Time in status on monday.com, saved as numbers

Regular monday.com boards have no time-in-status column, and each workaround breaks somewhere. A date stamped by an automation holds one date, so when an item returns to a status the earlier stay is lost. Formula results can't trigger automations and aren't always recognised by charts. A Time Tracking column counts only once its timer stops, so nothing can act on "in Review for more than 24 hours". The activity log has every change, but as a list, and older changes drop off when the plan's retention window passes. The SLA column exists only in monday service.

Teams want one number per status that formulas, filters and dashboards can use, and an alert when something waits too long.

• Numbers, not text. Hours go into number columns the app creates, or adopts when one with the same title exists. It writes to no other column, and if someone deletes a managed column, it refreshes the board's columns and tries again.
• History from before the install. The first time the app sees an item, it reads the board's activity log and rebuilds each stay in that status column from when the item was created, so totals are right from day one. If part of the history is already past the retention window, the current stay starts from when the status was last set. After that the app keeps its own record of every change, so totals survive once the log has dropped them.
• Late and duplicate events. A change can arrive from an automation and from a board webhook, sometimes twice or late. Repeats change nothing, older changes are ignored, and each item record carries a version number, so two updates at once can't overwrite each other.
• No stored monday tokens. monday sends a short-lived token, valid for about five minutes, with each request. The app does its reads and writes inside that request and never stores a long-lived token. Every request from monday is verified, by its signed token or a webhook id the app registered, before anything is read or written.
• A scheduled SLA scan. Turning on the trigger registers a board webhook on the status column and creates a timer for each item already in the watched status. A monday code scheduler job runs every five minutes and fires each due timer once, through monday's trigger webhook with a signed request. A failed call is retried up to five times with growing delays.
• AI that never sees item names. The analysis sends one board's column titles, status names, item counts and hours per status. No item names, ids, people, updates or files, and nothing until someone clicks the button. A board's result is reused for up to six hours, each account gets up to 25 analyses a day, and every reply is checked against a fixed format.
• Working inside monday code. monday code's document database supports only simple filters, so records are keyed for direct lookups by account, board, item and column. An uninstall event deletes every record for that account at once.