Stop 9 of 9 · Weekly
Farida's month and the capstone
The three scheduled routines that carry twenty client files through a month, then the graded capstone: one client from audit to advisory pack with the guardrail log.
What it is
Eight stops of parts, arranged into a month that runs whether or not Farida has a good week. Three routines carry it, and the capstone proves the whole chain on one real client.
The three routines
| When | What it does unattended |
|---|---|
| Every day 06:00 | The anomaly delta: signals across twenty files that are new or still unresolved since yesterday, written to one file |
| Wednesday 05:00 | The categorisation delegation, proposals and questions per client, nothing posted |
| Second working day 06:00 | The close report: status, exceptions and days since last close for all twenty, ranked |
Each of the three reads, compares and drafts. None of them posts a transaction, reclasses an account or changes a bank rule, because that is the rule the whole practice rests on and a timer is where people let it slip.
Every day at 06:00, run the five-signal sweep across every connected client file, read only, and compare against yesterday output. Write one file with two sections: signals that are new since yesterday, and signals that were raised more than three days ago and are still unresolved, each with client code, transactions, dates, amounts and dollars. Sort both by dollars. Contact nobody, post nothing, reclass nothing and change no bank rule.
On the second working day of the month at 06:00, for every connected client file, read only: return one line per client with close status, uncategorised rows, unreconciled accounts, open exceptions from the fifteen checks, open client questions, and days since the last close. Rank by open exceptions then by dollars. Add three lines at the end: how many clients are ready to close, how many are blocked on a client answer, and how many are blocked on us.
The capstone
Pick one real client and run a full month end to end, keeping the evidence. Five artefacts: the ten-minute audit with its findings table, the categorisation review queue with your accept and reject marks, the fifteen-check exceptions report, the advisory pack after its licence pass, and the guardrail log covering every proposal in the month.
[upload the five capstone artefacts: audit, marked review queue, exceptions report, advisory pack, guardrail log] Grade these against the rubric below out of 100, scoring each criterion separately and naming the evidence you used. For every point not awarded, say exactly what is missing and what would earn it. Be strict about figures with no transactions behind them, about anything posted without a logged approval, and about any sentence that crosses the practice licence.
The rubric
| Criterion | Points | Full marks means |
|---|---|---|
| Findings are verifiable | 20 | Every figure has transactions, dates and the report it came from, and nothing unverified reaches the client |
| Guardrail integrity | 25 | Nothing posted, reclassed or rule-changed without a logged human approval, and at least one refusal with a written reason |
| Review is a sample, not a scan | 15 | Unresolved, medium, over threshold and a random draw, with an acceptance rate recorded per tier |
| Exceptions are honest | 15 | Every failed check carries its evidence, and the report claims consistency rather than completeness |
| The pack earns the meeting | 25 | Cash, margin and two decisions, each tied to transactions, with the licence line held and questions routed |
Eighty passes. Below eighty the gap is almost always the same two: a figure with nothing behind it, and something posted before it was approved.