LO
PilotCognitive automation2014-04-10
Character recognition matched keyed values about 73 percent of the time on census forms, and fields it could not read went to a keyer, the Census Bureau found
Data entry clerkoccupation page →
Event date / reported
2014-04-10
Evidence stage
PilotSmall-scale trial in a real setting. Tells us the deployment conditions are being tested, not that they hold — so one pilot is never enough on its own; two independent ones are.
Tasks this bears on
Keying data from documents
Typing data from paper forms, scans and statements into systems.
Automating✓ Evidence-backed
Handling exceptions
Dealing with documents the system cannot read, sensitive data and unusual cases.
Still human-led≈ Platform inference
Where this applies
United States, the Census Bureau's processing centre, where clerks key all write-in responses from scanned questionnaires. In a test across more than 71,000 values, character recognition matched the keyed value about 73 percent of the time; almost 27 percent it did not read, and when it cannot determine a field's value the field goes to a keyer. Of all values it read, 0.3 percent were in error, less than the 1 percent error rate required for keyed batches. It is an older test on numeric fields only.
What this means
The design that later spread everywhere: the machine reads what it is confident about and a person keys the rest. Keying shrinks to the fields the machine cannot read.
What it does not yet show
An older test on numeric fields; today's tools read more, and it does not measure jobs.
What you can check
Open Census Bureau memorandum ACS14-RER-13 and find "the field goes to a keyer".
Does it change the assessment?
No. The impact index is never moved by a single event, and this stage does not move one on its own: a "Pilot" record counts toward a judgement but needs a second, independent record before the judgement rests on evidence. This one is counted; on its own it changed nothing.
Source
U.S. Census Bureau — Evaluation of the use of Optical Character Recognition to capture American Community Survey numeric write-ins in the 2013 Questionnaire Design Test, ACS Research and Evaluation Memorandum #ACS14-RER-13 (04/10/2014) · verified 2026-09-30 · Claude (VOLO agent) · interpreted 2026-09-30 · Claude (VOLO agent)
Primary source — published by the party that did this, or the authority of record. No co-signature needed.