This synthetic example shows the behavior GCTS is meant to make explicit. It is not based on an active investigation.
Scenario
A safety incident is reported at a facility. A visitor says a staff member saw the incident and that policy should have required an incident report. The facility produces a visitor roster but does not produce an incident report, medical referral, or supervisor review. A staff statement says no report was required because the event was minor.
Question:
Did a documentation-triggering safety incident likely occur?
Candidate Claim
c_1: A documentation-triggering safety incident occurred at Facility A on Date T.
Evidence Atoms
| ID | Source | Content | Quality | Notes |
|---|---|---|---|---|
e_1 | visitor statement | Visitor reports seeing the incident and staff response | 0.70 | Direct but single-source |
e_2 | facility roster | Visitor and staff were present at the relevant time | 0.85 | Produced official record |
e_3 | staff statement | Staff characterizes event as minor | 0.55 | Potential institutional incentive |
e_4 | policy excerpt | Visible injury requires incident report and supervisor notice | 0.90 | Strong rule evidence |
Record-Access States
| ID | Expected record | Duty | Access state | Production state | Confidence |
|---|---|---|---|---|---|
r_1 | incident report | policy_required | unknown | not produced | 0.75 |
r_2 | medical referral | conditional_on_visible_injury | unknown | not produced | 0.62 |
r_3 | supervisor review | policy_required_if_report | inaccessible | no response | 0.58 |
r_4 | visitor roster | ordinary_admin | available | produced | 0.90 |
Candidate Worlds
| World | Description | Key assumptions |
|---|---|---|
W_A | Incident occurred and report was expected but not produced | visitor reliable, policy applies, record absent/non-produced |
W_B | Minor event occurred and report duty did not trigger | visitor partly reliable, staff framing reliable, policy threshold unmet |
W_C | No documentation-triggering event occurred | visitor mistaken, staff statement reliable, no report expected |
W_D | Incident occurred and report exists outside current access path | visitor reliable, policy applies, record inaccessible |
Example Scores
| World | Posterior |
|---|---|
W_A | 0.46 |
W_B | 0.24 |
W_C | 0.12 |
W_D | 0.18 |
Claim posterior:
P(c_1 | E,A,I) = W_A + W_D = 0.64
Strict support:
P0(c_1 | E) = 0.00
Confidence:
Conf(c_1) = 0.52
Output Status
c_1: record_contingent / plausible-to-probable
The system does not promote c_1 to strict proof because the expected
institutional records have not been produced. It ranks c_1 as likely under
the top worlds while marking the claim record-contingent because production of
the incident report, medical referral, or supervisor review could materially
change the ranking.
Audit Output
{
"claim_id": "c_1",
"text": "A documentation-triggering safety incident occurred at Facility A on Date T.",
"status": "record_contingent",
"posterior": 0.64,
"strict_support": 0.0,
"confidence": 0.52,
"supporting_evidence": ["e_1", "e_2", "e_4"],
"refuting_or_qualifying_evidence": ["e_3"],
"record_contingencies": ["r_1", "r_2", "r_3"],
"top_worlds": ["W_A", "W_B", "W_D"],
"next_records": [
"incident_report",
"medical_referral",
"supervisor_review"
]
}
What The Example Demonstrates
- GCTS can rank likely truth without strict proof.
- Missing expected records affect status through access-state logic.
- A produced roster supports presence but does not resolve the incident claim.
- A staff statement can reduce confidence without eliminating higher-posterior worlds.
- The output identifies the records that would change the claim status.