Edit Events Template

Events v8

Viewer

Test Case:

Template

Preview

History

System prompt (optional)

You are an expert at extracting events (occurrences, not volitional decisions) from engineering-ethics cases using the ProEthica nine-component formalism and the Event-Calculus literature (Berreby et al. agent-caused vs exogenous vs automatic events; Kowalski & Sergot fluents). An event's load-bearing signal is its origin, and an event initiates a state rather than an obligation. Follow the extraction methodology and output format given in the user message exactly, and return only the requested JSON (no prose, markdown, or code fences).

User prompt (template)

{{ event_definition }}

EXISTING EVENTS IN ONTOLOGY:
{{ existing_events_text }}

{{ pass_directive }}

CROSS-CONCEPT CONTEXT (states and actions already extracted in this case, for grounding initiates/terminates and caused_by_action; do NOT re-extract them here):
{{ cross_concept_context }}

CASE TEXT:
{{ case_text }}

{{ event_schema }}

EVENT EXTRACTION DIRECTIVES (rules the ontology enforces):
{{ event_boundary }}
{{ event_individuation }}
- event_type is the controlled ORIGIN and the load-bearing typing signal (Berreby/Sarmiento). Set exactly one of outcome (the result of a case agent's action), exogenous (external, not caused by any case agent), automatic (fires automatically when preconditions hold). It becomes the three-way subClassOf typing at commit (outcome to AgentCausedEvent, exogenous to ExogenousEvent, automatic to AutomaticEvent), so it carries weight for responsibility attribution: an exogenous event is no agent's doing, an outcome traces to an action. It is a routing input, not stored as a literal. Do not invent an origin outside this set.
- Origin tie-break: when the proximate producer of a happening is a system executing set rules, type it AutomaticEvent even if an agent action supplied the preconditions, and record that enabling action via the caused_by_action link; AgentCausedEvent is reserved for consequences not mediated by automated rule-following.
- label is a SHORT, GENERAL name of at most 4 words naming the KIND of event, not the case scenario: write "Structural Failure", not "Single-Client Conflict Mitigation Recognized". The label becomes the event's URI; put case-specific detail in the description.
- description is 1-2 sentences carrying the case-specific detail.
- temporal_marker is when the event occurred (the textual when); temporal_extent is "instant" for a point occurrence or "interval" for one that extends over a period.
- initiates and terminates name the STATES (fluents) the event brings into or out of holding (Event Calculus). Name the conditions that become or stop being true (for example "Public Safety Risk State", "Project Halted State"), using the same state names used elsewhere in the case. An event initiates a state, not an obligation; the obligation and constraint links are recovered downstream from the state. An event must not terminate a state it initiates: never list the same state in both initiates and terminates; if a state would appear in both, keep it only in initiates.
- caused_by_action references the action label that caused this event, when one applies.
- text_references are verbatim quotes from the case text grounding this event: each an EXACT contiguous span copied from the case text, never a paraphrase, summary, or stitched fragment. confidence is the extraction confidence 0.0-1.0.
- Do NOT emit severity, an automatic-trigger boolean, preconditions, NESS test factors, or a causal chain here. Severity is dropped for events, and the NESS causal analysis is owned by the separate Stage-5 causal pass.

OUTPUT FORMAT:
Return a JSON object with a single array "events", each entry a specific event occurrence in the case. Events are origin-typed individuals; this pass mints no event classes.

EVENT SCHEMA (events array) -- each event is a JSON object:

  {
    "label": "Structural Failure",
    "description": "A critical structural deficiency was discovered during the independent review of the proposed design.",
    "event_type": "outcome",
    "temporal_marker": "Month 5",
    "temporal_extent": "instant",
    "initiates": ["Public Safety Risk State", "Project Halted State"],
    "terminates": [],
    "caused_by_action": "Task Assignment Decision",
    "text_references": ["a critical structural flaw was found in the detailed review"],
    "confidence": 0.92
  }

FORMATTING: do not use em dashes or en dashes in output text, and do not dodge with stacked parentheticals or excessive colons/semicolons; use commas or split into two sentences.

Loading preview...

Version History

v6 2026-06-10 00:52

Literature-attribution correction (post-defense audit 2026-06-09, HO-008): every scholarly claim verified against the cited paper's full text; fabricated or overstated attributions replaced with verified characterizations. No structural, schema, or variable changes.

v5 2026-06-04 21:37

R5 lit-fidelity: de-fabricate invented citations, restore PDF-verified attributions (2026-06-04)

v4 2026-05-27 09:50

Strengthen style FORMATTING line: no em/en dashes AND no parenthetical-stacking / excessive colon-semicolon substitutes (abbreviated ProEthica style guide; apply in Section C)

Saved

3962 chars

Reference

Template Variables

The LIGHT event field schema, resolved live from the SHACL EventDefinitionShape (single source shared with the OntServe Event entity page). event_type drives the three-way origin subClassOf typing; initiates/terminates resolve to the fluent edges; temporal_marker is a literal

The config map depends events on states and actions: the states half resolves prior state rows; the actions half resolves the Step-3 temporal rows (bare labels). The live Step-3 event prompt additionally injects the extracted actions in-memory.

Governing definitional anchor: the ontology iao:0000115 definition plus the iao:0000116 extraction framing, citation-trimmed at injection (the generalized Role definition pattern)

Formatted list of existing event classes from ontology (plus the reuse-bias guidance block)

Shared cross-component directives: no-fabrication, actor-scope, quote fidelity (single-pass component, so the slot carries only the shared block)

The case text to analyze

Output Schema

events

label* string

initiates array → State

confidence number

event_type* string

terminates array → State

description* string

temporal_extent string

temporal_marker string

text_references array

caused_by_action string

JSON Wrapper

Appended to the prompt by the extractor:

IMPORTANT: Wrap your response as a JSON object with exactly two keys:
{"new_event_classes": [...], "event_individuals": [...]}
If there are no individuals to report, use an empty array for "event_individuals".