a platform choice, an architecture question, which of three fixes to take. Contextual, not default: if the criteria recur, use dmn; if the question is certainty of evidence for a recommendation, use grade.

Origin. Charles H. Kepner and Benjamin B. Tregoe, The Rational Manager (1965); The New Rational Manager (1981)

The method, as extracted: read it

No ingested source. The origin above names one; nothing in this checkout holds it. literature-search is the skill that closes one of these.

Latent Semantic Indexing — retrieve and relate by co-occurrence structure, not by shared words

lsi — declared by cat-harness — source held

Applies when. Finding or relating units of text that discuss the same thing in different words, across a corpus too large to read whole, where no controlled vocabulary has been assigned. Use it to ask “what else in this graph is about this?”, to propose a home for an unfiled item among existing groups, to find near-duplicates, and to surface clusters nobody named. It answers which units are close in co-occurrence structure. It does NOT answer whether a unit is relevant, correct, or a dependency: every output is a PROPOSAL a person or an agent confirms. Not for assigning terms from a controlled vocabulary (skill-pipeline-subject-indexing), not for judging an assignment (consensus-grounded-subject-evaluation), and not for any decision (kepner-tregoe, dmn).

Origin. Scott Deerwester, Susan T. Dumais, George W. Furnas, Thomas K. Landauer and Richard A. Harshman, “Indexing by Latent Semantic Analysis”, Journal of the American Society for Information Science 41(6):391–407 (1990), for the method. Susan T. Dumais, “Improving the retrieval of information from external sources”, Behavior Research Methods, Instruments & Computers 23(2):229–236 (1991), for the log-entropy term weighting. Michael W. Berry, Susan T. Dumais and Gavin W. O’Brien, “Using Linear Algebra for Intelligent Information Retrieval”, SIAM Review 37(4):573–595 (1995), for folding-in and updating. Thomas K. Landauer, Peter W. Foltz and Darrell Laham, “An Introduction to Latent Semantic Analysis”, Discourse Processes 25:259–284 (1998), and Thomas K. Landauer and Susan T. Dumais, “A Solution to Plato’s Problem”, Psychological Review 104(2):211–240 (1997), for the cognitive reading of the same computation (there called LSA), which this node does NOT adopt. Nathan Halko, Per-Gunnar Martinsson and Joel A. Tropp, “Finding Structure with Randomness”, SIAM Review 53(2):217–288 (2011; arXiv:0909.4061v2), for the randomized SVD this platform computes it with — an implementation source, not part of the method.

The method, as extracted: read it

Ingested sources:

• library/deerwester-1990-indexing-by-lsa · view · item page
• library/landauer-foltz-laham-1998-intro-lsa · view · item page
• library/arxiv-0909.4061v2 · view · item page
• library/qi-hessen-vanderheijden-2023-ca-vs-lsa · view · item page · source
• library/arxiv-2202.02427v1 · view · item page · source

MADR — Markdown Architectural Decision Records

madr — declared by cat-harness — cited, not ingested

Applies when. Bean context — the owner’s binding, 2026-09-20. When a bean records a decision, this is the form. Not for the decision METHOD (see kepner-tregoe) and not for a recurring rule (see dmn).

Origin. Michael Nygard, “Documenting Architecture Decisions” (2011), for the ADR form; MADR (github.com/adr/madr) for the Markdown template with an explicit Considered-Options section.

The method, as extracted: read it

No ingested source. The origin above names one; nothing in this checkout holds it. literature-search is the skill that closes one of these.

Aggregation-type MCDM — an alternatives-criteria matrix collapsed to one score per alternative

mcdm-aggregation — declared by cat-harness — source held

Applies when. A FIXED, FINITE set of alternatives is to be ranked against several explicit criteria, all known up front. The input is an alternatives-criteria matrix: every alternative scored on every criterion, with weights obtainable. The output is a ranking. Use it when the decision is made ONCE, from data already in hand. Choose within the family by what you can supply. SAW and MEW need only weights and normalised scores. AHP needs pairwise comparisons, which grow as the square of the criteria count, and gives a consistency ratio in return. ANP needs those plus the interdependence structure, and is the only member that admits feedback between criteria. COPRAS, MOORA, FUCA and WASPAS sit between SAW and AHP in what they ask for. Do NOT reach for it when the alternatives are explored REPEATEDLY and information accrues as you go — that is probabilistic-decision-analysis. Do not use it for a recurring rule (dmn), for a decision whose criteria are MUSTs and WANTs rather than weighted scores (kepner-tregoe), or for grading evidence. It also assumes the criteria set is complete: a criterion nobody wrote down is weighted zero, silently.

