Owl, hands on.
Nine working mechanisms — the same rules the production composers enforce, on sandbox data in your browser; nothing you type is sent anywhere. Notice what the machine refuses to do: that is the product.
1 · Score one domain’s continuous survey readiness
mirrors the production readiness composer · sandbox
the rule it enforces, cited
The rule is continuous readiness rather than a point-in-time mock survey; the production composer attaches this citation to every score it returns.“Continuous survey readiness vs point-in-time mock survey; Joint Commission "always survey-ready" expectation; accreditation continuous-compliance model — readiness is a running state, not an annual event.”
Readiness is a running state, not an annual event. The composer starts from your mock-survey pass rate for a domain and subtracts what you already know is open: −6 per open finding, −4 per gap, −2 per stale evidence item, then clamps to 0–100 and rounds. The bands are exact: ready ≥ 90 · watch 75–89 · at-risk < 75.
The production module’s own honest limit, quoted: “Honest limit: the score reflects the signals supplied — garbage in, garbage out; it is a prioritization aid, not a guarantee of survey outcome.” A readiness indicator, never compliance evidence — and the sentence is documentation, not an enforced control.
2 · Place a finding on the SAFER matrix
mirrors the production placement composer · likelihood × scope → priority band · sandbox
the rule it enforces, cited
The rule is the accrediting body’s own placement grid — likelihood to harm crossed with scope — and the production composer carries this citation.“The Joint Commission SAFER™ matrix (2017–) — every Requirement for Improvement is placed by likelihood to harm a patient/staff/visitor (low/moderate/high) × scope (limited/pattern/widespread); placement drives Evidence of Standards Compliance expectations.”
A cell is a system priority, not an individual score — remediation routes through a plan of correction, not blame.
3 · Run an Appendix Q immediate-jeopardy assessment
mirrors the production IJ composer · CMS Core Appendix Q (2019) · honest both ways
the rule it enforces, cited
The rule is all-three-or-nothing: two of the three Appendix Q components is not immediate jeopardy.“CMS State Operations Manual, Core Appendix Q (2019): Immediate Jeopardy requires ALL THREE — (1) noncompliance, (2) serious adverse outcome or likelihood thereof (incl. serious injury, serious harm, serious impairment, death, or serious psychosocial harm), and (3) need for immediate action. The removal plan must state the actions taken, dates, and how recipients are protected and effectiveness monitored.”
Declaring or lifting immediate jeopardy is the survey team’s authority — never the platform’s. The composer says is-IJ / not-IJ and shows its reasons.
4 · Model your registrar workload
peer-reviewed regression, J Trauma Nursing 2020 (PubMed 33156254) · modeled — labeled as such
the rule it enforces, cited
The rule here is a labeling rule: modeled is not measured.sandbox model — no production citation constant; the numbers on this card carry their own sources.
5 · Try to tamper with a receipt chain
SHA-256 hash chain via your browser’s WebCrypto · the same principle as the production decision ledger
the rule it enforces, cited
The rule is the chain arithmetic itself: the production verifier returns a verdict — verified, total, breaks — and attaches no citation field of its own.sandbox model — no production citation constant; the numbers on this card carry their own sources.
Each receipt’s hash covers its text and the previous hash. Edit any earlier receipt — then verify, and watch every receipt after it break.
6 · Hold a vendor AI model to its claims
mirrors the production accountability scorecard · FDA GMLP (2021) · prepare-only — it never disables anything
the rule it enforces, cited
The rule is prepare-only governance: a scorecard informs a named owner, and the production composer carries this citation.“FDA/Health Canada/MHRA Good Machine Learning Practice (2021); Joint Commission Responsible Use of AI in Healthcare (RUAIH) guidance; Nightingale calibration lineage (vendorModelAccountability).”
Your EHR’s sepsis model claims a sensitivity and a PPV. Adjudicate its alerts against observed outcomes and let the scorecard judge the claim. Observed below 80% of claimed is a flag; fewer than 30 adjudicated cases is not evidence.
A failing scorecard informs a named governance owner — the platform never gates another vendor’s clinical output on its own. Ungoverned ≠ fine: a model with no scorecard sits in watch, not ready.
7 · Band an infection ratio the way NHSN reads it
mirrors the production HAI benchmark composer · CDC NHSN / CMS Care Compare / SHEA-IDSA 2014 · sandbox
the rule it enforces, cited
The rule is the national benchmark set the ratio is read against; the production composer attaches this citation to every band and every refusal.“CDC NHSN HAI Metrics / CMS Hospital Care Compare / SHEA-IDSA 2014”
SIR below 1.0 means fewer infections than predicted for your patient mix. The cutoffs are exact: 0.5 and 1.0 and 1.5 each belong to the band above them.
8 · Watch the k≥5 floor protect a small cohort
the house anonymity floor · HIPAA de-identification / WORKER_BILL_OF_RIGHTS · masked, never fudged
the rule it enforces, cited
The rule is the k≥5 floor and the ban on individual attribution; production attaches this citation to the OWL_NAV_002 refusal.“WORKER_BILL_OF_RIGHTS — rollups must be aggregate-only with k>=5 floor; no individual staff attribution permitted in dashboard surfaces.”
Every aggregate the platform publishes suppresses any cell reflecting fewer than five people. Drag the cohort below five and watch the number itself disappear — a suppressed cell never even reports how small it is.
Subgroup A: n 9 · ppv 0.82. Slide subgroup B — under five it is suppressed and excluded from the equity math; at five or more its PPV gap (0.82 − 0.58 = 0.24 > 0.15) flags differential validity.
Two different protections: a small cell is quietly masked; individual attribution in an aggregate is refused outright — the whole rollup, not just the field.
9 · Compose a plan of correction a surveyor would accept
mirrors the production PoC composer · CMS-2567 / Joint Commission ESC · validates completeness — it does not approve
the rule it enforces, cited
The rule is what a plan of correction must contain before anyone will accept it; the production composer carries this citation.“CMS Form CMS-2567 Statement of Deficiencies & Plan of Correction; Joint Commission Evidence of Standards Compliance (ESC) — a POC must state corrective action, responsible party, completion date, and ongoing monitoring (sustainability). Findings framed just-culture.”
Blank the corrective actions or the responsible role and run it again — the gate names every missing element. A complete plan auto-composes the sustainability / monitoring section surveyors reject plans for lacking.
These nine run on sandbox data by design — the same refusal the SAFER demo shows guards the real evidence chain in production. To see the mechanisms run against your own designation portfolio, start with the digital pilot.
Leave with the receipt
Everything you did on this page — every input you moved, every button you pressed, every refusal you triggered — can be hash-chained in your browser and handed to you as a file, so you can see the shape of the evidence a receipt chain produces.
Sandbox artifact: hash-chained in your browser, UNSIGNED, and it records only what you did on this page — nothing was transmitted. The production evidence packet is different: it is produced server-side on the audited evidence spine and carries a signed attestation. This file only shows the shape of a receipt chain.
Direct links: each mechanism’s title carries a copy-link button, so you can send one mechanism to a colleague. The guided tours open straight from a URL — ?tour=quality · ?tour=ciso · ?tour=investor — and every copied link points at nightingaleos.com, never at the file you may be reading this from.