Risk Tier · low · mid · high

The doctrine

“Not all penalties are the same. The flight sheet knows what we’re looking for, and so does the owner reading the report.” — Mr. Defendable

Each rule on a Flight Sheet declares a tier — its pre-weighted risk weight. Flags inherit the tier of the rule they raised. The owner’s report ranks flags high → low so a catastrophic event (a CRE 5-cap penciled at 10-cap, a $5k check that should be $100) surfaces before a citation typo.

The three tiers

Tier	Weight	When to declare it	Example flags
high	5×	Game-changing miss — catastrophic if undetected.	Math off ≥ 10% on a monetary value · core lending gate (DSCR < 1.20) · structure / schema breakage · missing required input · fabricated evidence
mid	2×	Material but recoverable — should not pass review.	Math miss in the 5–10% band · evidence missing for one claim · secondary policy gate violated · assumption unlabeled
low	1×	Nit — quality polish, not a decision.	Citation typo · format nit · cosmetic value out of canonical case

How tier drives severity

Severity	Trigger
honey	No flags present + human approved.
jelly	Mid- or low-tier flags only.
propolis	Any high-tier flag.

A single high-tier flag is enough for propolis — that’s the point of the tier. A high-tier rule is a rule the owner cares about catastrophically.

How magnitude becomes tier (variable penalty)

For math and approx checks, the engine maps the size of the miss to a tier (the football “spot of the foul”):

within 1%        → pass    (rounding / immaterial)
1 – 5%           → low      (honey · "minor variance")
5 – 10%          → mid      (jelly)
≥ 10% rel        → high     (propolis · "material miss")
material $ miss  → high     (propolis · "high-dollar impact")

Magnitude bands are declarable per Flight Sheet via eval_spec.penalty. The real keys (from app/executor.py _penalty()) with their defaults:

Key	Default	Meaning
`tol_rel`	`0.01`	At/under this relative miss → pass (rounding/immaterial).
`low_rel`	`0.05`	At/over this → at least a low flag (below it but over `tol_rel` is low).
`critical_rel`	`0.10`	At/over this relative miss → high flag.
`critical_abs`	`1000.0`	Absolute-dollar floor for the high-dollar escalation.
`abs_floor_rel`	`0.02`	Relative floor that must also be met for the high-dollar escalation to fire.

A miss escalates to high when rel ≥ critical_rel, or when the value is monetary and abs_miss ≥ critical_abs and rel ≥ abs_floor_rel (the high-dollar rule). Between low_rel and critical_rel the flag is mid; between tol_rel and low_rel it is low.

Risk breakdown — the grouped flags

Every verdict carries a risk_breakdown — the raised flags grouped by tier, each tier a list of flag labels (not a count):

"risk_breakdown": { "high": ["DSCR gate"], "mid": ["Comp set staleness"], "low": [] }

The grouped flags are the read for the owner: “one game-changer (the DSCR gate), one material issue, zero nits — and here is exactly which rule each one is.” Combined with the ranked findings, the owner can decide the next move in seconds.

🐝 Tier the flag. Rank the report. Catastrophic events first. To the shed.