Origin. Zhiyuan Wang (Singapore University of Social Sciences) and Gade Pandu Rangaiah (National University of Singapore; Vellore Institute of Technology), “Multi-Criteria Decision-Making: Aggregation-Type Methods”, Chapter 8 of a forthcoming volume; arXiv:2509.06388v1, 2026. The copy ingested is the authors’ preliminary draft manuscript, produced in Word and carrying no arXiv stamp, headed “Preliminary Draft Manuscript” and paginated 8-1 onward. _pdf_doc_id.py reads the arXiv id off page one’s text layer, so this copy derived no arxiv- slug and fell back to the basename. Filed instead under the author-year convention the repository’s other non-arXiv entries use (owner, 2026-09-29), which is why the bib-slug and the citation differ: cite arXiv:2509.06388v1; the slug is a filing key and nothing more.

The method, as extracted: read it

Ingested sources:

• library/wang-rangaiah-2026-mcdm-aggregation · view · item page

ODRL 2.2 — what a party may do, as permission and prohibition rules over actions

odrl-policies — declared by cat-harness — source held

Applies when. What an actor may DO — may this party perform this action, here? Use it when a grant or a refusal has to be written down so a machine can evaluate it: who holds an action, scoped to a process, a task or a role, and which broader action it falls under. It answers permit, deny or nobody has said. It does NOT answer who is involved in an activity (raci), which lane an actor may act in (role-model), or how a decision is reached (kepner-tregoe, dmn). It is not used here for source licences, which are recorded as SPDX ids in each library entry’s licence.json.

Origin. W3C, “ODRL Information Model 2.2”, W3C Recommendation 15 February 2018, edited by Renato Iannella and Serena Villata, produced by the W3C Permissions and Obligations Expression Working Group. The held copy is the Working Group’s own staged REC snapshot (github.com/w3c/poe), not compared byte for byte with the w3.org copy and without later errata — see the entry’s licence.json. The companion “ODRL Vocabulary & Expression 2.2”, which the model defers to for normative serialisation and the Common Vocabulary, is NOT held.

The method, as extracted: read it

Ingested sources:

• library/w3c-2018-odrl-model-2-2 · view · item page

Probabilistic decision-making algorithms — and what their regret bounds are claims ABOUT

probabilistic-decision-analysis — declared by cat-harness — source held

Applies when. The alternatives can be TRIED, repeatedly, and what you learn from one try changes what you should try next. Bandits, Bayesian optimisation and tree search all live here. The defining features are that the alternative set is sampled rather than scored, that information is gathered adaptively, and that each trial costs something — the monograph’s motivating setting is scientific discovery, “where experiments are costly”. Reach for it also when the question is not which algorithm to run but what an existing bound entitles anyone to say: the monograph is an analysis text first, so it is the right source for reading a regret guarantee rather than quoting one. Do NOT reach for it for a one-shot choice from a fixed matrix — that is mcdm-aggregation — nor for a recurring rule (dmn), a MUST/WANT decision (kepner-tregoe), or grading evidence. And do not reach for it when you cannot actually run the alternatives: adaptivity is the whole premise, and without it every guarantee in the book is vacuous.

Origin. Agustinus Kristiadi (Western University and Vector Institute, Canada), “Introduction to the Analysis of Probabilistic Decision-Making Algorithms”; arXiv:2508.21620v2 [cs.LG], 23 May 2026. A monograph, open access, ingested whole. Its stated aim is accessibility: “theoretical analyses in the literature are often inaccessible to non-experts”, and it assumes only basic probability and statistics plus some Gaussian processes.

The method, as extracted: read it

Ingested sources:

• library/arxiv-2508.21620v2 · view · item page · source

Process-driven autoformalization — judge a formalised statement by compiling it WITH a proof, and read the first error as the step signal

process-driven-autoformalization — declared by folio-assistant-sci — source held

Applies when. A natural-language statement is being turned into a Lean statement and the question is how to test the candidate, or a Lean statement is being turned into prose (a blueprint, a docstring, a narrative) and the question is how to keep that prose independent of the Lean it came from. Use it for the two checks it names — compile the statement together with a proof, and decompose informalization so the result is not a paraphrase of the syntax. Do NOT use it as evidence that a statement is FAITHFUL. The paper says itself that the compiler “can only validate the formal proof’s correctness, not its semantic correspondence to the original natural language” (§5.1.2). Not for choosing, training or ranking models: the numbers in it are about the authors’ models on the authors’ dataset.

