Actor
A person, organization, or other participant in a source.
Example: A ministry announces a consultation.
Review question: Similar names do not establish identity; roles can change over time.
Section pages
Record model · revised September 29, 2026
Define the distinctions policy analysis depends on.
The Agentic Conflict Ontology (ACO) defines actors, claims, commitments and change. This proposed vocabulary gives graph records an explicit meaning, so an interpretation can be questioned, reused and revised.
Research and architecture in development. This page explains our design direction. It is not a list of completed PRAXIS capabilities. See the documented product capabilities and open work.
A well-shaped claim can still be wrong.
C04 has an ID, a claim type and a source relation. Those fields can be structurally valid while “January is adopted” contradicts S01 §1: no rule has been adopted. The reviewer rejects that interpretation using E01, without deleting the attributed record. If a domain extension adds an adopted-instrument requirement in v2, old records retain their v1 definition; a reviewed migration marks missing instruments unresolved and invalidates affected current approvals. It must not promote proposal dates into effect. Independent annotation and downstream benchmark results remain unpublished.
Practitioner summary · the question, evidence and limits
- Practical question
- How do reviewers preserve the difference between a statement, an event and an inferred motive?
- Proposed mechanism
- Give actors, claims, commitments and temporal relationships explicit definitions, source links and review questions.
- Current evidence
- The examples teach a vocabulary; they do not establish annotation agreement or complete domain coverage.
- Strongest limitation
- A precise-looking graph can still encode a mistaken interpretation. Expert agreement needs testing on real source sets.
Revision responsibility: TACITUS Research. Summary prepared September 30, 2026; paper dates remain separate. Independent peer review is not claimed.
Eight starting points, not a complete theory of human behavior.
A vocabulary helps two people notice when they are using the same word for different things. Here, an actor making a claim, someone inferring an interest, and a party accepting a commitment are separate records. The boundaries are useful only if annotators can apply them consistently to actual sources.
Claim
A statement attributed to a speaker or source.
Example: The ministry says the measure will reduce costs.
Review question: Record attribution and modality without treating the statement as established fact.
Interest
An expressed objective or a proposed interpretation of motivation.
Example: A trade body seeks a longer transition period.
Review question: An inferred motive needs its supporting observations and plausible alternatives.
Constraint
A condition that limits an action or choice.
Example: Implementation depends on legislative approval.
Review question: Distinguish legal requirements, practical limits, and someone’s asserted constraint.
Leverage
A proposed means by which an actor can affect another’s choices.
Example: A supplier can delay a critical shipment.
Review question: A claimed capability is not proof of willingness, control, or causal effect.
Commitment
An undertaking with an actor, scope, conditions, and relevant time.
Example: The agency undertakes to publish a revision after consultation.
Review question: A forecast, aspiration, or assignment by a third party may not be an accepted obligation.
Event
An occurrence as described or evidenced by sources.
Example: The committee voted on September 12.
Review question: Multiple reports can refer to the same event or disagree about whether it occurred.
Narrative
A recurring framing that connects selected claims or events.
Example: A reform is framed as either modernization or a loss of accountability.
Review question: Shared language alone does not establish coordination or malicious intent.
These are working definitions for the research program. ACO is not an externally standardized ontology. Domains may require other concepts, and a useful model must allow a reviewer to mark an interpretation as uncertain or outside the vocabulary.
The edge makes an analytical claim too.
A graph can look precise while encoding a weak interpretation. Linking two events with a causal edge is a much stronger statement than putting them in date order. Relationship labels therefore need definitions, permitted endpoints, source references, and a way to record disagreement about the link itself.
| Relationship | Connects | Question | Failure to watch for |
|---|---|---|---|
| ASSERTS | Actor → Claim | Who is responsible for the statement? | Do not assign the journalist’s attribution to the quoted speaker without checking the passage. |
| SUPPORTS / CHALLENGES | Evidence or Claim → Claim | What bears on the assessment? | Support can be partial; two sources may repeat the same original report. |
| PRECEDES | Event → Event | What happened earlier? | Ordering does not establish that one event caused the other. |
| CAUSES (proposed) | Event → Event | What mechanism and conditions are alleged? | Store causal claims as attributed hypotheses unless the evidentiary standard justifies more. |
| CONDITIONAL_ON | Commitment → Condition | When does an obligation apply? | Missing conditions can turn a qualified promise into an unconditional one. |
| SUPERSEDES | Record version → Record version | What changed, and when was the change recorded? | Retain the earlier record for historical questions. |
Repeated reports should not automatically become independent support. A provenance path should distinguish a direct observation from a quotation, a repost, or an inference. W3C PROV-O provides established terms for derivation and attribution; adopting those terms still requires reliable capture and application-specific interpretation.
When it happened and when we learned it are different questions.
