RelayPay v0.14.0 — Settlement Intelligence Agent

RelayPay v0.14.0 — Settlement Intelligence Agent

RelayPay v0.14.0 adds a deterministic settlement intelligence agent on the shared
PostgreSQL-authoritative runtime while preserving /api/v1 and every earlier lost-response
guarantee.

Highlights:

• immutable, versioned merchant settlement policies with one IANA timezone, one daily cutoff,
  T+0/T+1/T+2 weekday delay, and include-or-skip weekends; the default is 17:00 Asia/Kolkata, T+1,
  weekends skipped;
• immutable pre-cutoff settlement forecasts with deterministic capture, refund, and
  receivable-offset line items; re-forecasts append a new sequence and never rewrite history;
• exactly four supported question intents — why is today's settlement lower, which payments are
  still unsettled, what is the refund cash impact, and when is the expected arrival tomorrow —
  with a typed clarification response for anything unsupported or ambiguous;
• a model contract that can only classify intent and dates; models never write SQL, compute
  financial values, or mutate state;
• fixed tenant-scoped query functions that produce every number, calculation step, citation
  (recordType, recordId, fieldPaths, snapshotSha256), and trace;
• lower-settlement answers compared against the latest immutable pre-cutoff forecast with capture,
  refund, receivable-offset, and settlement deltas;
• post-generation narrative validation that rejects altered numerics, unsupported claims, missing
  citations, cross-tenant record references, and malformed output;
• idempotent questions with workflow runs, model invocations, immutable prompt and pricing
  versions, and business events;
• admin APIs, generated SDK models, a console settlement workspace, local demo data, and browser
  coverage with accessibility checks.

This is synthetic INR portfolio software. It handles no real money, identity, customer, banking, or
provider data. All model output is deterministically validated and every integration is synthetic.
No deployment or hosting is included.