Kind decides the layout
A project’s kind is fundamental identity, declared at creation and not a reader preference — there is no relabel control.evaluation used to be called inference. Every such project was converted, and inference is still accepted on input as an alias that is stored as evaluation. An experiment is itself stored as a project of kind experiment, filed under its parent.
Kind reaches the project page only. It never restructures an experiment or a run — anything type-specific inside an experiment is driven by that experiment’s own data.
The project page
The header carries the project’s name, kind, research tags and contributors. Below it sit the page tabs: Overview, then Data, Subprojects, Integrations and Sessions as they apply. Data appears once the project has recorded data, Subprojects once it has subprojects or references, and Sessions once a coding-agent session is attributed to it. Overview has three parts:1
Explorer, on the left
The project’s experiments with their runs, Live runs (running now) above Project runs (runs with no experiment), and on a
research project its papers. Filter… matches experiments, runs and papers. Lists show 10 runs at a time; Show more runs reveals 10 more, and selecting an experiment shows all of its runs.2
Center
The Metrics center, or the papers on a
research project. Selecting an experiment or run in the explorer narrows the center to it. With the canvas on, an Overview / Canvas toggle switches the project scope to a board of its experiments and runs.3
Context pane, on the right
Summary, Notes and Files for the project or the selected experiment; a selected run shows its own tabs here. The Summary is the project’s or experiment’s page, written by the recorders; an entity nobody has written up yet says “No summary yet.” See summaries. When the center has nothing to show, the Summary moves into the center instead.
The Metrics center
training, evaluation and general projects share one Metrics center. Each experiment gets a collapsible section that compares its runs on shared charts, with its saved question above them. Runs with no experiment share a Project runs section before the experiments.
- Tables and charts an agent picked (tables, metric grids, bar, scatter, heatmap, box, histogram and parallel charts) render inside their experiment’s section, before its metric charts. Project-owned tables keep their own headings above the experiment sections.
- One search above every section matches metric keys and reopens matching tables.
- The legend and the explorer’s eye controls hide and show runs across a section’s charts. Hidden runs stay listed so you can bring them back.
- Show more runs, at the bottom, reveals 50 more runs from the activity-sorted roster.
Papers
On aresearch project, papers are listed in the Overview explorer, even before the project has experiments or runs. Selecting a paper opens it in the context pane: its summary, its source and repository links, its tags and its sharing action. An old /papers/{id} link opens the owning project with that paper selected.
The Canvas view of a research project is one board of its papers. Each paper sits to the right of the earlier paper its first recorded link points to, so the board reads as a tree of how the reading went. Arrows are only links an agent or person recorded (relates_to, discovered_via, informed_by); hover one for its reason.
Recording a paper is an act by whoever did the reading, so the dashboard has no add-paper form. See Papers.