Suppose a restriction takes effect on September 12, appears in a notice on September 14, and enters an analyst's record on September 16. A historical assessment made on September 13 must not silently inherit the later notice. Event or valid time, source publication time, and system record time should remain distinguishable.
Valid time
September 12
Restriction takes effect
Source time
September 14
Notice is published
Record time
September 16
Record enters the case
Dates may be incomplete or disputed. Preserve the original expression, its timezone or calendar when known, and the reason for any normalization. An unknown end date should not be replaced by an invented point estimate to satisfy a field requirement.
The interval explorer below illustrates relations between bounded intervals. It does not infer dates or resolve uncertainty from documents.
KAIROS · Allen-13 interactive
Move two intervals. Inspect their relation.
A
B
What it says
A begins first, ends inside B.
A = [14, 46]
B = [34, 72]
relation = OVERLAPS
B = [34, 72]
relation = OVERLAPS
Drag the bars or use the labeled sliders with arrow keys. This illustration compares two bounded intervals on an arbitrary scale, using a one-unit tolerance for matching endpoints. It does not extract dates from sources or demonstrate an integrated KAIROS service.
A commitment needs scope, conditions, and evidence of change.
“We intend to deliver in June,” “we will deliver if approval arrives,” and “delivery was completed” describe different things. A proposed commitment record should preserve who undertook what, to whom, under which conditions, and the source of each interpretation.
Fictional case · six entries
Preserve the undertaking and the dispute.
This proposed record keeps a commitment, an alleged breach, and a denial distinguishable. Entries retain their attributed source and date; the sequence does not settle whether a breach occurred.
- PROPOSED12 Mar
Nordalia delegate proposes a 5-km halt on offensive operations for 72 hours.
- ACKNOWLEDGED12 Mar
DRM delegate acknowledges proposal in plenary session; transcript p. 14.
- COMMITTED13 Mar
Joint communiqué codifies the 5-km halt with verification by international observers.
- CONTESTEDAudit trail14 Mar
Nordalia files breach complaint citing observed movement near 4.2-km line.
- DENIED14 Mar
DRM denies breach; characterizes movement as administrative reposition.
- REAFFIRMED15 Mar
Both parties reaffirm dialogue commitment; halt extended 48h pending verification.
The ledger is a fictional design example. Its transitions are explanatory, not evidence that the product automatically determines whether a real obligation has been fulfilled.
A later statement can amend the scope, allege a breach, or dispute whether there was an undertaking at all. Those reports should remain attributable. A validator can require evidence for a proposed transition; deciding whether the evidence establishes fulfillment is a separate analytical or legal question.
Share a small core. Let domains supply their own tests.
Regulatory work
A proposed RegulatoryCommitment might add jurisdiction, rule reference, consultation window, and commencement conditions. A legal reviewer defines how those fields should be interpreted.
Commercial contracts
A ServiceLevelCommitment might add a measurement interval, threshold, exclusions, and remedy. A monthly uptime percentage means little without the denominator and exception rules.
Mediation
A WorkAllocationCommitment might distinguish requested, acknowledged, and accepted scope. A third-party assignment should not become a promise by the assigned person.
Information environments
A Narrative record might link framings, attributed speakers, channels, and dates. Similarity is evidence of resemblance; coordination requires additional evidence.
Our proposed extension process records the parent type, definitions, allowed values, examples, counterexamples, and a version. A change that alters meaning needs an explicit migration decision. Passing a schema check is not enough to establish that two teams use a field in the same way.
Schema validity is one layer of review.
SHACL formalizes validation of RDF graphs against declared conditions. JSON Schema can likewise check the shape of JSON records. These are useful engineering checks: required fields, datatypes, cardinality, and references can be inspected consistently.
They do not settle whether two names identify the same person, whether a quotation is represented fairly, or whether a causal explanation is justified. We therefore separate structural validity, source fidelity, and interpretive review. A record may pass the first while failing the other two.
To evaluate this vocabulary, independent annotators should label the same held-out passages, explain disagreements, and test whether distinctions improve a downstream task. Add negative examples: predictions that resemble promises, denials that repeat the denied claim, and reported dates that differ from event dates. Persistent disagreement may require revising the ontology rather than simply retraining an extractor.
Standards to build with, questions still to test.
Primary standards reviewed September 29, 2026. Referencing them does not establish TACITUS conformance.
- Shapes Constraint Language (SHACL)
W3C Recommendation · 2017
Defines validation of RDF graphs against declared conditions and a vocabulary for validation reports. Conformance to shapes does not establish the truth of the data or the adequacy of the shapes.
- PROV-O: The PROV Ontology
W3C Recommendation · 2013
Provides a vocabulary for representing and exchanging provenance across systems. A provenance record describes derivation and attribution; the record itself still needs trustworthy capture and review.
Inspect the ontology repository for its published artifacts and version history. This page explains the evolving model; repository files determine the exact schema for a particular experiment.