Endpoint sentinel

The same prompt, at temperature=0 with a fixed seed, sent to each endpoint every day, with the provider and quantization pinned and the served provider read back. Each panel is one endpoint: the per-day median completion length with its interquartile band, the share of a day’s probes that repeated an earlier probe byte for byte, and where a single step in the series would explain the most variance.

Loading data…

Endpoints compared

All days in the selected range, per endpoint. Compare within one model only: different models produce different lengths by design. Identity rate is the share of a day's probes that repeat an earlier probe of that day byte for byte; 0 means every completion differed.

How to read it

• A step in the median is the signature of the serving configuration changing. The screen reports where a single step would explain the most variance; that marks where to look for an identifiable event and is not evidence of one.
• Identity rate at temperature 0 is how far an endpoint is from deterministic. Two endpoints serving the same weights can differ from “never repeats” to “repeats half the time”.
• Errors are data. A day of timeouts is a finding about the endpoint.
• The record before 2026-09-07 is oversampled: a scheduling defect let up to four rounds overlap on 09-05 and 09-06. Per-probe values stand; day-level counts do not.

Method, probe text, and the raw record: the repository will be linked here when it is public.