projectsmanager.ai
Method

How we trace every number

projectsmanager.ai refuses to make up a number. Every figure it shows can be opened to the records it came from. When something is unknown, it says so by name. This page explains how that works and what we will never do.

Try it in the demo Request a walkthrough

1. Provenance: every figure links to records

A total is never stored as a typed-in number. It is computed from records: payments, advances, approvals, indicator values and the evidence attached to them. Open any figure and you see the list of records it was built from, and from each record you can reach its documents and its history.

This means a report and the work behind it cannot drift apart. If a payment is corrected, every report that includes it reflects the correction, and the audit trail shows that it changed.

2. Named absence: unknown is not zero

Spreadsheets treat an empty cell as zero when they add it up. That turns "we don't know" into "nothing happened". We use named states instead:

A total that includes a missing value says so, and shows which parts are missing. It is not presented as complete.

3. Refusal by name

When access rules prevent someone from seeing a record, the product says that access was refused and why, rather than showing an empty screen or a smaller total that looks complete. An implementing unit sees only its own work; a donor viewer can read but not change. Either way, the person knows what they are not seeing.

4. AI that proposes, and a system that checks

The AI assistants can draft text, answer questions about your records and suggest entries. Every answer must cite the records it relies on. If the records do not support an answer, the assistant declines and says what is missing. An AI suggestion becomes a record only when a person accepts it, and the audit trail shows that it began as a suggestion.

5. Audit trail

Every create, change and approval is logged with who did it and when. Approvals are tied to the version that was approved, so a later change is visible as a change.

Worked example

Drilling one IFR line down to the payments

An illustration with fictional figures, of the kind you can repeat in the demo.

  1. The IFR line

    Component 2, Training, expenditure this quarter. The figure shows a marker: one of four districts is "not reported".

  2. Open the line

    It breaks down by district. Three districts list their spending. The fourth shows "not reported", not zero, so the reviewer knows the total is partial.

  3. Open a district

    The district total breaks down into individual payments and advance liquidations, each with date, payee, amount and approver.

  4. Open a payment

    The payment shows its supporting documents, the work plan activity it belongs to, and its history, including who approved it and when.

  5. Ask the assistant

    "Why is Component 2 lower than planned?" The assistant answers with citations to the work plan and the missing district report, or declines if the records do not explain it.

Commitments

What we will never do

Default a missing value to zero

Unknown stays visibly unknown, with its name, until someone records it.

Show a score that hides its inputs

No traffic light, rating or index appears without the figures that produced it being one click away.

Invent a figure

Neither the software nor its AI assistants will produce a number that is not computed from your records.

Some parts of the product are still in development, including a double-entry ledger and field monitoring with GPS-tagged photos. They will follow the same rules when they ship, and we label them as in development until then.

Open a figure and follow it down

The demo uses fictional data. Pick any total and open it to its records.

Open the live demo Request a walkthrough