A pilot charter: entry, stop and reconsideration conditions
6 min read
Lesson objectives
- Decide pilot entry, containment and stop rules from named evidence and accountable review.
Please sign in to use all the tools
Once you sign in, the reading tools are yours: larger or smaller text, typeface, line spacing, focus mode, highlighting and bold, and notes saved to your account and gathered in My study.
A bounded pilot is not a launch without accountability. Its charter defines the function, users, period and fallback, then names evidence for entry, containment and stopping. An entry condition shows that a pathway works well enough to begin limited learning; a stop condition protects users when a critical assumption fails. A single completion percentage is insufficient because non-urgent scheduling and potentially time-sensitive follow-up messaging have different risks. Passing a synthetic test does not establish safety in ordinary operation. It may permit monitored limited use when a real ability to stop exists.
Write a separate decision for each function. For non-urgent rescheduling, check task completion, whether failure is visible to the user, and whether a backup channel accepts a request without losing it. For follow-up messages, check receipt ownership and escalation when the duty holder is absent. Define an early operational signal, such as requests without a final status, and a later outcome, such as a complaint verified as arising from a dropped request. Specify data source and cutoff, who reads the signal daily, and who can stop the function. Do not present a locally chosen threshold as an international standard. If the institution adopts an operational trigger, record why, how it was tested, and who may revise it. A potentially consequential safety event goes to its qualified human pathway without waiting for a numerical trigger to be crossed.
The repaired charter covers one non-urgent appointment change per request. The digital channel displays a tracking number and request status, and a failure opens a backup ticket visible to the operations owner. Operations reviews open requests daily; the technology owner repairs transport; authorized leadership stops the function if a request cannot be traced or a report disappears from both paths. Follow-up messaging remains outside this pilot until duty-holder acknowledgment and an absence pathway are demonstrated. Before scheduling starts, retest the four failed cases and show either closure of the fourth or a visible open status with a named owner. Merely changing the screen to 'done' is insufficient. At the end of ten days, compare completed and pending requests and complaints under stable definitions, then decide whether to continue narrowly, redesign or stop.
Daily review does not imply that every event can wait until the next day. The charter separates non-urgent appointment requests reviewed on a service cycle communicated to users from messages whose failed receipt calls for prompt qualified human handling through the institution's pathway. Do not transfer a 90% figure from messaging tests into a scheduling standard: 18÷20 describes acknowledgments of test messages and says nothing about appointment quality. Write a measurement table with requests started, completed digitally, sent to backup, closed there and still open; separately record messages dispatched, arrived, acknowledged and escalated. The branches of each pathway should reconcile with its starting count or show why they do not. If a user presses submit twice, distinguish attempts from logical requests so a changing denominator cannot create an artificial gain or loss.
An operational stop rule states what happens next, rather than merely saying 'notify the committee'. If operations cannot find a request in either channel, pause new intake for the affected function, preserve identifiers for open requests, show a suitable alternative contact route to users, and have the technology owner investigate the gap. If a follow-up message loses its route, use the institution's escalation path to a qualified duty holder with minimum necessary data; the technical committee does not decide the health response. Restart requires a test of the actual failure cause, reconciliation of open cases, and a recorded decision by the authorized owner. A server restart is insufficient while old cases remain unaccounted for. If a new failure pattern appears during the pilot, review the gate then rather than waiting for the ten-day end point.
The worked example completed 26 of 30 scheduling attempts digitally. Four went to backup and three closed, leaving one unresolved. All 20 messages reached an inbox, but only 18 had acknowledgment; those are different events.
Delay scheduling until the fourth request is reconciled, and do not begin messaging until duty-holder acknowledgment and a backup handoff are demonstrated. If repair succeeds, run a bounded scheduling pilot with daily open-request review and a stop rule.
Attempts, messages and the period are fictional. The gates are operational choices for this case, not universal thresholds. Synthetic success cannot establish health outcomes or performance under every real operating condition.
If the fourth backup request closed with a clear handoff record, all four failed attempts would be accounted for and the bounded scheduling pilot could start. Two messages without duty-holder acknowledgment would still block the messaging function.
Lesson summary
The worked example completed 26 of 30 scheduling attempts digitally. Four went to backup and three closed, leaving one unresolved. All 20 messages reached an inbox, but only 18 had acknowledgment; those are different events. Delay scheduling until the fourth request is reconciled, and do not begin messaging until duty-holder acknowledgment and a backup handoff are demonstrated. If repair succeeds, run a bounded scheduling pilot with daily open-request review and a stop rule. Attempts, messages and the period are fictional. The gates are operational choices for this case, not universal thresholds. Synthetic success cannot establish health outcomes or performance under every real operating condition. If the fourth backup request closed with a clear handoff record, all four failed attempts would be accounted for and the bounded scheduling pilot could start. Two messages without duty-holder acknowledgment would still block the messaging function.