Sample engagement
Sample — corporate PD annual validation
A full model check, run live in your browser on AlphaIQ's own made-up data — no real bank or customer in it, and nothing saved anywhere.
Model
Corporate PD Model v3.2
Ownership
internal
Status
draft
Owner's severe-case claim
2.1%
Running the replica, the four scenarios and the metric pack in your browser…
| Fact | State | What the record says |
|---|---|---|
| Family | Credit & capital risk | Credit & capital risk |
| Ownership | Internal build | Internal build |
| Accountable lead | Ilan | Ilan |
| Declared specification | Corporate PD Model v3.2 (v1, 1fa221c3be64) | Corporate PD Model v3.2 (v1, 1fa221c3be64) |
| Owner attestation | Not established | The specification is at status "draft". Until the model owner declares it, AlphaIQ is the author of the artefact under validation and there is no independent specification to certify against. Obtain the model owner's declaration of the current specification. |
| Scored population | 4,000 scored rows as at 2026-06-30 | 4,000 scored rows as at 2026-06-30 |
| Replication check | Not established | The ownership arrangement permits independent replication and none has been run, so nothing on file says our replica reproduces the owner's model. Run the replication gate against the declared specification and the scored population. |
| Uncertainty simulation | Not applicable | Posterior uncertainty is quantified by running the model's own generative mechanism many times, which only a simulation engine can do for models built on it. This model was not built on one, so uncertainty around its parameters cannot be sampled — a fact about how the model was built, and not something the engagement can supply its way out of. |
| Outcomes analysis | Not established | No metric run has been recorded for this engagement, so no discriminatory power, calibration or stability figure exists for this model on file. Run the metric pack; a figure in a screenshot is not a run. |
| Last validation | Not established | No certificate has ever been issued for this model, so it has not been validated. This is not a late validation — it is a model the validation function has never concluded on. Take the engagement through to certification. |
| Review status | Not established | There is no last validation, so there is no interval to measure. This model is not within its review cycle and it is not overdue — it has never entered one. Validate it once; the clock starts at the first certificate. |
Attestation
NOT ATTESTED — The specification is at status "draft". Until the model owner declares it, AlphaIQ is the author of the artefact under validation and there is no independent specification to certify against.
Scenarios this run used
This engagement has no stressed scenario of its own, so it runs on the illustrative library. Those trajectories were drawn by AlphaIQ to the shape of published ones and cannot support a certified figure — import a supervisor's or the client's own to replace them.
- base-2026 · base · illustrative
- adverse-2026 · adverse · illustrative
- severe-2026 · severe · illustrative
- stagflation-2026 · custom · illustrative
What the sample engagement does not have
Eight tables hold no rows for it. They are listed rather than hidden, because an absence a screen does not mention is the absence a reader assumes was filled.
- certifications — The sample engagement has never been certified. It cannot be: the certification precondition requires the model owner's attestation of the specification, and deliberately removed the seeded one because fixture text naming a person is not an attestation.
- certification_revocations — Nothing has been certified, so there is nothing to revoke. The absence here is a consequence of the absence above rather than an independent gap, and it is listed separately because a reader counting evidence should not have to infer one from the other.
- metric_runs — No metric run has ever been written down for the sample engagement. The demo COMPUTES one in the browser from the specification and the population, and shows it as a run in progress rather than as a record on file.
- simulation_runs — No simulation run is stored. The demo runs the same engine in the browser; storing a run is a write, and a fixture-backed surface performs none.
- replication_checks — No replication check is stored. The demo re-derives one from the population, whose `score` column is the model's own output for the same row, so replication is genuinely computable rather than asserted.
- simulation_findings — Findings are attached to a stored simulation run, and there is none.
- finding_records — The durable register is empty. Every finding the demo shows is one THIS run raised, not one anybody has accepted, assigned or closed.
- finding_proposals — A proposal is raised against a stored run, and there is none.
No audit trail is shown. The trail records mutations to stored rows, and nothing on this page is stored — every figure was computed in your browser and none of it was written anywhere. This is not the sample engagement's trail rendered as empty; it is a screen with no trail to render.
The other families
This is the credit sample. The same walkthrough exists for every other family that ships one, each with its own evidence, its own thresholds and its own planted findings.
Every figure on this page was computed in your browser from files in this repository. No request was made to a database, and nothing you do here is stored anywhere.