Readings and judgment

ActionsArtifactVerification

Task and place

Sometimes the act text admits more than one understanding, and the modeler is not entitled to settle the dispute. Sometimes a premise needs a human decision no desk record can supply. This stage sits at the model step when competent readers diverge, and at the case step when a question waits on an authority. Its output is not an answer but an honest open point: each rival reading executable, each missing decision flagged by name.

The selection rule of a reading group is a contract of the answering process, not a property of the dispute. It answers three questions: who must choose a reading, when the choice is due, and what an unanswered dispute does to the rest of the document. The same rivalry can live under either contract; the profiles below run the same dispute and the same facts under both.

Inputs

• The disputed fragment and the rival understandings, each statable as rules.
• The authority map: who decides what in a live case — a hearing officer, a court, an agency — including who selects the governing reading.
• The two selection profiles quoted below, differing only in one selection line, all observed passing.
• The Northbridge appeal drill as the mandatory-profile worked example, and the running example as the negative case: no open reading, checked and stated as fact.

Actions

• Put each rival understanding in its own named reading, with the disputed fragment cited and an explicit status. A disputed rule left outside every reading is an ordinary package norm and applies in all cases — the quietest mistake on this page.
• Close the list of rivals in a group that names the alternatives and the selection rule. The case selects through the context axis, alongside the time axes — never as a fact, or the selection silently fails to act.
• Write the mandatory scenes for every group: each reading on identical facts with different answers — the selection alone flips the outcome — plus the unselected behavior the declared contract promises, plus one scene proving the undisputed part works unselected. A group under which the answer never changes is decoration.
• Choose the selection rule by the process contract, not by the kind of dispute. A rivalry between readings and a choice between editions can each live under either rule; what differs is what the process owes before it speaks.
• Vest authority-only facts in judgment relations naming their authority. Only that authority’s supply satisfies the premise; anything else leaves the question waiting with its flag up.
• Carry past decisions as precedents over explicit factors: the shared pattern plus the facts that tell cases apart, so a new case either follows or is distinguished back to the general rule.

Selection rule	Process contract	Empty selection
exactly_one	the process answers the disputed question only after the authorized selector names one reading	the dependent question is refused with INTERPRETATION_REQUIRED; independent questions still answer
any_of	the process may answer around the dispute without choosing	the disputed part stays silent (NEITHER, COMPUTED); the rest answers

Demanding one named choice never says which reading is legally right. Both readings stay executable, and the suite runs both — the flag on the unselected case is precisely the refusal to settle the dispute by silence. What would settle it is defaulting to one reading without a selector; neither contract does that.

Decisions

• Which disputes earn readings, and which close calls settle as decision records instead. A boundary both sides can live with is a record; an open rivalry is readings.
• The selection rule per group, with the process contract from the table and the name of the selector written next to it.
• Who selects a reading in a live case and on what ground — a process question the core leaves to the process around it.
• Which relations are judgment-vested and in which authority each one rests.

Artifact

The artifact is readings with their groups, judgment declarations with their authorities, and the scene table proving each group bites. The running example contributes an empty artifact here: a search over the package sources finds no reading declarations, no judgment relations, and no precedent blocks — every rule applies in every case, and that absence is itself the checked statement.

EAI example

The employee accident insurance package, version 0.1.0, language 0.2, zero dependencies, explicit local imports, pinned to edition EAI_EDITION with materialization PINNED_UNOFFICIAL_COPY — an Adilet API copy retrieved 2026-09-13, sha256 pinned, local copy kept in the package — models the duty, tariff, payout, and penalty branches of five articles with no open reading. The employer duty, the twenty-two tariff rows with premium base as insured sum times rate plus the minimum floor, payout for loss from thirty through one hundred percent, penalty as unpaid times 0.015 times days: none admits a rival understanding in the package, so none is wrapped in a reading. The inclusive thirty boundary was a close call, and it settled as a record (EAI-D2), not as readings — compare the EAI threshold decision. All seven scenarios answer with plain case facts and no selection context, which is exactly what a package with no dispute should do.

Pitfall

Four failures share one root: letting the dispute leak out of its container. A disputed rule left outside its reading always applies. A reading passed as a case fact instead of a context choice silently fails to act, and the dispute looks settled when nothing was selected. An any_of group where the process needed a mandatory stop launders the dispute into silence — the document answers around a question it was obliged to refuse. And an exactly_one group is sometimes cited as proof that one reading won: the contract demanded a choice, but the suite still runs both readings, and the winner is named by the selector, never by the declaration.