Origin. Jianqiao Lu, Yingjia Wan, Zhengying Liu, Yinya Huang, Jing Xiong, Chengwu Liu, Jianhao Shen, Hui Jin, Jipeng Zhang, Haiming Wang, Zhicheng Yang, Jing Tang and Zhijiang Guo, “Process-Driven Autoformalization in Lean 4” (arXiv:2406.01940v2 [cs.CL], 14 October 2024; version 1, June 2024). The paper labels itself “Work in progress”. Open access. It is a MACHINE-LEARNING paper: it contributes a dataset (FormL4, built by informalizing Mathlib 4 theorems) and a training loop (an autoformalizer and a verifier fine-tuned against Lean compiler feedback). This platform trains no models, so what is adopted below is the small part of the method that survives without training; §”Where this rendering stops” is long on purpose.

The method, as extracted: read it

Ingested sources:

• library/arxiv-2406.01940v2 · view · item page · source

PROV-O — the record of who did what, in which role, under which plan

prov-o-provenance — declared by cat-harness — source held

Applies when. Recording, after or as it happens, what was done — which activity ran, which agent was responsible for it, in which role, following which plan, and what it read and wrote — in a form another system can read without knowing this one. Here that is the log of every BPMN task run, which the QA/QC after-check grades against policy, and the derivation and alternate-of links on the published knowledge graph. It answers what happened and who answers for it. It does NOT answer whether the action was permitted (that is the ODRL policy the record points at), how to decide anything (kepner-tregoe, dmn), or who should be involved in a task before it runs (raci).

Origin. W3C, “PROV-O: The PROV Ontology”, W3C Recommendation 30 April 2013 (http://www.w3.org/TR/2013/REC-prov-o-20130430/), editors Timothy Lebo, Satya Sahoo and Deborah McGuinness, for the Provenance Working Group. It is the OWL2 encoding of the PROV Data Model (PROV-DM), which is cited by it and not held here.

The method, as extracted: read it

Ingested sources:

• library/w3c-2013-prov-o · view · item page
• library/w3c-2024-prov-jsonld · view · item page

RACI — who is involved in an activity, and in which of four ways

raci — declared by cat-harness — source held

Applies when. Who is involved in an activity, and how. Use it when a process or a breakdown exists and the question is participation — who answers for this, who must be asked first, who is told afterwards. Not a decision method: a one-off choice among options is kepner-tregoe, a recurring rule is dmn, certainty of evidence is grade, the record of a decision is madr, and situation analysis before any of them is swot. RACI answers who, never what or whether.

Origin. NO PRIMARY SOURCE IS HELD HERE, and the ingested one is SECONDARY. RACI is a responsibility assignment matrix from project-management practice. The text this node cites renders the four roles but does not originate them — it attributes them onward to PMI’s PMBOK Guide (2021), which is a paid standard nobody here has opened. The earlier Linear Responsibility Chart literature is the other commonly named ancestor. Both stay in §”The primary is still not held” as CANDIDATES TO FETCH, unverified, rather than asserted here as provenance. Rendering from an open secondary was the owner’s ruling of 2026-09-23 — the route swot took. See literature-search.

The method, as extracted: read it

Ingested sources:

• library/dusengumuremyi-2026-ai-mediated-raci · view · item page

RASCI — RACI plus Supportive, for when doing the work and owning it come apart

rasci — declared by cat-harness — cited, not ingested

Applies when. Who is involved, when a role does the work without owning the deliverable. Use it where a separate Supportive party is real — someone who contributes effort or resources to an activity that another role answers for. If no such party exists, use raci instead: a fifth letter nobody fills is a column that makes the chart look more considered than it is. Like raci it answers who, never what or whether — a one-off choice among options is kepner-tregoe, a recurring rule is dmn, certainty of evidence is grade, the record of a decision is madr, situation analysis is swot.

