HOW IT WORKS

From filing to verdict in a single deterministic path.

Every verdict is a deterministic function of the inputs, the methodology, and the version. Replayable. Citable. Auditable. The same six steps for every security, every methodology, every tenant.

1. Source extraction. Public regulatory filings (ESEF, EDGAR) ingested deterministically. iXBRL facts extracted by fact ID. Narrative sections processed by a constrained-schema LLM pipeline with confidence floors. Every figure is cited back to its fact ID, page number, or text snippet offset.
2. Confidence floor and human review. Each extracted figure carries a confidence score. Composite verdict confidence below the tenant’s configured floor routes the case to a human-review queue before the verdict is issued. No silent low-confidence approvals.
3. Methodology evaluation. The signed YAML methodology document is loaded by version. Sector screens run first. Financial ratios are computed. Thresholds are compared. Every rule’s pass/fail status is recorded with its numerator, denominator, and margin.
4. Parallel methodologies. The same security runs against all four methodology engines in parallel — AAOIFI, DJIM, S&P Shariah, and the tenant’s House methodology. Divergences are exposed at the rule level, so the institutional choice between methodologies is visible, not hidden.
5. Verdict and purification. The verdict is issued with full structural transparency: methodology version, input figures with citations, rules evaluated, ratios computed, thresholds applied, and purification ratio per AAOIFI Standard 21 § 3/4 where applicable.
6. Audit log and replay. Every verdict is written to the tamper-evident, hash-chained audit log. The replay endpoint reproduces any historical verdict from its inputs. This is the same view your internal auditor will use — and the same view a competent authority would request in any ex-post review.

Built on a TRL 6 working system.

4,400+ LOC backend across the four methodology engines, replay, and audit log. 13 API endpoints. 190+ test assertions on edge cases. Available for live demonstration to qualified institutional teams.