Screen 1 of 8§1 in the guidelines ↗Passed ✓
You are writing the answer key
An AI takes a finance brief and produces a deck and a model. You write the rubric; an AI judge works through it line by line, marking pass or fail. If your rubric passes work a partner would fail, the rubric is wrong — not the partner.
Six words, used precisely — tap each
The flow — tap the step where your work happens
You write before you see how the agent did — a rubric written against an attempt you have read grades that attempt, not the task.
ST1 is the pilot. Its pass rates set the pattern — the cheapest time in the project.
Screen 2 of 8§2 in the guidelines ↗Passed ✓
First decide what it is testing
Before any rows: read the prompt and golden together and write the intent in one sentence — what capability is this testing? Then sort. Crucial: every good build has it. Not crucial: the prompt left it open. Never test: the agent was handed it.
The prompt. To support the take-private thesis for Northvale, the IC has asked us to build the market context and the operating analytics. Build out the 'market build' tab sizing off the market inputs, and the 'KPI' and 'KPI benchmarking' sheets that show how Northvale performs in the industry. This will form the base evidence for the growth and margin story underpinning the investment case.
What it is testing: can the agent size the market from the given inputs and turn Northvale's own financials into KPIs that stand up against the industry — evidence a growth and margin thesis can rest on.
Every crucial line survives a completely different build; anything that would fail another good build is not crucial.
Screen 3 of 8§3 in the guidelines ↗Passed ✓
Six columns, nothing else
Build row 4 of the worked example. The check: the revenue chart on the slide titled Business Model & Historical Operating Performance states FY2023A revenue of £544m and FY2024A revenue of £581m.
Integers only
| # | Section / Sheet | Scope | Rubric | Category | Tier |
|---|---|---|---|---|---|
Row 4, as §10 wrote it:
The revenue chart states FY2023A revenue of £544m and FY2024A revenue of £581m.
Screen 4 of 8§4 in the guidelines ↗Passed ✓
One sentence that stands on its own
A criterion says what must appear, names what it points at, and carries the expected answer. This one is broken — fix it until the checker goes quiet.
§4's version — specific about what must appear, silent on what to write:
The commentary attributes the FY2025A EBITDA margin movement to at least two specific operational drivers, each named and quantified.
The judge counts named drivers and checks each carries a number. Nobody passes by copying the rubric.
Screen 5 of 8§5 in the guidelines ↗Passed ✓
One row, or two?
One test decides: would you ever want to pass one part and fail the other? If the report should tell the failures apart, split. If splitting lets a broken build keep scoring, keep it whole.
Screen 6 of 8§6 in the guidelines ↗Passed ✓
Tolerance follows the source
Not how confident you feel. Pick where the number came from, type the value, and read the fragment the judge needs — the figure and both bounds, written out. Visit all five rungs.
Too narrow fails work that is right. Too wide passes work that is wrong. Neither is safe — the ladder decides.
Screen 7 of 8§7 in the guidelines ↗Passed ✓
Critical gates
One failed Critical fails the sub-task — nineteen of twenty is not 95%. Critical takes all three: Mandatory — every correct answer has it. Objective — checkable in the work itself. Stable — separate judges agree. Miss one: Non-Critical. Sort these §10 rows.
At least five Criticals per sub-task. A strong attempt should still fail about one in five.
Screen 8 of 8§10 in the guidelines ↗Passed ✓
Write the row
From §10's prompt: the slide Historical Profitability & Capital Structure shows Net Debt / EBITDA. The golden has FY2025A at 3.5x — a ratio the analyst works out. Write the complete row.
§10's row 8:
The slide states FY2025A Net Debt / EBITDA of 3.5x (acceptable range 3.3x to 3.7x).
Self-check — tick each against your row:
That is the loop. Before a real handover: the five-minute read and the five gates — §9 ↗.