Pipeline health

Cross-corridor. This is the one screen that is not scoped to a corridor — a stuck job on NH44 has to be visible while you are looking at NH16. It is composed from six calls (GET /health, GET /ready, GET /whoami, GET /capture/sessions, GET /schedule?shape=matrix, GET /admin/config) plus one progress probe per launched session. There is no one-shot GET /jobs, so figures may be seconds apart and each panel can fail on its own — a panel that failed says so; it is never drawn as a quiet pipeline.

Roll-up

What happened to my upload?

one box · filename (exact or the start of one), session id, or vehicle · GET /capture/sessions?q=

Sessions in the pipeline

stall threshold passed to the probe as ?stall_after_s=; the server reports the basis it used.

Overdue across corridors

every corridor × every stream · from GET /schedule?shape=matrix

Deployment diagnostics

what this environment actually is · GET /admin/config · admin-guarded

Cost

what it cost — and why this board will not tell you

Measured spend is not in this system

Nothing in RoadIntel records what was billed. RunTask (capture.py:684-700) carries no cost-allocation tag for corridor, session or stage, and no Cost Explorer or CUR feed is wired in. There is therefore no figure here for spend month-to-date, spend by corridor, or spend by stage — not a zero, and not a dash standing in for one.

The four figures an ops cost panel is expected to carry, and the state of each in this build.
Figure	State	Why
Spend, month to date	⊘ not measured	no billing source is connected to this deployment
Spend by corridor	⊘ not measured	processor tasks carry no cost-allocation tag, so spend cannot be attributed even if a feed existed
Spend by stage	⊘ not measured	the stage is not on the task either
Next survey — forward estimate	▲ estimate only, and not served	`estimate_survey_cost` / `frames_for_survey` (`roadintel/pipeline/guardrails.py`) compute a forward estimate from a frame count, not measured spend. They have no route and no caller anywhere in `app/`, so this board has nothing to call.

▲ Why no number is shown even for the estimate. The same estimator answers very differently depending on one keyword argument — seconds_per_frame at the measured 0.67 s/frame on CPU versus the batched-GPU default is more than an order of magnitude apart for the same drive. A single currency figure printed here, with the assumption hidden, would not be reporting cost; it would be reporting a guess with a currency symbol in front of it. When the estimator gets a route, this panel gets the dial and the assumption beside the number.

Reading this board. Counts are sessions, never files or frames. Timestamps are UTC, exactly as the API stores them — nothing here is converted to a local clock it cannot prove. Stalled is the server's derivation from S3 artifact times, and every stalled row carries the basis it used; the browser never decides it. Not probed is a state of this page, not a state of the session. No currency figure appears anywhere on this page, and that is deliberate — see Cost.