Method

How Default Liberty Is Built

A source-controlled research architecture for developing a source-first theory.

NONCANONICAL PRESENTATION

Publication and technical presentation, with authority effect none. This page describes how the project works. It is not part of the theory and changes nothing in it.

Default Liberty asks a demanding question of normative authority: where did this constraint come from, what activates it, and how far does it reach? The project applies the same question to its own development. Claims are separated by status, their dependencies are preserved, proposed changes cannot promote themselves, and retrieval or software output cannot manufacture authority.

Canonical stateWhat the project currently accepts
Derivational structureThe chain and the Fidelity Gate that every proposition must state
ResearchCandidates, objections, stress tests
CRI / CodexRetrieval, indexing, accepted contextInforms. No authority effect.
Evidence reviewIndependent, one checkpoint at a time
Accept
Explicit promotionOptional. A separate, hash-checked act
Canonical state
RejectSupersedeReturn for workThe path ends here. The canon is unchanged.

Retrieval, model output, graph position, similarity, and technical validation have no authority effect.

In words: the canonical state controls the derivational structure. Research works within that structure and produces candidates; the retrieval layer informs the work and carries no authority. Both lead only to Evidence review. Review ends in one of four dispositions. Three of them leave the canon unchanged. Acceptance may authorize an explicit promotion; it does not perform one, and some accepted work is never promoted. Only a promotion changes the canonical state.

The reason for building it this way is the theory itself. A project that tells institutions to show the source of every power would be poorly placed if its own doctrine could grow by repetition, by a persuasive draft, or by whatever a search happened to return. So the working method is designed to submit to the discipline it argues for.

Canon is the controlling state

Default Liberty maintains a defined canonical state. Accepted propositions are distinguished from drafts, research candidates, external scholarship, retrieval results, historical records, and software output.

A proposition does not become part of Default Liberty merely because it has been written, retrieved, indexed, ranked highly, connected in a graph, or generated by a model. Canonical status is explicit: something either has been accepted and promoted, or it has not.

When two records disagree, the higher one controls:

  1. Project state
  2. Canonical documents
  3. Codex decisions
  4. Accepted review records
  5. Research and history

Candidates and pending proposals sit below all of these. What the canon currently contains is set out on the Canon page.

The dependency structure is explicit

Every proposition in the theory depends on something: an earlier position, a source, the facts that trigger it. The project keeps those dependencies in a fixed order, so that a conclusion can be followed back to where it came from and a missing step shows as a gap.

Each stage must be established separately. Validity at one stage is not authority at the next.

  1. SubjecthoodWho is the normative subject?
  2. Prior positionWhat valid position did they already hold?
  3. Source or mechanismWhat accepted source or valid mechanism is relied on, and what facts trigger it?
  4. Hohfeldian positionWhat exact claim, duty, liberty, power, liability, immunity or disability results, and how far does it reach?
  5. 5ViolationWas that position actually violated?
  6. 6LiabilityWhat separately supports any secondary liability?
  7. 7JurisdictionWhat establishes authority over this person and this matter?
  8. 8ProcedureWhat procedure is validly authorized?
  9. 9RemedyWhat remedial source supports the remedy requested?
  10. 10EnforcementWhat separate source supports enforcement?

If a link fails, the derivation stops there. The prior position remains.

  • violation is not liability
  • liability is not jurisdiction
  • jurisdiction is not remedy
  • remedy is not enforcement

Before any proposed rule, mechanism or power can go to review, it has to be stated in the same nine parts and in the same order. The project calls this the Fidelity Gate. Nothing downstream is considered until the primary position has been established and bounded.

  1. Prior position. The baseline liberty, or the valid position already in place.
  2. Alteration. Exactly what is proposed to change.
  3. Source. The accepted source or valid mechanism relied on.
  4. Trigger facts. The facts that must be proven for the source to operate.
  5. Hohfeldian result. The precise relation produced, and between whom.
  6. Scope ceiling. The furthest the source and facts can carry it.
  7. No-smuggling audit. A check that no unaccepted premise is doing hidden work.
  8. Residual liberty. What remains unaltered.
  9. Downstream. Only then: violation, liability, jurisdiction, procedure, remedy and enforcement, each on its own source.

This structure is also the basis of a planned formal graph: a directed record in which each position points to what it follows from. The project has accepted that graph as a research roadmap. It has not been built, and no schema has been fixed. Today the dependencies live in the chain and the gate above.

Research cannot promote itself

Research is the exploratory lane. It can derive candidates, identify gaps, construct objections, test edge cases, compare external theories, and expose possible defects. It cannot decide that its own conclusions have become part of Default Liberty.

ResearchCandidate or open questionEvidence

Researchno pathCanon

Unresolved work stays visibly unresolved. Every proposition carries one of four statuses, and none changes silently:

AcceptedCandidateOpenSuperseded

The open and unsupported questions are published as such on the Research page.

Adoption is a separate act

A second lane, called Evidence, reviews what Research proposes. It is invoked deliberately, takes one checkpoint at a time, and is kept apart from the lane that wrote the proposal.

