DGT · Explore the control record

FICTIONAL DEMONSTRATION · EXPERIMENTAL SOFTWARE · NOT FOR PRODUCTION PAYROLL USE

Records with duplicate or blank IDs are not compared with the other file. Every listed conflict must be investigated by hand.

100Approved source rows

102Submitted source rows

7Reported changes (5 modified, 1 removed, 1 added)

2Identity conflicts requiring investigation

What would the operator do?

An approved payroll export and a submitted export may have the same records in a different order. The task is to account for differences without assuming that a changed or duplicated record is authorised.

Before · raw-file inspection

1. Open two CSV exports

2. Match transaction IDs across reordered rows

3. Check every field for additions, deletions or edits

4. Investigate ambiguous identities

5. Document what was actually checked

After · proposed control record

1. Read the comparison scope: COMPLETE for shared columns

2. Inspect the 7 reported changes

3. Manually investigate 2 identity conflicts

4. Trace exceptions back to source Excel rows

5. Retain a ledger accounting for all 202 physical rows

This is a proposed workflow improvement. No operator time savings have been measured. Excel and Power Query can perform comparable work.

Exception explorer

Select an exception to inspect its exact recorded values and source references. These records are read from the original PWR57 exceptions CSV; this page does not rerun the comparison.

93 unchanged identities are not listed here. The original engine's ledger accounted for all 202 physical source rows in its tested fixture; this preview is not a ledger verifier.

Where the tool can mislead

Independent adversarial C3 case — separate from the original 100/102-row fixture.

Approved snapshot

T003

Amount: 3200.00

One approved row

Submitted snapshot

T003 × 2

Amount: 9999.00 on each copy

Duplicate identifier

What the tool actually shows: It flags the duplicate and references all three physical rows, but says “0 changes” and does not display 9999.00 in the readable report or the exceptions CSV. The approved-to-submitted value difference is not shown.

Records with duplicate or blank IDs are not compared with the other file. Every listed conflict must be investigated by hand.

Someone who removes one duplicate without comparing the values could leave the inflated amount unresolved. A zero-change headline does not mean zero unresolved discrepancies.

What this experiment supports

Demonstrated: On the tested fictional fixture, the engine reported 5 modifications, 1 removal, 1 addition, 2 identity conflicts and 93 unchanged identities, with all 202 physical rows accounted for. Independent testing found no changed value silently marked unchanged in realistic unique-ID cases.

What it does not establish

No production reliability, security certification, authorisation decision, real-world time savings or evidence of buyer interest. Duplicate and blank IDs require manual inspection. Formatting-only changes may be reported; Excel may strip leading zeros from CSV values. An independent reviewer found the automated scorer detected only 3 of 10 deliberate report corruptions.

What we would learn from external reactions

Would this make an existing control easier to perform? Is the output genuinely useful, or already handled adequately with Excel or Power Query? Which parts of the exception record would be confusing or untrustworthy?

No personal information is collected by this page. There are no external links, uploads or network calls. The 30-day observation period begins when the proposition first becomes publicly accessible.

Digital Grease Trap — We do the digital dirty work. · Fictional presentation only · Interactions are local to this page, not an operational payroll tool.