{"id":"stability.run_length.v1","title":"Stable for N days (behaviour clock)","unit":"days","chain_set":"eip155:8453 (Base)","window":"the agent's whole counted history, one value per complete UTC day","definition":"For each agent at L7 or above, the number of complete UTC days since the last break in its own payment behaviour. A break is a day on which a two-sided CUSUM (reference 0.5, decision 4.74) on any of four daily features of its counted payments (payments, USD, distinct payees, share of payees never paid before; counts and USD on a log1p scale) crosses the decision threshold against the median and MAD of the agent's own previous days in the current regime (at most 60), with the MAD floored at 1.0 (0.25 for the share). A break restarts the baseline, so the new behaviour becomes the norm it is next judged against. The clock starts 14 days after the agent's first counted payment.","method":"afg-stability, daily at 02:20 UTC, queries over money_events only. The decision threshold gives a one-sided average run length of about 730 days under Gaussian noise (Siegmund's approximation), about one false break per agent-year two-sided; payment counts are not Gaussian, so the measured break rate is published beside the designed one (false_alarm in /api/stability). A single large payment moves the statistic for one day and washes out: it is a point anomaly, not a regime.","excludes":["Agents below L7, and agents with fewer than 15 complete days since their first counted payment (reported as too new).","Today's partial day."],"does_not_assert":["That a break is a harm: it is a change in the agent's own behaviour, which a new customer, a price change or a drain can all cause.","That the designed false-alarm rate holds: the measured rate is the honest one and is published."],"source":"AFG-Pulse stability clock","since":"2026-10-13","revisions":[],"body_sha256":"e4f7b57a4bf32b8b01fde94b0964d6f6d9cd0a9ef175b966f59561e665678699","url":"https://agenticfinancegraph.com/def/stability.run_length.v1","ladder":"https://agenticfinancegraph.com/ai-agent-liveness-methodology-l0-l9-ladder","license":"https://creativecommons.org/licenses/by/4.0/","attribution":"Agentic Finance Graph — https://agenticfinancegraph.com"}