A vendor shows a demo on a clean PDF. GSTIN extracted. Total extracted. Looks like 98%. Then you send your photocopies. 150 dpi. Crooked. Phone photos. The number drops to 72%. The demo was not a lie. The demo was not your paper.
Straight-through rate is the only number that means anything here, and it has to be measured on your paper. This post is how it is counted. Your console shows it, measured on your own paper. There is no committed minimum. The number is what it is.
ST = AUTO_EXPORTED with no human touch. The validator passed on all fields. The row exported to Tally. No clerk confirmed or corrected. Not a confidence score. Not "probably right." Either the gate passed or it did not.
Numerator and denominator
Denominator: the invoices you process in a period. Not a cherry-picked sample. Not the clean ones. The whole batch.
Numerator: invoices that auto-exported without a human touch. The validator passed GSTIN, HSN, arithmetic, and regime. Required fields were present. The row exported to Tally.
ST = numerator / denominator. 340 of 400 auto-exported is 85%. 320 of 400 is 80%. Same arithmetic, whichever way your paper goes.
340
auto-exported, no human
400
invoices in the corpus
85%
340 / 400
Example. Your numbers will differ. The denominator is everything you processed.
What ST is not
ST is not field-level fuzzy accuracy. It is not "92% of fields extracted correctly." It is not a confidence score. It is a binary: the row exported without a human or it did not.
ST is not IRN lookup rate. If your invoices carry IRN and you want IRN validation, that is a different path. ST is the scan-to-Tally path on photocopies without IRN.
ST is not a lab score. The demo on a clean PDF is not ST. Your photocopies are ST. If the demo shows 98% and your paper shows 72%, 72% is the honest number.
What we measure
Your own scans. Straight-through rate, flagged invoices, operator time. All of it measured by instrumented pipeline events and shown in your console.
The number is the deliverable. You watch it with your own paper in front of you. We do not set a floor, because a number picked before seeing your paper would not be worth much.
Billing is per page actually processed, at ₹1.40 a page plus GST, dropping to ₹1.20 beyond 40,000 pages a month.
What if the paper is too bad?
If your photocopies are 80 dpi, handwritten, or in a language we do not read, ST may be low. That is the honest answer. The console will show you.
The review queue also tells you which suppliers are the problem. If 30% of the failures come from one supplier, that is a supplier conversation. Better scans. Digital invoices. IRN. The product does not fix the supplier's paper. It tells you which supplier needs fixing.
- EntryLedger/validation
Straight-through is measured on your paper and shown in the console.
- EntryLedger/validation
AUTO_EXPORTED or NEEDS_REVIEW. Binary. No slider.
What is ST?+
AUTO_EXPORTED on your paper. No human touch. Not a lab PDF.
What counts in the denominator?+
Everything you processed in the period. The whole batch, not a clean sample.
Where do I see the number?+
In the console, measured on your own scans, from the first batch.