The method is public. Weights stay visible.

Survey Posture Score, Methodology (sps-v1)

Status: canonical. This is THE facility survey score across every Credalyst product (Survey Command, Due Diligence, Organizational Capability). Its version, sources, weights, and absence rules are published below.

Principles

• Deterministic. Same inputs, same score, byte for byte. No wall clock (the caller supplies asOf), no randomness, and no LLM anywhere in the path.
• Explainable. Every component reports basis strings citing the exact values used and the arithmetic applied. There is no hidden weighting.
• Traceable. Every component carries EvidenceCoordinates ({source, snapshotId, recordId, field}) into immutable, dated CMS snapshots (dated evidence snapshots). A score can be reproduced as-of any snapshot set; the snapshotIds on the score are the reproducibility anchor.
• Honest about absence. A component whose evidence is not on record reports observed: false, contributes 0 points, and says why. Missing evidence is never scored as safety or as risk.

Inputs (all public CMS data, zero PHI)

Components (total 100)

1. F-tag signature, 25 points

Severity/scope weight per deficiency from the CMS A, L grid: A, C = 1, D, F = 2, G, I = 3, J, L = 4 (severity band; scope rides within the band). Each citation is recency-decayed by survey age at asOf: ≤12 months ×1.0, ≤24 ×0.6, ≤36 ×0.3, older ×0.15. base = min(20, 0.8 × Σ weight × decay). Recurrence bonus (+1 each, max +3): the same F-tag cited on ≥2 distinct survey dates. Cluster bonus (+1 each, max +2): a CMS deficiency category with ≥3 citations.

2. Enforcement trajectory, 15 points

Deficiencies are grouped into survey cycles (CMS inspection_cycle when present, else by survey age: ≤12mo → cycle 1, ≤24 → 2, else 3) and severity-weighted per cycle. Current ≥1.5× prior cycle → +8; merely above prior → +5; strictly rising across three cycles → +3; complaint-survey share of the current cycle ≥30% → +4. Capped at 15. Fewer than two observed cycles → observed: false (no slope from one point).

3. SFF proximity, 15 points

Verbatim special_focus_status: SFF → 15; any non-empty candidate status → 12. Otherwise a published proxy for candidate-pool distance: 1-star health inspection rating → 8; 2-star → 4 (the SFF candidate list draws from the worst health-inspection scores). No profile on record → observed: false.

4. Financial exposure, 15 points

Realized CMP total (fact, from the Penalties dataset): ≥$250k → 8; ≥$100k → 6; ≥$25k → 4; >$0 → 2. Any payment-denial action → +4. Any penalty dated within 24 months of asOf → +3. Capped at 15. This component reports realized enforcement only; modeled forward exposure lives in the exposure engine, not the score.

5. Staffing signal, 10 points

Reported total-nurse HPRD below the published 3.48 federal minimum staffing standard (89 FR 40876) → +3; else at/below p25 of the state cohort → +2. Nursing turnover at/above p75 of the state cohort → +3. PBJ contract-labor share of nursing hours (Σ _ctr hours ÷ Σ hours over the observed days): ≥20% → +3; ≥10% → +1. Capped at 10. No staffing evidence at all → observed: false.

6. Peer context, 10 points

Percentile of health-inspection badness (6 − rating) within the latest ≥20-facility state cohort snapshot: ≥p90 → 10; ≥p75 → 7; ≥p50 → 4. No qualifying cohort snapshot → observed: false (never scored against a fabricated peer group).

7. Deficiency-cycle window, 10 points

Months from the latest standard (non-complaint) survey to asOf, against the statutory survey cadence (42 USC 1395i-3(g)(2)(A): ≤15.9 months between standard surveys): ≥15.9 → 10 (survey due); 12, 15.9 → 7 (inside the window); 9, 12 → 4; else 0. No standard survey on record → observed: false.

Bands

0, 19 low · 20, 39 guarded · 40, 59 elevated · 60, 79 high · 80, 100 critical

Bands are presentation over the transparent total; the components are the substance.

Operator fingerprint (sps-v1)

Entity resolution v1: exact normalized (trimmed, uppercased, whitespace-collapsed) owner_name equality across the latest Ownership snapshot per CCN. The fingerprint reports the portfolio CCN list, mean and worst Survey Posture totals, band distribution, and SFF / SFF-candidate counts, every edge carrying its ownership coordinates. Known limitation (stated, not hidden): name-variant owners (e.g. punctuation differences) resolve as distinct entities until a stronger resolution rule ships in a future version.

Versioning

Any change to weights, thresholds, decay, or inputs bumps methodologyVersion (sps-v1 → sps-v2) and this document. Scores are comparable only within a version.

The founding law

The company constitution was ratified in July 2026 and closed with three founder corrections. A fourth correction — the deepest — added its first page: the Law of Earned Movement. Amendments require a founder correction of equal weight, never drift.

• We do not manufacture urgency. We do not manufacture certainty. We do not manufacture activity.
• We advance only what reality has earned the right to move.
• When reality does not justify movement, we remain quiet. Silence is not inactivity: silence is honesty.

How a capability earns its name

Inside Credalyst, no capability is declared: it settles. A capability advances through stages — designed, running, verified, compounding — and advancement is decided by code against a burden of proof, not by a slide.

• Verified requires EVERY requirement met: at least 5 governed settlements, evidence coverage of 70% or more, invalidated experiments at 20% or less, independent evidence subjects, at least 2 independent settlement windows, and no recent contradictory trend.
• Compounding additionally requires at least 10 settlements, 3 independent windows, and 70% or more of predictions validated.
• Missing evidence is never scored as safety or as risk — the same rule this document applies to your facility.

Capability Settlement v1

Organizational capability is read from a linked record: a governed decision, a frozen prediction, recorded execution, outcome evidence, and a dated settlement. Activity alone does not satisfy the standard.

• Settlement states: validated, contradicted, invalidated, or honestly unsettled. Invalidated means the experiment could not test the claim; it does not count as a contradiction.
• Inputs: decision, prediction, and subject identity; observation window; metric; observed value; outcome references; settlement date; and the action’s traceable receipt.
• Known limit: one settlement does not guarantee a future result. Capabilities without a governed metric and canonical source remain running, never verified by inference.
• Change record: v1, July 2026 — required identity linkage, explicit burden of proof, and separation between an invalid experiment and a contradicted prediction.

Provenance

This section is refreshed editorially from the operating records. Nothing here is generated or projected; if an artifact stops being true, it is removed.