Skip to content

Your first run in the viewer#

The viewer is a web page that shows a run as it works and after it ends. This page walks through a finished run of the example_study, a two-variable quadratic with its minimum at (1, −2). The run used claude-haiku-4-5-20251001, took half an hour, and cost $1.12.

Open the viewer#

pip install "adda[viewer]"
python -m adda.viewer studies/example_study

Open the http://127.0.0.1:8765/session?token=… URL that the command prints. The viewer lists every run in the study's runs/ directory on the left. Pick one.

Read the header#

The header answers four questions about the selected run:

  • How long has it run, compared with its budget (Wall clock)?
  • How much has it cost (Cost)?
  • How many delegations are done or running (Delegations)?
  • What is the best counted result so far (Best row)?

The green badge beside the run name says the run closed as Gated: the reproduction gate accepted the deliverable. Other runs show running, stopped, halted, or crashed.

Find who did what in the Timeline#

The Timeline view. Three delegation cards run in two columns, and three gate marks sit at the bottom.

Each card is one delegation. Here the data generator checked the oracle (D001), an implementer sampled 50 points (D002), and another implementer delegation ran Bayesian optimization (D003) for 14 minutes. The three marks at the bottom are the acceptance reviews of the deliverable: one asked for a revision, one rejected, and the last passed.

Check what the run believed in Hypotheses#

The Hypotheses view. Four hypotheses are supported, each with a confidence bar and a link to its delegation.

Every claim the run relies on is a hypothesis with a status and a confidence that moves as evidence arrives. Each row links to the delegation that tested it. See How a run is kept honest for the rules a hypothesis must meet before the run marks it supported.

Check the evidence in Data#

The Data view. A scatter plot of every evaluation and a line for the best value so far, then a table of stored rows.

The chart plots every evaluation the run paid for. The step line is the best value so far. In this run the best value dropped at evaluation 97, from below 1 to 0.0006. The table below lists the stored rows, and each row names the delegation that produced it.

Read the answer in Deliverable#

The Deliverable view. The problem statement, then a table of hypotheses with their evidence.

The deliverable is pipeline.ipynb, rendered in the page. The Re-execute button runs the notebook again.

Go further#

The Logs and Setup views, and the controls that start, stop, and steer a run, are described in Watch and steer a run.