JSON in. Benchmark report out.
No images. No OCR. No live parser calls.
QA Lite works on structured JSON from any parser or vendor. Send 25–50 redacted outputs. We run QA checks and return a privacy-safe report showing what passed, what needs review, and what should be rejected — with reason codes for every flag.
Accept, review, or reject —
with reason codes.
Every benchmark report shows which outputs passed QA checks, which need manual review, and which should be rejected before downstream processing. Each flag includes a reason code so your team knows exactly what failed and why.
- Accept / review / reject verdict for every sample in the batch
- Reason codes per flag — math failure, missing field, structural issue, buyer-critical
- Buyer-critical issue rate — the share of outputs that could affect payouts or disbursements
- Redacted examples and a gap summary included in every report
subtotal_tax_total_mismatch
line_sum_mismatch
Structured checks across
fields, math, and schema.
QA Lite runs a defined set of checks on every JSON output in your batch. No configuration required — checks apply to every sample automatically. Each check maps to a reason code in the report.
- Required fields (total, line_items) checked for presence and validity; merchant and date fields checked for presence where required
- Math reconciliation: line item sum must equal subtotal; subtotal + tax must equal total
- Structural checks: empty arrays, null values in required fields, duplicate line items
- Buyer-critical flags: missing total, zero-item receipt, structural failures that could affect payout
Send redacted JSON.
No images required. Low-retention benchmark handling.
QA Lite works on JSON structure and math only. No receipt images, no OCR text, no raw capture data needed. Redact or anonymize before sending — we don’t need personal identifiers to run checks. Raw inputs can be deleted after the benchmark.
- No receipt images or OCR text required — JSON structure only
- Redact or anonymize samples before sending — PII not needed for QA checks
- Reports are designed to avoid raw payloads and sensitive fields
- Raw inputs can be deleted after the benchmark is complete
No images required
QA runs on JSON structure and math. Receipt images are not required and are not part of the QA Lite process.
Send redacted samples
Anonymize before sending. Personal identifiers are not needed to run QA checks.
Low-retention handling
Benchmark handling is designed to minimize raw receipt exposure. Raw inputs can be deleted after the benchmark.
Privacy-safe reports
Reports are designed to avoid raw payloads, sensitive identifiers, prompts, model responses, and model envelopes.
What ReceiptCheck is building.
ReceiptCheck is starting with QA Lite, a JSON-only benchmark for parser and vendor outputs. Additional QA workflows are in development for teams that need ongoing monitoring, vendor comparison, or evidence-based checks against OCR text or receipt images.
Request a sample benchmark report.
Send 25–50 redacted JSON outputs from your current parser or vendor. We’ll return a privacy-safe report with accept / review / reject counts, reason codes, buyer-critical issue rate, and a gap summary.