Showcase: From Person to Verdict (Val d'Oria)

This showcase walks every layer of the engine using a real build of the hrcentral pack against the Val d’Oria Academic Institutions tenant (slug: vai). The numbers, signals, and verdicts below are not mocked: they come from hrcentral_vai_kls.duckdb, the same KLS our internal JinDesk reads.

Val d’Oria is a synthetic but realistically-shaped cluster of academic institutions — workforce, contracts, projects, programs, infrastructure, publications. We use it to validate every release of the hrcentral pack and as the canonical demo.

The tenant

The shape of the dataset:

Entity	Rows
Persons	104,326
Person-role assignments	111,145
Affiliations	15,709
Assignments	13,569
Positions	12,659
Contracts	6,933
Bookings	2,384
Cohort	311
Enrolments	97,887
Publications	20,553
Events	14,065
Qualifications	3,739
Grants	2,059
Org units	189
Programs	36
Infrastructure	80
Projects	150
Project funding	60
Project budget	150

These are the Gold contract — the consumption layer. Signals operate exclusively on Gold; the medallion below is invisible to the analytical instruments.

Step 1: Signals — detect anomalies

The pack ships 17 signals for hrcentral. On the Val d’Oria tenant, 16 of them have findings:

Signal	Findings	Worst	What it detects
signal_affiliation_drift	4,812	medium	Persons whose recorded affiliation no longer matches their active role
signal_dormant_employee	1,329	high	Active contracts with no recent assignment, booking, or publication activity
signal_ghost_employee	768	high	Persons billed via payroll with no documented work record
signal_succession_risk	419	high	Key roles with no documented successor in the pipeline
signal_dual_affiliation	136	low	Persons holding affiliations at two institutions simultaneously
signal_credential_expiring	109	medium	Qualifications expiring within the policy window
signal_unrecorded_workforce	60	medium	Active assignments without a corresponding contract record
signal_project_funding_gap	18	high	Projects with committed work but no matching funding record
signal_project_carrier_single_person	12	high	Projects depending on a single principal investigator (no co-leader)
signal_infrastructure_unbooked	6	medium	Infrastructure assets with no booking history in the active period
signal_project_orphan	6	medium	Projects with no person assignment recorded
signal_infrastructure_no_owner	5	medium	Infrastructure assets with no assigned owner role
signal_offboarding_residue	5	high	Persons whose offboarding left active permissions or assignments behind
signal_overload_risk	3	high	Persons with cumulative assignment > 100% FTE
signal_pi_overcommitment_projects	1	high	Principal investigator committed to more projects than time permits

Step 2: Perspective — aggregate

One perspective rolls signal findings up to the person level:

Perspective	Entities	What it measures
perspective_person_health	5,630 persons	Overall person risk combining workforce visibility, succession, credential, and assignment signals

The entity_signal_summary table holds 6,158 entity-signal pairs — every person, project, or infrastructure asset that any signal touched, with severity and aggregated risk.

Step 3: Theses — confirmed

The build produced verdicts on 9 theses. Eight are confirmed with full evidence scores:

Thesis	Status	Findings	Evidence score
thesis_workforce_visibility	confirmed	6,969	1.00
thesis_affiliation_drift	confirmed	5,367	1.00
thesis_credential_compliance_gap	confirmed	664	1.00
thesis_succession_fragility	confirmed	419	1.00
thesis_workforce_overcommitment	confirmed	139	1.00
thesis_project_portfolio_gaps	confirmed	19	1.00
thesis_funding_commitment_gap	confirmed	18	1.00
thesis_data_protection_compliance	confirmed	5	1.00

One thesis (thesis_infrastructure_stewardship) lands not_observed — the engine evaluated it but the pattern didn’t materialise in this build.

The evidence scores all hit 1.00 because the signal panel for each thesis on this tenant has at least one primary signal with findings and the supporting signals each weigh in. On a tenant where the data is cleaner, the same theses would show lower scores — or not_observed.

Step 4: The full chain

For one confirmed thesis, the chain reads top to bottom:

Every box above is a row in the KLS. Click through them in JinDesk and you walk the same path the engine just walked — finding to thesis to verdict, with full provenance.

What this showcase teaches

• The same engine that ran the Inspire showcase runs here — the instrument chain is pack-agnostic
• Workforce signals carry no money_at_risk by default — the engine handles non-monetary risk without forcing a CHF figure
• A high evidence score (1.00) doesn’t mean a high-stakes problem — it means all the evidence points the same way; severity comes from the signal counts and entity scope
• Perspectives collapse 16 signals into one person-level read — the dashboard view a workforce manager would actually open first

Compare

See the same engine at work on a completely different domain: From Case to Verdict (Inspire) — the nuMetrix hospital material flow pack against the Inspire tenant. Different entities, different signals, same instrument chain.