Assignment Success

“Success isn’t ‘looks done.’ Success is: every rule the owner declared got satisfied, and no game-changer flag is on the board.” — Mr. Defendable

What “success” means under the rulebook

The referee does not judge whether the work feels finished. It applies the declared rulebook (the Flight Sheet) and checks one thing: did every REQUIRED declared rule pass, with no high-tier flag raised?

Concretely, an assignment is a success when all of these hold:

Required output fields present — every field the sheet’s required_output_schema.required lists is in the submission (required_fields_present passes).
Schema / structure intact — declared deterministic_checks (type, value-in-set, required-keys on array items, expected value) all pass.
Math re-derivation within tolerance — every calculation is recomputed from its own inputs + formula and lands inside the declared penalty band (no ≥ critical_rel or high-dollar miss).
Declared policy gates pass — every structured rule (the rule DSL — thresholds, cross-field, conditionals) the sheet declared evaluates True.
No high-tier flag — nothing escalated to high, so the verdict severity is honey.

Tied to the real verdict fields

Success maps to a specific verdict shape (see Verdict JSON):

Field	Success value
`outcome`	`pass`
`severity`	`honey`
`client_ready`	`"Yes"` (or `"Yes — minor notes only"` when only low-tier nits are present)
`risk_breakdown.high`	`[]` — empty

A single high-tier flag flips severity to propolis and outcome to fail — that is not a success, no matter how complete the rest of the submission looks. Mid-tier flags land at jelly / risk — usable with limitations or after repair, but not a clean pass.

And the final gate: a human still approves. A receipt only mints on decision='approved'. Success under the books-and-records doctrine is honey verdict + human approval, not the machine score alone.

Roadmap

🐝 Operator-grade · books and records · to the shed.