Margin by job, with the loss-makers flagged
The request
The task request supplied to the adapter, alongside the workspace below. Runtime instructions, authorized tools and skills also affect the system. Containerized runs withhold host-side references; local mode is not an isolation boundary.
Fatima wants to see, job by job, what we have earned against what the job has cost us, with the ones losing money flagged. Save it as margins.xlsx with live formulas and write memo.md with the headline. Her note says what counts as revenue and cost.
Workspace
- note_from_the_principal.txt
- project_ledger_2026_h1.xlsx
- unbilled_time_to_date.csv
Expected deliverables
- margins.xlsx
- memo.md
Recorded acceptance checks
The table uses check names and outcomes from the frozen campaign ledger. Acceptance requires all required predicates to pass; a predicate may itself contain a tolerance or coverage threshold. Its scope is part of the measurement contract.
The frozen scorer's check set differs from this task's source declaration. These are different evaluator versions. The counts below report only checks actually recorded; a missing check is never counted as a pass.
| Recorded check | Type | Required | Failed / observed |
|---|---|---|---|
| margins.xlsx exists | file_exists | yes | 0 / 6 |
| live formulas | xlsx_has_formulas | yes | 0 / 6 |
| no error cells | xlsx_no_errors | yes | 0 / 6 |
| Harborview revenue (invoiced only) | xlsx_value_present | yes | 0 / 6 |
| Cedar Point cost (unbilled time in, credits netted) | xlsx_value_present | yes | 0 / 6 |
| worst job margin | xlsx_value_present | yes | 0 / 6 |
| total margin across all jobs | xlsx_value_present | yes | 0 / 6 |
| memo carries the studio totals and the worst margin | text_numbers_present | yes | 0 / 6 |
| memo names both loss-making jobs | text_contains_all | yes | 0 / 6 |
| memo says the worst job is losing money | custom | yes | 0 / 6 |
Current task declaration (10 checks)
From task.yaml. Original runner grading and frozen retrospective scoring are separate. Read the frozen source for the rules behind the reported results.
- margins.xlsx exists —
file_exists, required - live formulas —
xlsx_has_formulas, required - no error cells —
xlsx_no_errors, required - Harborview revenue (invoiced only) —
xlsx_value_present, required - Cedar Point cost (unbilled time in, credits netted) —
xlsx_value_present, required - worst job margin —
xlsx_value_present, required - total margin across all jobs —
xlsx_value_present, required - memo carries the studio totals and the worst margin —
text_numbers_present, required - memo names both loss-making jobs —
text_contains_all, required - memo says the worst job is losing money —
text_sentence_matches, required
Recorded results, desk release
| System | Repetitions | Frozen / 3 | Original / 3 | Median minutes | Mean model estimate (USD) |
|---|---|---|---|---|---|
| Proto / DeepSeek V4.1 Flash | 3 | 1 | 3.7 | 0.0449 | |
| Codex / GPT-5.6 sol | 3 | 0 | 6.1 | 0.7064 |
These are repeated executions on the released fixture, not independent new business cases. Cost is captured model usage priced under the recorded assumptions, excluding infrastructure and human review. Per-attempt check verdicts, raw and frozen verdicts, usage, receipt and artifact hashes for this task are in results/latest/attempts.jsonl under "task": "project-margin".
Recorded results, desk comparison, 28 September 2026
| System | Repetitions | Frozen / 3 | Original / 3 | Median minutes |
|---|---|---|---|---|
| Proto / GPT-6 sol | 3 | 0 | 2.4 | |
| Codex / GPT-6 sol | 3 | 0 | 3.2 | |
| Proto / DeepSeek V4.1 Flash | 3 | 1 | 1.4 |
A separate campaign under the same frozen scorer, not pooled with the release. Conditions and data.