Keep evidence visible.
Values link to source lines. Original sample files stay available, and corrections preserve the initial extraction.
FieldnoteFrom test sheets
to a reviewed register.
A working example of how I turn an operational problem into a usable tool: organize inconsistent records, surface the exceptions, and keep the evidence close to every decision.
No account, setup, or paid AI required.
Load five fictional records, compare extracted values with their sources, resolve or exclude exceptions, and export a reviewed CSV. Progress stays in your browser. This is an independent demonstration, not commissioned client work.
Walkthrough / 65 seconds
See the workflow before you try it.Follow a real session in the sample demo: check an ambiguous date, explain a correction, hold unresolved records, and export the reviewed register. Silent video with on-screen captions.
Fictional sample data · Rules-based prototypeDownload MP4 · 3 MB ↓
Read the walkthrough transcript
- 00:00 / The problem. Scattered sheets. Missing details. A register someone still has to check.
- 00:06 / Organize. Load five fictional sheets. Keep the extracted fields and their sources together.
- 00:14 / Check the source. 09/10/2026 is ambiguous. The visit schedule gives the confirmed date, 2026-09-10.
- 00:24 / Explain the change. Correct the date, link the evidence, and save the reason for the correction.
- 00:30 / Human review. A person checks the fields and measurements. This is not equipment certification.
- 00:36 / Hold unresolved records. No instrument identity is recorded. Hold the sheet for confirmation; do not invent a value.
- 00:43 / Keep exceptions visible. The conflicting asset and possible repeat also need confirmation. Their explanations are retained.
- 00:50 / Reviewed output. Two reviewed sheets. Three held for confirmation. Export only the reviewed measurements.
- 00:57 / Start with your workflow. Try the demo at matlock.work/fieldnote. Then discuss a pilot around your team’s process.
01 / The business opportunity
Less time
rebuilding the record.
More time reviewing it.
Service teams may collect test sheets with different labels, missing details, conflicting identifiers, and repeat visits. Someone has to reconcile those records before they can become a useful register.
Fieldnote explores that handoff. It preserves the source text, organizes supported fields, and makes unresolved details visible before export. It does not determine equipment condition or make technical pass/fail decisions.
The intended value is less re-entry and a clearer review process. A pilot would measure those outcomes against the team’s existing process before claiming time savings.
THE WORKING SEQUENCE TRY IT YOURSELF
- 01Load the sheetsFive fictional records with realistic exceptions.
- 02Check the evidenceCompare fields with captured source lines.
- 03Resolve and reviewExplain corrections or exclude an incomplete record.
- 04Export the registerCarry reviewed measurements and source references into CSV.
02 / What this shows about my work
The useful part
is the whole workflow.
Automation needs a way for people to inspect, correct, and trust the result. Those steps are built into this prototype.
Make uncertainty actionable.
Missing references, conflicting IDs, ambiguous dates, and possible repeats require a reviewer’s decision.
Review the current record.
Saving a correction removes the previous review. CSV export opens only when every included sheet has been reviewed.
03 / Where AI could fit
Prove the process.
Then broaden
what it can read.
The public demo runs extraction rules over text prepared from the bundled originals. It does not call an AI model, read visitor uploads, or process scans.
A company pilot could evaluate model-assisted extraction across varied forms while retaining the same source checks and human review. Training a specialized model would be a later decision, based on reviewed examples and measured performance.
I would start with the team’s real workflow, its existing software, and a small set of approved examples. Success means reducing the time to a reviewed register without adding missed errors or extra correction work.
Working in this demo
- Source-linked extraction from five bundled samples.
- Corrections with reasons and explicit exclusions.
- Review gates, CSV export, and browser-local progress.
Needs a company pilot
- AI or OCR for varied layouts and scanned sheets.
- Independent accuracy and reviewer-time evaluation.
- Shared access, system integrations, and a durable audit trail.
04 / Try one decision
A date that needs
a second look.
Load the samples and select TR-201. Its date, 09/10/2026, is ambiguous. The visit schedule in source line 7 gives you the evidence to correct it to 2026-09-10.
Save the correction with an explanation, then record a review as Demo Reviewer. Review or exclude the other sheets to unlock export. A walkthrough is included in the demo.
Have a workflow that deserves a closer look?