Agents that have been stuck for more than 5 minutes (or 300 simulated seconds) are listed below. You can hand them off to a human.
The saved run monitor begins with seven local fixtures. Two are completed, two running, two stuck and one failed. At forty pixels per run the first three status counts each measure eighty while failed measures forty. Filtering changes the visible rows but the statistics count the entire runs array. These are browser records rather than observed remote processes, and a displayed running status cannot prove a worker is actually generating. The page does not connect to a supervisor or production agent backend. The seven fixtures contain twenty eight thousand seven hundred fifty combined input and output tokens and stored costs totaling zero point one one eight dollars. The add-run function instead assigns zero point zero zero two dollars per thousand tokens. Applying that rate to the existing token total would yield zero point zero five seven five dollars, leaving a difference of zero point zero six zero five from stored costs. At two thousand pixels per dollar the stored, recalculated and difference bars measure two hundred thirty six, one hundred fifteen and one hundred twenty one. The sample cost fields are not invoices or consistent billing evidence. The two stuck fixtures have last activity four hundred and five hundred seconds before initialization. Both exceed the three hundred second handoff threshold by one hundred and two hundred seconds. At half a pixel per second their age bars measure two hundred and two hundred fifty, and threshold one hundred fifty. Handoff merely changes a local status to completed and shows an alert; it does not contact a person or prove resolution. Chart construction runs before handoff and share rendering, so missing Chart leaves those later sections empty offline. Native filtering and row deletion still update earlier lists and statistics before the same chart error.