Sample Verdict

A live verdict from cre_memo_dscr_ltv_v1 (the DSCR and LTV Analysis Flight Sheet) on a deal whose math is correct but whose DSCR comes in below the declared lending gate.

The referee returns the rolled-up verdict plus the full list of applied rules. Below is the verdict object as emitted by GET /runs/{id}/verdict:

{
  "outcome": "fail",
  "summary": "Math reconciled; DSCR below the declared 1.20 gate.",
  "score": 0.875,
  "score_100": 88,
  "severity": "propolis",
  "client_ready": "Not client-ready: one high-tier policy flag.",
  "recommended_action": "review",
  "checks_passed": 7,
  "checks_failed": 1,
  "risk_breakdown": { "high": ["dscr_gate"], "mid": [], "low": [] }
}

The applied rules (GET /runs/{id}/checks) — each rule passes or raises a flag, deterministically. check_key is the rule id; severity here is the rule’s declared pre-weight (a RuleSeverity), which tier_of() normalizes to a tier:

[
  { "check_key": "sections_present",   "label": "Required sections present", "category": "structure", "status": "pass", "severity": "critical",    "detail": null },
  { "check_key": "citations_present",  "label": "Evidence is referenced",    "category": "evidence",  "status": "pass", "severity": "noncritical", "detail": null },
  { "check_key": "math_ltv",           "label": "LTV recompute",             "category": "math",      "status": "pass", "severity": "critical",    "detail": "ltv=0.7500" },
  { "check_key": "math_debt_service",  "label": "Debt service recompute",    "category": "math",      "status": "pass", "severity": "critical",    "detail": "annual_debt_service=703452" },
  { "check_key": "math_dscr",          "label": "DSCR recompute",            "category": "math",      "status": "pass", "severity": "critical",    "detail": "dscr=1.022" },
  { "check_key": "ltv_gate",           "label": "LTV <= 0.80",               "category": "policy",    "status": "pass", "severity": "noncritical", "detail": "ltv=0.7500 <= 0.80" },
  { "check_key": "coc_nonneg",         "label": "Cash-on-cash >= 0",         "category": "policy",    "status": "pass", "severity": "noncritical", "detail": null }
]

And the one flag — the findings array is the subset of checks whose status is flag:

[
  {
    "check_key": "dscr_gate",
    "label": "DSCR >= 1.20",
    "category": "policy",
    "status": "flag",
    "severity": "critical",
    "detail": "DSCR recomputed 1.022 — gate 1.20 — MISMATCH"
  }
]

How to read this

The math is right. math_ltv, math_debt_service, math_dscr all pass — the referee recomputed every claimed number from the agent’s own inputs and they reconciled.
The deal does not pencil under the declared gate. dscr_gate flagged because the recomputed DSCR (1.022) is below the lane’s declared threshold (1.20).
outcome: "fail", severity: "propolis" because the flag normalizes to a high tier. One high-tier flag is enough. (outcome is pass | risk | fail; severity is honey | jelly | propolis.)
This is a deal-finding, not a rework. A policy-category flag means the math is right and the rule says no. Report it to the client.
score_100: 88, score: 0.875 — 7 of 8 declared rules satisfied, weighted by tier. Score is mechanical, not a quality grade.
risk_breakdown is lists of check_keys by tier — { high: ["dscr_gate"], mid: [], low: [] } — not counts. client_ready is a string readiness note.
recommended_action: "review" — the rulebook surfaces this for client conversation, not for resubmission.

🐝 Math right · gate failed · deal-finding · to the client · to the shed.