Software can pass every test and still describe itself incorrectly.
One repository calls something a permission. Another calls the same thing a capability. A third calls it policy, then lets the component being governed report whether the policy succeeded. Every sentence remains locally plausible. Together they produce a system nobody can reason about without first guessing which words survived the trip.
Organon exists for that failure.
It is a collection of instruments for making a body of work internally legible: an ontology for stabilizing meaning, an editorial grammar for carrying difficult ideas, and formal artifacts for discovering where apparently compatible definitions collapse under pressure. The name comes from the traditional title for Aristotle's collected works on logic: instruments of reasoning, not the doctrine those instruments may later examine.
Idolum builds systems in which agents act across boundaries of authority, evidence, and responsibility. Those systems make unusually expensive demands on language. A capability is not automatically a permission. An output is not automatically completion. A record is not automatically evidence. When a project blurs those distinctions, the implementation eventually inherits the confusion.
Organon provides a common surface from which project documentation can be reviewed. It asks:
- Which actor possesses a Capability, and which Principal grants Permission to use it?
- Who makes a Claim, what Witness is independent for that Claim, and what Rule admits the result as Evidence?
- Where does a term change meaning as it crosses a repository boundary?
- Which uncertainty remains explicit, and which has been hidden by fluent prose?
The objective is not terminological obedience. Project vocabulary should describe project mechanisms. Organon makes the correspondence inspectable.
Organon is not an anonymous standard. Its distinctions come from a particular history of building developer platforms, AI systems, and institutions around them. Who is Daniel? records that position, the experience behind it, and the limits of his authority over work that other people maintain.
This matters because an ontology can be internally consistent and still reflect the attention of its author. Declaring that attention makes the work easier to challenge. It also prevents a useful instrument from acquiring the suspicious voice of a tablet recovered from a mountain.
ontology/ binding prose, stable term registry, profiles, and Lean experiment
editorial/ long-form, short-form, and relational answer instruments
evals/ one source-pinned lifecycle for generation, judgment, improvement, and review
project-ontologies/ source-pinned candidate mappings for downstream projects
provenance/ essay, editorial, and term-level evidence lineage
proposals/ nonbinding promotion dossiers for quarantined vocabulary
reviews/ reusable review method and completed project audits
schemas/ machine-readable adoption contract
examples/ example adoption manifests
scripts/ repository, semantic, and adoption checks
The machine-readable structure manifest declares
the role and focused-hydration rule for every top-level region. Evaluation
history remains under evals/*/results/, while each suite's current.json
selects one current run without pretending that a passing model judgment is
human adoption. Proposal lifecycle separately
records whether each retained dossier remains quarantined, was partially
promoted, or was promoted.
Contributing to Organon defines the review burden for binding changes. In particular, a new term must survive a termhood challenge, dependency and collapse audits, comparison with its closest intellectual shadows, and proportionate formal testing before it can be called promotion-ready.
The binding artifact is Daniel's Ontology v0.18. It remains readable as one Markdown document; the machine-readable registry and formal artifacts check and challenge it without replacing it. Its term registry assigns stable organon:* identifiers, typed claims, and explicit dependencies. The generated prompt projection carries every primary term statement and registered commitment in a smaller, explicitly lossy form; its manifest records exact source hashes, coverage, and omissions. The hidden-bridge audit records why three notions became terms while three were reduced to metalanguage or existing Relations. The Ritual and Meaning dossier records why Flow absorbed recurrence, Ritual and Meaning survived termhood, and Ritual Residue did not. The changelog keeps earlier arguments and rejected formulations out of the active system.
For focused context hydration, the same generator accepts one or more terms and emits only their transitive dependency closure plus applicable commitments:
python3 scripts/build-ontology-prompt.py \
--term organon:Intelligence \
--term organon:OperativeKnowledge \
--output /tmp/organon-knowledge-prompt.md \
--manifest /tmp/organon-knowledge-prompt-manifest.jsonTerms under quarantine are developed through proposal pull requests. A proposal may preserve candidate definitions, hypotheses, evidence requirements, and reasons for refusal without changing the binding ontology merely by existing.
Capitalization is not adoption. A downstream repository adopts Organon by naming a version and profiles, declaring governed paths, and explicitly mapping its local vocabulary through an adoption manifest. The example manifest is executable documentation.
