PixSettle v1.1.0 — historical shadow reconciliation and recorded replay

PixSettle v1.1.0 — historical shadow reconciliation and recorded model replay

The public pilot now reconstructs verified historical events in read-only shadow mode. Recorded model explanations cite evidence, and a Codex-assisted operator advisory appears in the case timeline. All six exceptions still require human review; no financial approval or settlement is recorded by the pilot.

Measured on the deployed release on 8 October 2026 at 03:53 UTC:

• 2,216 accepted historical events across five scoped store aliases, covering 1,195 orders.
• 586 observed Pix payments totaling R$ 425,119.94; 582 exact order/payment matches totaling R$ 422,310.95. These figures describe validated observations, not bank settlement.
• Four amount mismatches and two MED cases. Six model advisories and one Codex-assisted advisory; zero resolved cases.

This release validates signed shadow batches, records durable ingestion receipts, preserves exact confidence values and hashes, and serves stored advisories without public model calls or financial effects. The historical projection is an explicitly traced subset of the source; a live incremental feed is not verified.

The separate simulated demo retains its existing Tempo testnet receipt and three webhook deliveries, with two duplicates ignored. Its transfer is not a settlement of the historical pilot.

Validation: 286 product tests passed and one Moderato network integration test was skipped; all three TypeScript builds passed. Migrations 006–012 were applied and reversed against the restored demo snapshot with financial data preserved. The deployed release passed all 60 smoke checks, including zero public pilot model and chain attempts during the measured request window. Six private recorded model calls cost US$ 0.0024676.

Source: 5686a76. Render deployment: dep-db3h6c8m7kps73em4q20.

Public demo: https://pixsettle-demo.onrender.com/demo
Historical pilot: https://pixsettle-demo.onrender.com/pilot

The attached two-minute video contains 480 real browser captures, exported as 120 seconds of 1080p H.264 footage with English captions. It labels the recorded model replay, Codex-assisted advisory, and the separate simulated Tempo testnet receipt. No new order, webhook or financial transaction was created for filming.

Video SHA-256: f14becd19064bc970e7ef9784d410b0050718ebff1d2dcf303d0ab7db1bb0933

Updated walkthrough (8 October 2026): PixSettle v5 — recorded shadow pilot, human review and replay. This 120.28-second walkthrough is a later recording of the public pilot and the separate simulated demo. It does not change the release source, deployment, dated measurements or original attached video above. The pilot remains read-only shadow reconciliation, with human review and no financial settlement.

Updated video file SHA-256: fe7d4d198357081437a9228387bc55bc8f393067e5c83cc9b10f1b7edb615c3a.