Trust score
3.3 Per-receipt quality and per-user trust
Trust is tracked at two connected layers. Each receipt that exits the pipeline receives a quality assessment: the layer checks how complete the extracted record is (merchant, date, time, totals, line items) and whether the receipt's amounts reconcile internally. The assessment resolves to a quality tier for the receipt. That tier then feeds a cumulative per-user trust standing, updated synchronously in the same processing pass: a stream of clean, complete receipts raises standing over time; low-quality or inconsistent uploads slow or reverse it. Downstream systems consume the standing as tiers rather than raw values.
Signal families
The quality assessment in the current release draws on record completeness and amount reconciliation, together with duplicate detection (3.9). The following signal families extend the model and are planned; they are not active in the current release:
| Family (planned) | What it observes |
|---|---|
| Pipeline confidence | How confident the upstream pipeline was in the extraction (OCR confidence, LLM confidence, rule-layer reconciliation). |
| Merchant consistency | Whether the merchant, branch, and receipt template match what we have seen before from this merchant. |
| Temporal plausibility | Whether the receipt's date, time, and the user's upload pattern are consistent with normal behaviour. |
| User history | The user's recent contribution quality, scoped to a rolling window. |
The exact scoring composition, tier boundaries, and per-tier effects are managed in the internal operations layer.
Quality tiers
The per-receipt assessment resolves to an ordered set of quality tiers. Higher tiers reflect complete, internally consistent receipts and strengthen the user's trust standing more; lower tiers reflect sparse or inconsistent records and contribute less. The tier definitions and their exact effects are calibrated in production and not published.
3.4 What the receipt record carries
The receipt's quality block records the assessed tier and the completeness observations that produced it. Tier boundaries and per-tier effects live in the internal trust configuration. The user-facing surface communicates the outcome (bINT amount, reject) and the reason category when relevant.
This is intentional: public outcomes provide clarity while internal calibration values preserve the calibration surface.