Practice
Practice Library
“What can institutions do?”
Fifteen governance patterns drawn from the PAN framework's lever catalog and the documented case histories: what each one changes in the system, who has the authority to pull it, what it looks like institutionally — and, for every pattern, what can backfire.
Structural patterns
Change the system's wiring: what flows where, at what volume.
Pace the pipeline
Backfire notedMatch the flow of AI output to the real capacity of the people who must check it — throughput honesty as a safety control.
Put a verifier on the agent
Backfire notedAttach an independent checking step to the least-supervised operator — in PAN runs, the single highest-leverage move.
Improve the model
Backfire notedThe default instinct — buy or build a better model — is a real lever with an honest, limited reach.
Provenance labeling
Stamp unverified and AI-generated records so people and systems down-weight them instead of inheriting them as fact.
Data minimization
Backfire notedWrite less and keep less: the least data that does the job is the least there is to leak, to contaminate, and to purge later.
Peer-edge governance
Govern the sideways pathways — operator-to-operator forwarding and store-to-store replication — that multiply everything else.
Procedural patterns
Change what people are required to do, and are given capacity to do.
Risk-tiered oversight
Spend scarce verification where stakes are highest: mandatory correction plus ground-truth checks for the high-stakes tier.
Human-in-the-loop write gating
Backfire notedRequire verified sign-off before anything enters the official record — govern the write, not just the read.
Framing and mirroring reduction
Train people and agents to prompt without leading — cutting the loop where the user's belief manufactures its own confirmation.
Deskilling-arrest mandate
A scheduled step — recertification, blind re-checks, unassisted rotations — that stops the quiet drift toward rubber-stamping.
Understand the system
Fund continuous learning about what the AI deployment is actually doing across the sociotechnical system, so a limited budget can target hidden dynamics and real failure modes instead of guessing.
Feedback patterns
Change how the system responds to what monitoring observes.
Deployment circuit-breaker
Backfire notedA pre-authorized clamp-down that fires when a measured signal crosses a threshold — the control Robodebt never had.
State-feedback vigilance
Correction effort that rises when observed error rises, and relaxes when it subsides — oversight as a thermostat, not a setting.
Authority patterns
Change who may act — gates, grants, cadences, and review powers.
Connection authorization
Backfire notedNo data pathway exists until someone with authority approved it — the general edge-level access control.
Oversight cadence & retrospectives
Standing review on a mandated schedule — boards, audits, retrospectives — modeled on the actors who actually ended the documented failures.
Vendor quality gate
Procurement as governance: transparency, evaluation access, and exit terms decided while the institution still has leverage.
Conformity assessment gate
Backfire notedA formal pre-deployment authorization step — honestly framed: it changes who may act, not how the system behaves.
Patterns describe what institutions can do; the PAN Lab is where selected patterns can be stress-tested against scenarios — illustratively, with limits stated. For the documented histories behind them, see the Domain Atlas.