Verification

Both runs below were executed for this page with tool version law 0.1.0, semantics law.core/0.2, against the selection profiles — two synthetic packages identical except the selection line (mandatory declares exactly_one, open declares any_of), every scene asserting the same three facts on the same applicant:

Output

$ law test docs/handbook/files/fixtures/selection-profiles/mandatory
law test demo.handbook.selection_mandatory: world demo.handbook.selection_mandatory
  ok   [demo.handbook.selection_mandatory#authored] tests/selection.lawtest / narrow on full facts
  ok   [demo.handbook.selection_mandatory#authored] tests/selection.lawtest / broad on full facts
  ok   [demo.handbook.selection_mandatory#authored] tests/selection.lawtest / no selection — disputed flagged
  ok   [demo.handbook.selection_mandatory#authored] tests/selection.lawtest / no selection — independent answers
total: 4 checked, 4 passed, 0 failed, 0 not run; code 0
$ law test docs/handbook/files/fixtures/selection-profiles/open
law test demo.handbook.selection_open: world demo.handbook.selection_open
  ok   [demo.handbook.selection_open#authored] tests/selection.lawtest / narrow on full facts
  ok   [demo.handbook.selection_open#authored] tests/selection.lawtest / broad on full facts
  ok   [demo.handbook.selection_open#authored] tests/selection.lawtest / no selection — disputed silent
  ok   [demo.handbook.selection_open#authored] tests/selection.lawtest / no selection — independent answers
total: 4 checked, 4 passed, 0 failed, 0 not run; code 0

Each scene was inspected in the suite files; the setups and expectations read:

Scene	Setup, identical in all four scenes	Mandatory	Open
narrow on full facts	second vehicle, eligible, filed; Narrow selected	FALSE_ONLY	FALSE_ONLY
broad on full facts	same facts; Broad selected	TRUE_ONLY	TRUE_ONLY
no selection, disputed question	same facts; nothing selected	INTERPRETATION_REQUIRED, no truth value	NEITHER with COMPUTED
no selection, independent question	same facts; nothing selected; filing question	TRUE_ONLY with COMPUTED	TRUE_ONLY with COMPUTED

Three facts follow. First, the selection alone flips the answer: Narrow says false and Broad says true on byte-identical facts, under both contracts — the mandatory contract demanded a choice without picking a winner. Second, the contracts diverge exactly once, on the unselected disputed question: a flag under mandatory, silence under open. Third, the independent filing question answers in both profiles with nothing selected, so the mandatory flag refuses the dependent question, not the whole document.

The Northbridge appeal drill is the mandatory profile of the same dispute shape: its SecondVehicle group declares selection exactly_one over the Narrow and Broad readings, and its unselected scene expects INTERPRETATION_REQUIRED. Re-run with the pinned tool for this page, the suite passes seven of seven:

Output

$ law test packs/examples/language-demo/appeals
law test demo.northbridge.appeals: world demo.northbridge.appeals, demo.northbridge.calculations, demo.northbridge.permits, demo.northbridge.vocabulary
  ok   [demo.northbridge.appeals] tests/appeals.lawtest / broad reading
  ok   [demo.northbridge.appeals] tests/appeals.lawtest / narrow reading
  ok   [demo.northbridge.appeals] tests/appeals.lawtest / no selection — open question
  ok   [demo.northbridge.appeals] tests/appeals.lawtest / judgment not yet rendered
  ok   [demo.northbridge.appeals] tests/appeals.lawtest / judgment rendered
  ok   [demo.northbridge.appeals] tests/appeals.lawtest / precedent followed
  ok   [demo.northbridge.appeals] tests/appeals.lawtest / precedent distinguished
total: 7 checked, 7 passed, 0 failed, 0 not run; code 0

One honest limit of that drill: its Broad scene carries resident and vehicle facts that its Narrow scene omits, because Broad’s own premises require eligibility. The two scenes prove both readings run, but they do not isolate the selection — facts vary together with the choice. The profiles above exist to supply the missing isolation: identical facts, and the answer moves if and only if the selection moves.

Criterion: the rival readings answer identical facts differently under both contracts, the unselected case behaves as the declared contract promises, and the undisputed question answers unselected. The running example meets the mirror criterion: seven checked, seven passed with plain case facts and no selection context, observed with the same tool version.

Limits

Readings record a dispute; they never settle it and never say which side should win. The selecting authority and its grounds live in the process around the core, not in the package. A judgment relation is usable only where its authority has spoken, and a precedent binds only inside its factors pattern — outside that pattern it says nothing at all.

The flag’s reach is a fact about the pinned profile, observed above: the unselected mandatory group refuses the dependent question while the independent one answers COMPUTED. Descriptions that put the flag on every answer of the package do not describe this profile’s behavior; the independent scene is the standing counter-proof. Which reach is semantically right is a question for the language, not for this page — the handbook records what the pinned tool does and keeps the scene that would catch a change.

Next step

Continue with Evidence and admission, which decides what counts as established on the presented material.

Sources

• Northbridge course — the appeal article behind the seven-scene drill.
• Readings exercise — the drill for disputed constructions.
• Duty — the positions that appeals resolve disputes about.
• Language reference — syntax and semantics of every construct.
• Command line — the test command used above.
• Absence, negation, and conflict — the previous stage.