Origin. NOT ESTABLISHED FROM ANY SOURCE HELD HERE, and that is stated rather than guessed. RASCI (also written RASIC) is the five-letter member of the responsibility-assignment-matrix family from project-management practice. No text for it is ingested in this repository and none was reachable when this node was written. The one RACI source this checkout does hold — library/dusengumuremyi-2026-ai-mediated-raci — was searched and contains ZERO occurrences of rasci, racsi, supportive or five roles, so it backs raci and expressly not this. The candidates are the same paid standards and books listed in raci’s §”The primary is still not held”. See literature-search.

The method, as extracted: read it

No ingested source. The origin above names one; nothing in this checkout holds it. literature-search is the skill that closes one of these.

Assessment of open standards for requirements and methodologies fitting the bootstrap Requirement

requirements-standards-assessment — declared by cat-harness — source held

Applies when. Evaluating how requirements authored or verified in the harness map to or from open industry standards, interchange formats, statement grammars, and quality models — and deciding which transformations belong in higher harness layers (cat-harness, and content instances above it) while keeping the bootstrap Requirement schema strictly self-contained and free of external dependencies. It answers how bootstrap requirements translate to ReqIF, OSLC-RM, EARS, ISO 29148, and ISO 25010, what each standard costs in serialization and semantic complexity, and where mapping adapters must live. It does NOT govern runtime execution logs (prov-o-provenance), decision table structures (dmn), software package bills of materials (spdx-3), or access policy enforcement (odrl-policies).

Origin. Survey and comparative assessment of formal and open requirements standards: OMG Requirements Interchange Format (ReqIF 1.2, formal/2013-10-01); OASIS Open Services for Lifecycle Collaboration Requirements Management (OSLC-RM 2.1 / OSLC Core 3.0); Easy Approach to Requirements Syntax (EARS: Mavin et al., IEEE RE 2009); ISO/IEC/IEEE 29148:2018 (Systems and software engineering — Life cycle processes — Requirements engineering); IEEE Std 830-1998 (Recommended Practice for Software Requirements Specifications, superseded by ISO/IEC/IEEE 29148); ISO/IEC 25010:2011/2023 (Systems and software engineering — Systems and software Quality Requirements and Evaluation — System and software quality models); INCOSE Guide for Writing Requirements (INCOSE-TP-2010-006-04, 4th Edition, 2023). Evaluated against the bootstrap Requirement schema (bootstrap/schemas/requirement.schema.json, Issue #1164).

The method, as extracted: read it

Ingested sources:

• library/rfc2119-key-words-requirement-levels · view · item page
• library/rfc8174-uppercase-vs-lowercase-2119-key-words · view · item page

SDLC agentic landscape — mapping LLM multi-agent systems to the software development lifecycle

sdlc-agentic-landscape — declared by cat-harness — source held

Applies when. Assessing, designing, or auditing multi-agent systems and automated workflows across the software development lifecycle (SDLC). Use this framework to evaluate which SDLC phases an agentic platform covers, identify architectural gaps, and benchmark agent collaboration models (e.g., Waterfall sequential vs. Agile sprint cycles vs. dynamic workflows). Not applicable for selecting single-prompt generation techniques or isolated bug-fix heuristics.

Origin. Junda He, Christoph Treude, and David Lo, “LLM-Based Multi-Agent Systems for Software Engineering: Literature Review, Vision and the Road Ahead” (arXiv:2404.04834v4, 2025), Singapore Management University. Systematic review (30pp) covering multi-agent architectures, SDLC stage taxonomies, empirical case studies, and research agendas for agentic software engineering.

The method, as extracted: read it

Ingested sources:

• library/arxiv-2404.04834v4 · view · item page · source

Skill-pipeline subject indexing — one policy-grounded stage per cognitive step, each output inspectable

skill-pipeline-subject-indexing — declared by cat-harness — source held

Applies when. Assigning terms from a closed, rule-governed controlled vocabulary to a work: deciding what it is about and saying so in the vocabulary’s own authorised form. Use it when the vocabulary comes with a written policy manual (LCSH has the Subject Headings Manual; MeSH, AAT and FAST are the source’s own named candidates) and an authority file that can be queried, and a wrong but plausible term would be acted on in cataloguing, retrieval or linking. It answers how to produce the assignment. It does not answer how to tell whether an assignment is good: that is consensus-grounded-subject-evaluation, which is a parallel node and is not folded into this one. Not for choosing between options (kepner-tregoe), a recurring decision rule (dmn), or making a model emit a rule instead of a result (hybrid-llm-deterministic); §”Against hybrid-llm-deterministic” explains why the last is a separate method and not this one’s parent. Not for free keywording with no vocabulary to validate against, because the authority step is what the method rests on.

