The Canons as a Decision Procedure

The claim

A list of interpretive canons is usually treated as a set of arguments. An interpreter picks the ones that support a conclusion and cites them. Under that reading, the list has no control flow. It is a library, not a program.

Ross Brodskiy's Canons of Constitutional Interpretation of Gibsonia invite a different reading. The instrument is closed, layered, and ordered, and it binds every interpreter regardless of institution. That is enough structure to ask a precise question: to what extent do the canons define a decision procedure, meaning a fixed specification that maps inputs to outputs through operations stated in advance?

My answer: they define a partial decision procedure. They specify admissible inputs, orderings over conflicting provisions, defeaters, gates, and defaults. They do not compute the predicates those operations consume, and three canons resist formalization in ways worth stating exactly. Canon numbers below follow the Institute edition, which runs from 1 to 23.

Definitions

• Provision p: a unit of enacted text in a binding instrument.
• Candidate reading r: a proposed meaning of p.
• Canon: a rule that, given p, r, the facts, and the admitted sources, supports r, defeats r, or activates another canon.
• Predicate: a classification the canons require but do not define operationally, such as "ordinary public meaning," "genuinely silent," or "least restrictive means reasonably available."
• Output: one of three values. Adopt r; reject r; or abstain, which returns the question to the empowered institution with the unresolved predicate named.

The third output matters. A procedure that can only adopt or reject will manufacture certainty whenever its inputs run out.

Inputs

Fontes Limitati (canon 21) is an input specification. It ranks four authoritative sources: the text read through the canons; the structure of the instrument; binding Gibsonian precedent consistent with the canons; and contemporaneous public meaning at enactment, admitted only "where plain meaning is unclear." Foreign law, comparative jurisprudence, academic commentary and moral philosophy may be consulted for illumination but carry no binding authority.

Formally, this is an allowlist with one conditional entry. The check it implies is mechanical: in any derivation, every supporting edge must terminate in an allowlisted source. Illumination-only material may appear in the record as context, but it cannot be the ground of a conclusion. Clarus Vincit Propositum (canon 20) adds a guard on a second class of input. Purpose, legislative history and drafters' intentions may illuminate ambiguity; they may not override clear text. So purpose-type inputs are admissible only on the branch where the textual layers have returned "ambiguous."

There is also a temporal index. Verba Suprema (canon 1) and Fixatio Significationis (canon 14) anchor meaning at enactment, not at application. The input is therefore not "the text" but "the text as of its enactment date." An implementation needs a dated corpus, and resolving a term against the current state of the language is a type error.

Precedence

The canons contain two different kinds of ordering, and they should not be conflated.

Orderings over provisions. Ordo Praevalet (canon 12) applies only where conflict "cannot be reconciled," and ranks four categories: fundamental individual rights; structural constraints on government; enumerated government powers; administrative and procedural provisions. Higher displaces lower. Within a category, Specialis Derogat Generali (canon 7) controls. That is a two-key lexicographic sort: category rank first, specificity second. Canon 7 carries its own scope condition, "within the same instrument and on the same subject," which is itself an input the procedure must establish before the second key applies.

The preamble adds a third ordering, over instruments: the Constitution, statutes, charters, treaties, and other binding instruments, "in that order of precedence." The text does not say how the instrument ordering composes with the category ordering when they disagree, for example a rights provision in a statute against an administrative provision in the Constitution. A formalization must choose a composition. It should record the choice as its own, not attribute it to the text.

Orderings over canons. Here the instrument does something more interesting. It rarely states that one canon beats another. Instead, most canons carry a guard: an activation condition that is satisfied only when an earlier stage has failed to decide.

• Libertas Praesumitur (canon 8) fires "when provisions bearing on individual rights are ambiguous."
• Silentium Pro Libertate (canon 10) fires "where the text is genuinely silent."
• Ordo Praevalet (canon 12) fires only after Concordia Praeferenda (canon 11) fails to reconcile.
• The fourth source in canon 21 is admitted only "where plain meaning is unclear."
• Absurditas Vitanda (canon 23) requires a "reasoned explanation of why no other canon resolves the matter," which makes it last by construction.

This is a cascade of partial functions, each defined on the residue its predecessors leave. It is also why Llewellyn's dueling-canons objection lands differently here: two canons can only duel if both guards are satisfied at once, and the guards are designed to make that rare.

Whether the eight layers themselves form a strict lexical priority is a separate question. The text orders the layers on the page, and the guards produce much of the effect of lexical order, but no canon says "earlier layers prevail." Some commentary reads the layer sequence as a superiority relation. That is a reasonable encoding. It is an encoding.

Defeaters, gates and defaults

Borrowing the vocabulary of defeasible logic, the canons sort into four formal roles.

Two details make the gates unusually checkable. First, each gate states its exclusions as well as its conditions. Canon 9 says vague standards such as "reasonable," "necessary," or "in the public interest" do not, standing alone, satisfy the textual basis requirement. Canon 17 says disagreement with an outcome is not manifest error. Canon 23 excludes unwise policy, undesirable consequences and moral discomfort. An exclusion list is a type constraint on an input, and type constraints are cheap to enforce.

Second, the gates are conjunctive. A restriction of a recognized right that fails any one of the five prongs is invalid. A derivation that addresses four prongs has not reached a conclusion. It is incomplete, and incompleteness is detectable without deciding anything about liberty.