The Long-Form Editorial Grammar describes how a reader comes to need, receive, and carry a difficult idea. The canonical Short Form governs sentence-scale delivery once that idea has been earned. The proposed Essay-Answer Form governs a different relation: answering a particular question through a restrained, evidenced hypothesis about its reader. None may silently redefine what the ontology says exists.
The evaluation methodology governs every retained DSPy experiment: pin the source, preflight deterministic facts, generate when appropriate, judge in non-overlapping layers, turn failures into bounded improvement work, compare exact candidates, and leave promotion to a human maintainer. The essay-question suite applies it to reader answers without treating generated interlocutor hypotheses as facts about actual readers.
The project ontologies apply the same discipline to Engram and Kenogram. Their two-stage suite checks Organon dependency closure before open-source documentation quality and emits explicit promotion gates. The editorial-artifact suite carries the same lifecycle into README and long-form candidates. None of these outputs becomes an adoption claim by passing an automated gate.
The ontology's essay evidence is indexed through the public Essay Corpus. Private drafts and review artifacts are declared through Editorial Provenance. Term Provenance records the lineage claimed for every stable term without presenting lineage as proof of truth.
The Lean formalization prices decisions that prose can hide. It encodes selected high-risk regions, including classical A3, Direction-indexed Transformations, explicit feeding between States, executable Specifications, contextual Capability, Order-indexed Permission, scoped Witness independence, a nontrivial Boundary model, discriminating Operationalization, finite World and Substrate witnesses, separate Truth, Trust, and Alignment countermodels, adaptive Intelligence plus operative-knowledge reconstruction, and finite epistemic, moral, sovereignty, and valuation profiles.
The formal directory also contains an Absence-free OrganonCore reduct and complete registry audit. The current downstream Lean classifiers and finite witnesses compile against that reduct without importing the Absence/Presence extension. A falsification-first seam preserves challenge classifiers for Presence, Missingness, Persistence, and Entity; Reality has one pending representation decision, and 103 binding terms remain explicitly unknown rather than inheriting a proof from similar vocabulary.
The nonbinding candidate algebra experiment applies the same falsification-first posture one level above individual terms. It translates nine load-bearing definitions into typed witness normal forms, generates one-step structural mutations and extensional query countermodels, checks six candidate-discipline annotations, and tests the unchanged taxonomy on held-out consciousness, valuation, sovereignty, Ritual, and Meaning cases. The taxonomy is an experimental review instrument, not an executable law engine or a replacement for the binding ontology or Lean shadow.
Lean is not canonical. Compilation proves that the declarations elaborate under the pinned compiler. It does not prove that the encoding exhausted the prose, that the prose exhausted Reality, or that the compiler has become a metaphysician. The formalization decisions and build receipt preserve that boundary.
Begin with the project. Identify its actual actors, boundaries, transformations, records, and authority paths. Then use the review template to classify each correspondence as exact, a refinement, a conflict, or unmapped. A defect may belong to the project, to Organon, or to the proposed mapping.
Repositories can adopt one or more profiles from profiles.json. The checker expands every selected profile through the registry's dependency closure and validates local mappings:
python3 scripts/check-adoption.py path/to/organon-adoption.json --repo-root path/to/repositoryFrom the repository root:
python3 scripts/check-links.py
python3 scripts/check-semantics.py
python3 scripts/build-ontology-prompt.py --check
python3 scripts/build-algebra.py --check
python3 scripts/build-reduction-audit.py --check
python3 scripts/build-positive-calculus.py --check
python3 scripts/check-proposals.py
python3 scripts/check-formal-receipt.py
python3 scripts/check-adoption.py examples/organon-adoption.json --repo-root .
(cd ontology/formal && lake build && lake exe ontology_check)The link check keeps public Markdown inside the portable Obsidian/GitHub intersection documented in Markdown Policy. The semantic check verifies stable markers, claim types, dependency order, governed projections, and term-level provenance. The adoption check verifies the contract a downstream repository would make.
Organon v0.18 is provisional and binding only where explicitly adopted. The ontology is not complete, universally true, or immune to revision. It is a versioned promise that named distinctions will not quietly change halfway through an argument.