Origin. Eric H. C. Chow, “A Skill-Based Agentic Pipeline for Library of Congress Subject Indexing” (arXiv:2605.03537v1), School of Humanities, The University of Hong Kong. Open access, ingested whole and read before this node was written. The method is rendered from that paper alone. Two parallel tracks, both entrants in SemEval-2025 Task 5, were ingested beside it and read so that §”The parallel tracks, and why they are not this node” rests on the papers and not on how Chow summarises them: Suominen, Inkinen & Lehtinen, “Annif at SemEval-2025 Task 5: Traditional XMTC augmented by LLMs” (arXiv:2504.19675v2), National Library of Finland; and Bayrami Asl Tekanlou et al., “Homa at SemEval-2025 Task 5: Aligning Librarian Records with OntoAligner for Subject Tagging” (arXiv:2504.21474v1).

The method, as extracted: read it

Ingested sources:

• library/arxiv-2605.03537v1 · view · item page · source
• library/arxiv-2504.19675v2 · view · item page · source
• library/arxiv-2504.21474v1 · view · item page · source

SPDX 3 — a bill of materials as a graph of elements, for what crosses a trust boundary

spdx-3 — declared by cat-harness — source held

Applies when. Describing an artefact that LEAVES this repository so that a party who does not run our tooling can check what is in it, what it depends on, under which licences, and that it arrived intact — a release, a signed package, a published instance’s dependency set, the licences of what we redistribute. It answers what is this, what is it made of, who supplied it, under what licence, with what known vulnerabilities. It does NOT answer what happened inside the repository or who answers for it (prov-o-provenance), what an actor may do (odrl-policies), how a node is serialised internally (json-ld-serialisation), or whether a block passed its QA criteria (the house QA schemas; see §”What it refuses”).

Origin. The Linux Foundation and its Contributors, with SPDX Model contributions from OMG, “System Package Data Exchange (SPDX) Specification Version 3.0”, OMG formal/24-11-01, March 2025 (https://www.omg.org/spec/SPDX); its model is the SPDX 3.0.1 model (Annex A points at https://spdx.org/rdf/3.0.1/spdx-model.ttl). Licensed Community-Spec-1.0, with pre-existing portions CC-BY-3.0. SPDX® is a registered trademark of The Linux Foundation. SPDX 3.1, a release candidate since 2026-01-26, is held only as a secondary source (a 2025 conference deck, CC-BY-SA-3.0) and as its RC1 machine-readable schema queued in uploads/, NOT as a specification.

The method, as extracted: read it

Ingested sources:

• library/omg-2024-spdx-3-0 · view · item page
• library/strauch-carbno-2025-spdx-3-1-supply-chain · view · item page

Specification-compiled agents — the control flow comes from the diagram, not from the model’s plan

specification-compiled-agents — declared by cat-harness — source held

Applies when. A process is already written down as a diagram, and something must now EXECUTE it. Use it when the control flow is external and authored — a BPMN process, a DMN table, a pipeline someone drew — and the question is how an LLM should act inside it. It answers where the plan comes from, never what the plan should be: authoring the diagram is bpmn-authoring, deciding a branch from a table is dmn, and choosing between options is kepner-tregoe. Its companion is hybrid-llm-deterministic, and the two are the same shape at different grains: that one has the model emit a RULE that machinery validates and runs; this one has the model act inside a CONTROL GRAPH that machinery derived from a specification. Reach for this one when the artefact that constrains the model already exists as a diagram. Do NOT reach for it where no specification exists, or where the process is genuinely open-ended. The paper’s own evaluation is scoped to deterministic workflows and says so; a setting with ambiguity, stochasticity or open-ended human decision-making is outside what it measured, and outside what this node claims.

Origin. Harris Borman, Herman Wandabwa, Fusun Yu, Sandeepa Kannangara, Justin Liu, Anna Leontjeva and Ritchie Ng, “Beyond Generalist LLMs: Specialist Agentic Systems for Structured Code Workflow Execution”, Commonwealth Bank of Australia. Published as a workshop paper at SCALE, ICML 2026 (PMLR 306); arXiv:2607.14456v1 [cs.SE], 16 July 2026. Open access, ingested whole and read before this node was written. It is a SYSTEM paper reporting one implementation against a benchmark its own authors built, so what is adopted below is the METHOD, and §”Where this rendering stops” says which parts were left behind. §”What the figures say that the prose does not” carries a reading of the paper’s own charts that changes how its headline numbers should be used, and it is the reason this node quotes almost none of them.