  1. Candidate. Research publishes a fixed checkpoint and stops.
  2. Inspect. Evidence claims one checkpoint and reads it as written.
  3. Verify. The sources it cites, and what it says they establish, are checked independently.
  4. Adjudicate. The derivation, objections, hard cases and status labels are tested, and one disposition is recorded.
  5. Disposition. Accept, reject, supersede, or return for work.
  6. Promotion. Only after acceptance, and only as a separate explicit act, where authorized.

Completion of research is not canonical adoption. Acceptance authorizes a change where appropriate; it does not make the change silently. A proposition that has been accepted but not yet promoted is authorized for the canon and is not yet in it.

Retrieval assists reasoning. It does not establish authority.

The project's corpus is large and keeps changing. To work with it, Default Liberty uses a dedicated retrieval layer, called CRI, together with a Codex: a structured record of accepted decisions and of where candidates and reviews stand. Between them they find current project material, accepted decisions, derivational context and relevant earlier work, so that the research does not depend on anyone's memory of it.

CRI is retrieval infrastructure. It has no normative authority. Similarity scores, ranking, graph reachability, currentness, memory matches, and model-generated summaries cannot establish or promote Default Liberty doctrine. What retrieval returns is material to read. Where it surfaces a conflict or a superseded position, the conflict stays visible.

The retrieval index is treated as derived state. When controlling project material changes, the index can become stale, and it must be rebuilt and resealed before it is again treated as the preferred retrieval route. Until then the work falls back to reading the controlling documents directly, in the same order of authority.

The history is preserved

A later conclusion should not erase the path by which the project reached it. Research checkpoints are kept as they were published. Review decisions are kept. Superseded work stays in the record, marked as superseded, and is not rewritten into having always been correct. Where two records conflict, the conflict remains inspectable.

The governing documents are kept distinct from research and history, so that a preserved draft is never mistaken for the present state of the theory. The same convention appears on this site: each published text gives its version and the hash of its source file.

Technical detail

Each research checkpoint is identified by a SHA-256 hash. Review produces receipts at three points: when a checkpoint is claimed, when a decision is completed, and when a change is promoted. Claim and completion receipts carry the checkpoint's hash, and completion also binds the hash of the decision.

Promotion is limited to the existing governing documents. Before it writes, it checks the checkpoint's identity and hash, the acceptance receipt, the exact decision and its hash, and the expected current hash of every document it will change. A mismatch at any point stops the whole promotion, and a failed write rolls back the set.

Retrieval indexes, caches and currentness records can be rebuilt from the project's documents at any time. The review queue itself is excluded from ordinary indexing.

AI assists the work. It does not govern the theory.

AI systems assist with research, retrieval, adversarial testing, drafting, formalization, and technical implementation. They do not possess authority to alter Default Liberty's canonical commitments.

A model can propose an argument, identify an inconsistency, retrieve a source, generate a stress test, or formalize a derivation. Those outputs remain inputs to the research process. Proposed changes remain proposed until they pass the project's explicit review and promotion rules.

Model output is evidence or analysis to inspect, not authority to obey.

Nothing in review runs on its own. No checkpoint is claimed, adjudicated, completed or promoted automatically. Foundational revision is not delegated to the AI or provider. It requires the project's valid Track B constitutional-change procedure, including the required explicit post-ACCEPT Constitutional Principal ratification.

External scholarship tests the theory; it does not supply missing foundations

Substantive research follows a fixed order.

  1. Frame the problem using the accepted Default Liberty sources.
  2. Attempt the derivation internally.
  3. Where the derivation fails, preserve residual liberty or identify the gap.
  4. Stress-test the result.
  5. Only then consult external scholarship: for objections, analogues, counterexamples, terminology, consequences and novelty.

Citation can challenge a Default Liberty derivation. It cannot become a Default Liberty source merely because the cited authority is prestigious, conventional, judicial, statutory, or widely accepted.

The project relies heavily on scholarship, and its own account of its debts is long. The distinction is between two kinds of weight. A scholar's argument can show that a derivation is wrong, that a premise is contested, or that an idea is not new, and the project is bound to take that seriously. The argument cannot stand in for a source the theory has not accepted. The objections recorded against the foundations are listed under Research.

The method is becoming executable

The next stage of the project is to make this structure checkable by machine. The first two steps below are in place. The rest are planned, and none of them has been built.

  1. Human-readable theory

    In place

    The articles, guide and governing documents.

  2. Controlled canon

    In place

    Explicit status, a fixed order of authority, and separate review.

  3. Derivational graph

    Planned

    Each position recorded with what it follows from. Accepted as a research roadmap.

  4. Formal ledger

    Planned

    A specification able to represent a derivation without hiding or manufacturing a step.

  5. Self-reconstruction

    Planned

    Rebuilding derivations the theory already accepts, and classifying any that fail.

  6. External reconstruction audits

    Planned

    The same test applied to frameworks from outside the project.

  7. Derivation engine

    Planned

    Deterministic checks over the validated structure.

  8. Default Liberty Derivation Lab

    Planned

    Public inspection of real derivations.

The aim is not to ask a language model what the law or morality should be. The aim is to make derivational claims inspectable: identify the prior position, locate the claimed source, test the trigger, follow the alterations, expose missing steps, and show where the derivation stops.

Explore the Derivation Lab