Scheduled Detector Minute-Window Alignment Automation: Ten Detectors, Zero Firings, Ever

Building a rule that says "run this at 9am" is only half of what a scheduled system needs to get right — the other half is making sure the actual mechanism doing the checking runs AT a moment that can genuinely satisfy that rule. A scheduler that sweeps periodically, rather than firing at the exact literal instant configured, needs its own check to account for that gap deliberately; comparing against an exact instant when the checking mechanism itself never lands on that exact instant is a rule that reads as correct and can never actually be satisfied.

The defect is stated directly in the fix's own documentation: an "hour matched" check inside these detector functions compared the current hour and minute, formatted as exact text, against an organization's own configured hour — a comparison that only succeeds if the check happens to run in the exact same minute the configured hour begins. The scheduler actually triggering these checks runs on a fixed periodic sweep landing at minutes 2, 12, 22, 32, 42 and 52 of every hour, never at minute 0 or 30. A configured hour like 9:00 or 9:30 is precisely the kind of value this scheduler can never land on exactly — meaning the comparison was written correctly, and could still never once succeed, for the most natural, common configuration values a person would actually choose.

The measured consequence, stated directly: a family of ten billing-adjacent detectors had zero firings recorded in their entire history, and a separate monthly content sequence due to begin the following month — with stages configured for specific days and times — would have launched exactly as silently mute, for the identical structural reason, if the underlying comparison hadn't been fixed first.

The fix replaces the exact-equality check with one shared function answering a different, correctly-scoped question: is the current moment within a 10-minute window starting at the configured hour? Ten minutes is deliberately wide enough that exactly one of the scheduler's own periodic sweeps is guaranteed to fall inside it, while a separate, already-existing deduplication mechanism prevents that same window from triggering the same action more than once. Because six separate functions each contained their own copy of the same broken comparison, and together represent tens of thousands of characters of largely identical code, the fix is applied as a single, programmatic, pattern-based rewrite across all six functions at once — reading each function's own current definition directly from the database, finding the specific broken comparison pattern inside it, and replacing it with a call to the new shared window function, function by function, in one pass. The same migration includes its own built-in negative-proof step: after applying the rewrite everywhere, it explicitly searches the six functions again for any remaining trace of the original exact-match pattern and raises an error, refusing to complete, if it finds one — turning "we're confident we got all six" into a check the migration itself performs and would fail loudly on, rather than a belief nobody independently verified.

An organization configures a billing-related detector to check in at 9:00 each relevant day. The scheduler triggering that check runs at minutes 2, 12, 22, 32, 42 and 52 — never at minute 0. Before the fix, the exact-match comparison against "09:00" could never succeed at any of those actual running times, so the detector never fired, for the entire time it was configured this way. After the fix, the same detector correctly recognizes it is within its 10-minute window during the 9:02 sweep and fires exactly once.

Six of Centriu Axis's scheduled detector functions — covering a family of billing-adjacent checks and other timed, proactive nudges — now genuinely fire at their configured hour, closing a gap where a naturally-chosen, round configuration value could structurally never have worked, verified by a built-in check confirming the fix was applied everywhere it needed to be, not just where the investigation started.

This fix addresses WHEN a detector's own internal time check succeeds relative to the scheduler invoking it — it does not change what any individual detector actually checks for or recommends once it does correctly fire, which remains each detector's own separate, unmodified logic.

A time comparison written as exact equality looks unambiguous and easy to reason about in isolation — the defect only exists in the relationship between that comparison and the specific, periodic way the surrounding scheduler actually operates. Centriu Axis's fix replaces an exact check with one scoped to match how the scheduler genuinely runs, and proves the fix was applied everywhere the same defect existed with an explicit, automated check built into the same migration, rather than trusting that six manual edits all landed correctly.