Methodology

A selection mission binds a programme, organization, goal, candidate set, decision protocol, and human review boundary. Its lifecycle is explicit: draft, active, waiting for review, blocked, paused, completed, or cancelled. Completion can mean either an approved shortlist or an explicit conclusion that no line is ready.

Riyaan may gather, structure, reconcile, monitor, and propose only inside that approved scope. A document cannot expand the scope, invoke a tool, or change permissions.

1. Observe

   Accept an approved signal: mission start, source intake or change, evidence review, proof outcome, approval resolution, or due task.

2. Reconcile

   Compare the signal with the current evidence snapshot and versioned protocol.

3. Advance

   Run the next bounded, idempotent task that does not require new authority.

4. Ask

   Create a targeted approval request when named expert judgment is required.

5. Resume

   Continue from the recorded resolution without reconstructing prior context.

Durable tasks use bounded retries and visible failure states. Overlapping work for the same mission is prevented, and overdue leased work is eligible for one catch-up attempt after restart.

Private uploads are quarantined until declared and detected MIME types match and malware scanning passes. Registered URLs must use HTTPS, match an allowlist, survive redirect checks, and resolve outside private or local networks. Snapshots retain a content hash, version, and retrieval time.

Model assistance is limited to extraction and drafting. Structured output must identify the document, document version, candidate, page or section, extraction version, and confidence. A missing citation fails validation. Every accepted draft starts as unreviewed and cannot silently replace canonical evidence.

A versioned decision protocol—not a model draft—determines blockers, readiness bands, candidate ranking bands, and completion eligibility. Explicit exclusions and unresolved required evidence are evaluated before supporting signals, so a blocker cannot be averaged away by favorable evidence elsewhere.

A proposed next proof may use transparent impact, resolution, timing, and cost bands to order work. These are programme planning inputs, not measured biological probabilities or validated outcome predictions.

Separate approval records cover evidence promotion, proof plans, candidate dispositions, and final briefs. Approval records retain the target, reviewer identity, resolution, and time. Rejecting an item leaves the mission waiting for review; it does not promote the proposed content.

“Reviewed” records a workflow decision in a stated programme context. It does not assert independent reproduction, laboratory validation, regulatory acceptance, freedom to operate, clinical suitability, or universal fitness of a cell line.

Each brief revision records its evidence snapshot hash, protocol key and version, unresolved blockers, model metadata where applicable, content, and reviewer approval. Earlier revisions remain available instead of being overwritten. A printable web view supports browser PDF export.

The finished brief shows candidates, exclusions, unresolved risks, source-linked rationale, and the next proof worth funding. It may conclude that no line is ready when required evidence or approvals remain insufficient.

• Riyaan is not a LIMS, QMS, clinical diagnostic, CoA generator, or patient-treatment selector.
• A source citation does not establish truth, commercial rights, regulatory status, or fitness for a particular purpose.
• Real programmes require independent source verification, laboratory testing, quality review, rights review, and appropriate clinical and regulatory expertise.
• The public catalogue and evidence surfaces remain illustrative supporting material; the public mission itself uses only synthetic identifiers and events.