Operating model

Running Parallel to the Paper

Your instinct is right. Two things are called scanning, though, and only one of them should happen everywhere on day one.

The answer

Yes — run parallel. But do not try to digitise the file. Digitise its movements first, and only its decision pages second. A registry that photographs 60 pages per jacket will quietly stop scanning by week three; a registry that taps a barcode twice a day will still be doing it in year two.

Scan type 1 · Track-scan A barcode on the jacket, tapped at every counter Two seconds. No typing, no judgement, no literacy in English required. It answers the only question the whole chain cannot currently answer: where is this file and who has it.

Every station, day one Non-negotiable

Scan type 2 · Page-scan A photograph of a document, attached to the record Slow, storage-hungry, and legally awkward the moment anyone asks which copy is the original. Genuinely valuable for about seven pages per case — the ones that carry a decision or prove a handover.

Selective Never "scan everything"

Why "scan everything" fails: a full jacket is 40–80 pages of handwriting, thumbprints and carbon copies. At a realistic 20 seconds a page that is half an hour per file at a counter that already has a queue. It will be done for the first fifty files and then abandoned, and an abandoned scanning policy is worse than none — the dashboards will show coverage falling and nobody will trust any of it.

What gets captured at each of the ten records

Seven page-scans across the whole life of a case. Everything else in the jacket stays paper and is referenced, not copied.

Record

Track-scan

Page-scan

Seconds

14track-scans per caseTwo per handover — one out, one in — across seven physical movements. Under thirty seconds of counter time for the entire life of the file.

7page-scans per caseCover minute, inward leaf, allocation sheet, perusal minute, sanctioned charge sheet, outward leaf, judgement. Roughly four minutes total, spread across four institutions.

0statements scannedForm 3, Form 4 and Form 18 stay paper in the pilot. They are the bulk of the pages, the least useful digitally, and the most sensitive to copy.

Six rules for the parallel period

Parallel running fails the same way everywhere: the digital side becomes a chore done at the end of the week from memory, and then the timestamps are fiction. These six rules are what stops that.

{{ r.body }} In practice — {{ r.practice }}

Four phases, with the gate that ends each one

Nobody moves to the next phase because a date arrived. They move because a number was met on the Jinja Road → Central Registry → RSA Nakawa → Nakawa Court corridor.

Gate to the next phase {{ p.gate }}

Paper status {{ p.paper }}

What never stops being paper The Station Diary, because it is a legally bound pre-numbered book. Signed and thumb-printed statements, because the signature is the evidence. The charge sheet bearing the ODPP Sanction Stamp, because the stamp is the authority. Retiring those is a legislative question, not a software one — and it is not on this roadmap. What can be retired is the duplicate tracking: the CID dispatch register, the inward register and the outward register are all just movement logs, and a scan does that job better than a signature nobody ever reads back.

What this adds to the Django schema

Three new models and one changed assumption. The changed assumption is the important part: FileMovement stops being something a clerk types and becomes something a pair of scans produces.

Five ways this goes wrong, and the control for each

Failure {{ r.title }} {{ r.body }}

Control {{ r.control }}

Kit per counter

Deliberately unglamorous. No dedicated scanners, no kiosks, no laptops at the counter — those get locked in a cupboard for safekeeping and never come out.

The one thing to spend real money on: a week of counter training per registry and a laminated one-page procedure taped to the desk. Every registry system that has ever failed in this region failed at the counter, not in the code. If clerks do not scan, none of the charts are real — and a chart that is wrong is worse than no chart, because somebody will make a decision with it.

Built by