Issue 20 · Crosswalk

Who Checks the AI in Your Medical Record?

Every Structure claim mapped to the outside pressure that supports, bounds, or cannot reach it.

CrosswalkStructure ↔ Reference

Reference Dependency Crosswalk

This connects claims to outside pressure, not to certainty.

For every Structure claim, this table shows whether Reference supplies outside-world support or marks the dependency as synthesis, scope limit, or process boundary.

Structure claimOutside supportReferenceWhat it does and does not cover
SL-020-01
Installation and reliance are different events; reliance must be earned and renewable.
Partly satisfied; standard is arguedRL-020-01, RL-020-02Adoption channels and performance-change risks are documented. Treating deployment and renewal as recurring evidence decisions is the essay’s standard.
SL-020-02
Quiet failures can hide behind familiar interfaces.
Satisfied, class-specificRL-020-02, RL-020-04, RL-020-05Calibration, note errors, automation effects, and alert behavior are sourced separately. They do not combine into one universal failure rate.
SL-020-03
The Epic sepsis chronology shows why local validation and continuing checks matter.
Satisfied with version/site limitsRL-020-03The v1 validation, Michigan pause, and later local-tuning record are supported, but they are not one controlled comparison and later performance is not retroactive validation.
SL-020-04
Hospital, clinician, and patient occupy different positions in the reliance chain.
Partly satisfied; conceptual synthesisRL-020-01, RL-020-06Deployment channels and patient attitudes are documented. The three-position distinction and the choice to call patient status exposure rather than decision are the essay’s synthesis.
SL-020-05
Behavioral measures do not directly reveal trust or review quality.
Satisfied as a measurement limitRL-020-04, RL-020-05The studies observe outputs, overrides, timing, or outcomes. None directly measures private belief or the quality of each review.
SL-020-06
The hospital owns first-line operational checking; vendors remain answerable.
Argued from documented control asymmetryRL-020-07, RL-020-08Contracts, disclosure constraints, doctrine, and outside rules support the control map. The responsibility assignment remains the essay’s governance judgment.
SL-020-07
Checking must scale with the tool’s power to change care or the permanent record.
Structural / synthesisRL-020-02, RL-020-04, RL-020-08Different tool classes and risks are documented. The stake-matched checking framework is a proposed design, not an existing uniform standard.
SL-020-08
Outside governance exists, but it does not replace local checking.
Satisfied as status; argued as sufficiencyRL-020-08Final rules, a proposed rule, and an announced initiative are status-checked. Their insufficiency for local performance is a reasoned limit, not a quoted finding.
SL-020-09
Process is not proof.
Structural, not externalThis recordThe record exposes choices, corrections, gaps, and review pressure. It does not make the published claims true.