A pilot only proves a sample, not a cadence
Corridor and confined-network pilots are often written as “we ran it once”. For operations and safety review, one success only shows that under a given permit, crew, weather and kit state, the job did not abort. Routine asks another question: can the same segment object be re-entered, re-evidenced and re-reviewed inside an agreed window, and still recovered if it fails.
Treat cadence as an asset calendar, not a flight-show timetable. The smallest unit is segment or bay, trade and evidence type — not how many times we flew this week. If a pilot week produces only highlight clips with no asset code or defect-order fields, later routine will be dragged down by its own footage.
Pre-entry reconnaissance, main collection, same-day coarse sort and cross-crew review should be frozen at the instruction-manual layer so pitch language cannot drift. Confined-space kit fits pre-entry and gap reconnaissance. It does not fit being advertised as a statutory-inspection substitute. Translating pilot counts into coverage while capability is unverified sits outside the evidence gate.
Window density matters more than a pretty trajectory
Routine usually fails because the window is dishonest, not because the algorithm is old. Night windows, overlapping construction and wet-season standing water all change enterability. Mark the days you cannot enter first, then schedule reconnaissance. Otherwise the pilot looks full and the operating quarter is blank.
In a Hong Kong–Shenzhen and Greater Bay Area setting, reviewers know permits, standby and recovery better than consumer-drone endurance numbers. Write cadence as: permit prep, pre-entry check, limited-duration reconnaissance, mandatory recovery, evidence ingest, review deadline. That is closer to engineering language than a smooth trajectory plot.
Sensor and video timestamps must align. Recovery records, interrupt drills, and an explicit “this system does not claim” list are the minimum evidence pack for graduating a pilot. Missing any one item, stay in the pilot. Do not graduate early in public-relations language.
How to read the article back to engineering
This article is a method note, not a field scorecard. It invents no customers, prices or unpublished trial conclusions. Reading an article is not a confined-space permit. If a public standard or funding scheme appears, treat it as a question list: what must our own rig answer, and what must public copy forbid.
Graduation from pilot to routine should let a second engineer retell window, object and failed recovery from the file alone, without sitting through a pitch. That is cadence. That is not a show count.
Fix on one page in a crew meeting: enterable segments this week, reasons not to enter, evidence fields, reviewer. One page survives a handover better than a long report. Copying an external figure into a spec sheet without a rig check conflicts with the evidence page.
First executable crew rule
Write object code, entry window, recovery owner and evidence fields on the same table. If the table does not exist, cadence does not exist. Review must have a deadline. Timeout may only escalate or reopen. It may not close in a chat tool. Interrupts, wet slip, a weak link and a gas alarm are learnable events. They are not an embarrassed silence. Public copy stays inside the evidence gate: unverified capability is not coverage; a work order that has not happened is not a customer success.
The English site carries the full method set: system architecture, safety bounds, target specifications, the 18-month roadmap and HK–SZ operations.