The method, as extracted: read it

Ingested sources:

• library/arxiv-2607.14456v1 · view · item page
• library/kg-folio-asst-2026-09-30 · view · item page
• library/omg-2013-bpmn-2-0-2 · view · item page

SWOT — situation analysis over internal and external factors

swot — declared by cat-harness — source held

Applies when. Situation analysis, before a decision — never instead of one. Use it to assemble what is true about a subject’s internal attributes and its external environment, when the point is to see the field whole rather than to choose between candidate options. Not for choosing: a one-off choice among options is kepner-tregoe, a recurring rule is dmn, certainty of evidence behind a health recommendation is grade, and the record of whatever is decided is madr. SWOT produces the INPUT to those; it is not a substitute for any of them.

Origin. Rendered from two ingested sources. Gürel, E. & Tat, M. (2017), “SWOT Analysis: A Theoretical Review”, The Journal of International Social Research 10(51), pp. 994–1006, doi:10.17719/jisr.2017.1832 — a theoretical review, carrying the history, the variants and the criticism. And Sammut-Bonnici, T. & Galea, D. (2015), “SWOT Analysis”, Wiley Encyclopedia of Management vol. 12 (Strategic Management), doi:10.1002/9781118785317.weom120103 — a reference chapter, carrying the conceptual framework and the practical discipline. The METHOD’s own origin is contested and neither source settles it — see §”Where the method came from, and why that is not a settled question”.

The method, as extracted: read it

Ingested sources:

• library/gurel-tat-2017-swot-analysis · view · item page
• library/sammut-bonnici-galea-2015-swot-analysis · view · item page

Verifiable AI guideline evaluation — multi-measurement debiased inference, peer councils, and hybrid no-gold-standard bounds for clinical guidelines

verifiable-ai-guideline-evaluation — declared by cat-harness — source held

Applies when. Evaluating language models, knowledge-graph pipelines, and multi-agent systems on clinical guidelines (such as WHO SMART Guidelines L1 -> L2/L3 authoring, editing, and national adaptation processes, and L5 clinical workflow adherence) where gold-standard ground truth is absent, scarce, or historically conflicting, and human SME capacity is a critical bottleneck. Use it to: (1) benchmark model adherence to clinical workflows without relying on naive LLM-as-a-judge; (2) detect and classify discrepancies in clinical recommendations (extraneous advice, omissions of contraindications, timing deviations, and cross-guideline contradictions); (3) measure the fidelity and reproducibility of automated L2 (DAK) and L3 (FHIR/CQL) authoring; (4) insulate authoring from the MCQA vs free-form generative performance cliff; (5) separate intentional national/local guideline adaptation from unintentional hallucination; (6) isolate the marginal causal contribution of Knowledge Graphs (Route B vs Route A); (7) structure layered validation (mechanical -> agentic -> human SME) to protect clinical expert time; (8) operationalize automated discrepancy gating across core BPMN workflows. Not for simple single-assertion syntax checks (use mechanical compiler gates like cql-translation or sushi), and not for general software unit tests.