Abstention

Where should the procedure abstain? The canons give a sharper answer than one might expect, because they distinguish two kinds of gap.

Rule-level gaps on rights questions are closed by defaults. If the text is genuinely silent on a question affecting individual rights, canon 10 resolves the silence in favor of liberty. If a rights provision is ambiguous, canon 8 favors the individual. If temporal scope is ambiguous, canon 16 resolves against retroactivity. In these cases the correct output is not abstain. It is a directed default. Silence is not "unknown"; it is "liberty." That is a deliberate design choice, and it makes the procedure total over a large class of questions that would otherwise stall.

Predicate-level gaps are not closed. The procedure consumes classifications it cannot compute: ordinary public meaning, legitimate public purpose, narrow tailoring, least restrictive means, proportionality, manifest error, whether the invalid part is "so central" that severance fails. When one of these cannot be established with warrant from admitted sources, the honest output is abstain, with the missing predicate named. Manufacturing a value to avoid abstaining converts a judgment gap into a silent assumption, which is the failure the Institute's Four Laws treat as foundational: no output with unwarranted confidence.

One subtlety connects the two. "Genuinely silent" is itself a predicate. The default in canon 10 is only as good as the finding that the textual and structural layers have actually run out. An implementation that treats "I found nothing" as "the text is silent" has substituted a retrieval failure for a legal conclusion.

What resists formalization

Three canons resist, each for a different reason.

Prohibitio Manipulationis (canon 18). A reading is pretextual "when its stated textual basis is a cover for a predetermined result." Predetermined is a property of how the decision was produced, not of the record it produced. Two opinions with identical text can differ in pretextuality. The property is therefore not decidable from the output. What a system can compute are symptoms: a guarded canon invoked without its guard satisfied (purpose relied on while the text is declared clear); a canon applied one way in one matter and the opposite way in a matched matter where only the outcome's beneficiary differs; counter-readings present in the sources but absent from the record. These are useful flags. They are not a pretext detector.

Bona Fide Interpretatio (canon 22). Good faith requires giving the text its "fair meaning," neither stretched nor shrunk, and presumes the drafters meant what they said. This is a standard about the interpreter's disposition. A system can check procedural markers that good faith would produce: textual basis stated, counter-readings addressed, ambiguity disclosed, sources cited. Satisfying the markers is necessary. It is not sufficient.

Absurditas Vitanda (canon 23). The trigger is partly formal. "Logically impossible" and "self-contradictory within the instrument itself" can, in principle, be tested as inconsistency in a formal model of the instrument, but only if that model exists and is faithful to the text. "Physically or mathematically absurd" requires world knowledge. The remedy is the harder part: "the nearest non-absurd reading consistent with the text" presupposes a distance measure over readings. The canon does not supply one. So the trigger is partially formalizable and the remedy is not.

The general lesson is that formalization relocates judgment into predicates. That is acceptable only if every predicate value in a derivation carries provenance: the text, evidence, finding or model output that supplied it.

The version-control reading

The companion essay Version Control for Law maps the canons onto a Git model: enacted law as the main branch, amendments as tagged merges, opinions as signed commits, precedent as commit history, overruling as a gated force push, dissents as preserved branches, and non-Gibsonian sources as disallowed dependencies. Read against the decision procedure, the mapping identifies which checks can run automatically at commit time.

• Protected main. Under Mutatio per Emendationem (canon 15), interpretation may apply text to new facts but may not change what it means. Meaning changes enter only as amendment merges with version tags. An interpretation commit that alters the meaning of a term is rejected at the branch rule, not argued about afterward.
• Signed commits. In the essay's model, a reasoned opinion states its textual basis, addresses counter-readings, discloses ambiguities, and cites sources. That is a schema. A commit missing a required field can be refused mechanically.
• Force-push gate. Canon 17 requires manifest error under the canons, fair treatment of reliance, and reasoning under the canons. The essay's gate asks the overruling opinion to name the violated canon, supply the canonical re-reading, address reliance interests, and reason textually. A pre-receive hook can verify that the named canon belongs to the closed set of twenty-three, that a re-reading is present, that a reliance section exists, and that no supporting edge runs to a disallowed source. It cannot verify that the error is manifest. That remains a predicate.
• Preserved branches. Dissensus Servandus (canon 19) forbids deleting reasoned dissents from the record. In its completed form, a preserved dissent has no binding force of its own and may be revived only when the overruling conditions of canon 17 are met. In Git terms: the branch is never deleted, and it can reach main only through the same protected gate as any other force push. There is no fast-forward path for a dissent.
• Dependency allowlist. Canon 21 is the dependency manifest.

This is where the decision-procedure reading pays off. The gates check form: fields present, sources admitted, guards satisfied, conjunctions complete. The predicates carry substance: meaning, purpose, proportion, error, good faith. A signature proves who committed and that the reasoning is attached. It does not prove the reasoning is honest.

What the procedure buys

The canons do not make constitutional interpretation automatic. They make much of it checkable. Source validity, guard satisfaction, priority compliance, gate completeness and record completeness can all be verified without deciding a single contested predicate. Good faith, pretext and the nearest non-absurd reading cannot.

That is a better result than it sounds. A procedure whose holes are named, located and logged can be audited, and disagreement about it can be directed to the right place: this input, this predicate, this ordering choice. An oracle whose holes are hidden cannot be audited at all.