Origin. Synthesized from eight peer-reviewed and open-access mathematical foundations for objective evaluation in the absence of gold standards and public health guideline verification: (1) David Nordfors, “The Metanym Game: An LLM Benchmark Without Ground Truth That Rises With the Models It Measures” (arXiv:2606.21008v4, 2026) — SVD of peer ratings, separating generation from evaluation skill, and swept calibration anchors; (2) Naoki Egami & Sooahn Shin, “Debiased Inference for AI-Generated Data without Gold-Standard Labels: Identification via Multiple Imperfect Measurements” (arXiv:2608.18294v2, 2026, MIT) — DMM framework via CP tensor decomposition conditioned on unit-level text features; (3) Yan Liu & Abhinav K. Jha, “How accurately can quantitative imaging methods be ranked without ground truth: An upper bound on no-gold-standard evaluation” (arXiv:2403.16873v1, 2024, SPIE/WashU) — Cramér-Rao bound on ranking error; (4) Yan Liu & Abhinav K. Jha, “Extending Regression Without Truth to Integrate Ground-Truth Measurements for Evaluating Quantitative Imaging Methods with Patient Data” (arXiv:2603.27124v1, 2026, WashU) — semi-supervised joint likelihood; (5) Ziping Liu, Zekun Li, Joyce C. Mhlanga, Barry A. Siegel & Abhinav K. Jha, “No-gold-standard evaluation of quantitative imaging methods in the presence of correlated noise” (arXiv:2203.02010v1, 2022, SPIE/WashU) — off-diagonal error covariance modeling; (6) Kaivalya Rawal, Zihao Fu, Eoin Delaney & Chris Russell, “Evaluating Model Explanations without Ground Truth” (arXiv:2505.10399v1, FAccT 2025, Oxford) — AXE on-manifold fidelity and anti-fairwashing principles; (7) Joshua Harris et al., UK Health Security Agency, “Evaluating Large Language Models for Public Health Classification and Extraction Tasks” (arXiv:2405.14766v2, 2025) — benchmarked recommendation extraction and clinical classification across 11 LLMs; (8) Joshua Harris et al., UK Health Security Agency, “HEALTHY LLMS? BENCHMARKING LLM KNOWLEDGE OF UK GOVERNMENT PUBLIC HEALTH INFORMATION” (arXiv:2505.06046v4, 2026) — PubHealthBench: 8,000+ questions across 687 government guidance documents, quantifying the MCQA-to-freeform degradation cliff and taxonomy of recommendation discrepancies.

The method, as extracted: read it

Ingested sources:

• library/arxiv-2606.21008v3 · view · item page · source
• library/arxiv-2608.18294v2 · view · item page · source
• library/arxiv-2403.16873v1 · view · item page · source
• library/arxiv-260327124v1 · view · item page
• library/arxiv-2203.02010v1 · view · item page · source
• library/arxiv-2505.10399v1 · view · item page · source
• library/arxiv-2405.14766v2 · view · item page · source
• library/arxiv-2505.06046v4 · view · item page · source

WireGen: wireframing from a written design intent

wiregen — declared by cat-harness — source held

Applies when. Designing a USER INTERFACE, for example a page layout, a navigation scheme or a visualiser, where the choice between candidate designs has to be reviewed and, when reviewers disagree, adjudicated. It is a design-generation and design-evaluation method. It does not choose between non-UI options (use the decision-analysis methodology) or grade evidence. It does not by itself settle a disagreement: that is adjudication.

Origin. Sidong Feng, Mingyue Yuan, Jieshan Chen, Zhenchang Xing and Chunyang Chen, “Designing with Language: Wireframing UI Design Intent with Generative Large Language Models”, arXiv:2312.07755v1 [cs.HC], 12 Dec 2023. Ingested in full at cat-harness/library/arxiv-2312.07755v1/. The source states no licence, and its licence could not be established (see check:source-licence).

The method, as extracted: read it

Ingested sources:

• library/arxiv-2312.07755v1 · view · item page · source

Zero trust architecture (NIST SP 800-207) — trust is never granted implicitly, it is evaluated per request at a decision point

zero-trust-architecture — declared by cat-harness — source held

Applies when. A participant is about to act on something another participant gave it, and the question is whether to trust it. That covers an access decision at a boundary, a hand-over between agents, a tool or a dependency graph reaching this platform, and the design of a check that decides any of these. It answers how trust is decided: never by location, per request, by policy, at a named decision and enforcement point. It is NOT a method for choosing among options (kepner-tregoe), for recording a decision (madr), for a recurring business rule (dmn), or for who is involved (raci). It is also not a threat-modelling or risk-scoring method: it says where trust is decided, not how likely an attack is.

Origin. NIST Special Publication 800-207, Zero Trust Architecture, by Scott Rose, Oliver Borchert, Stu Mitchell and Sean Connelly, August 2020, doi:10.6028/NIST.SP.800-207. The primary is held, in full, at library/nist-sp-800-207: 59 pages, 55 sections and all 12 figures described. The owner supplied it on 2026-10-07 after this container’s network policy denied both NIST hosts. It is a US-government work: the PDF states it “is not subject to copyright in the United States” (p. i). Every quotation below is from that copy, cited by section id.

The method, as extracted: read it

Ingested sources:

• library/nist-sp-800-207 · view · item page · source

Files in the graph that are not methodology nodes

None — every .md in the declared directories carries $schema: folio-methodology/v1.