/experiments
Most experiment tracking stops at artifacts and metrics. But the most valuable output of an experiment is the team’s interpretation: what worked, what didn’t, and what to do next. That context typically lives in someone’s head. When they leave, the next person starts from scratch.
Outcomes captures what your team tried, what happened, and what the team decided and why. Every decision logged is institutional knowledge that persists regardless of team changes.
Experiment Timeline
The main view shows all experiments for the currently selected project, sorted by date or impact.Experiment Cards
Each experiment appears as a card showing:- Name and creation date
- Source type icon — research resource, hypothesis, incident, or recommendation
- Status badge — one of the eight lifecycle states
- Target metric and observed delta (color-coded: green positive, red negative, gray pending)
- Tags for grouping and pattern detection
Experiment Lifecycle
An experiment moves through eight states:shipped, iterating, and rejected are reached from analyzing. abandoned can be reached from any non-terminal state.
Outcomes Analytics
A date-bucketed timeline sits at the top of the view, with three charts over your target metric across experiments:
The observed delta appears as a click-to-edit pill: click it to update
observed_delta inline without opening the experiment.
When you move an experiment into a terminal state (shipped, iterating, rejected), a decision-capture modal opens so the reasoning is recorded at the moment the decision is made.
Filtering and Search
Creating an Experiment
Click + New Experiment in the top-right corner. The create form adapts based on source type:- From a Resource
- Custom Hypothesis
- From an Incident
Select the Search papers source mode. Search for a paper, repo, or model by title. Remyx autocompletes from its resource index.
Experiment Detail
URL:/experiments/dashboard/<experiment_id>
A two-column layout showing the full lifecycle of a single experiment.
Origin Section
For research-sourced experiments, the Origin section shows the launch context — built automatically on first load (~2-4 seconds):
For hypothesis-sourced experiments, the Origin section shows the hypothesis text.
External Validation Data
When you’ve connected the relevant tools, the experiment detail view pulls in external validation data alongside the experiment’s own record:- Weights & Biases runs — tracked training and evaluation runs surfaced from your connected W&B account
- HuggingFace metadata — model and dataset details for linked HF resources
Analysis Card
Combined Hypothesis and Decision in a single card:- Hypothesis — the team’s prediction, always visible at top
- Decision — logged after results are in; includes text, author, and timestamp; click to edit
Implement Section
A compact bar for Claude Code integration:- Copy-paste command to run Claude Code with the Remyx MCP connection
- Link to Connectors for setup
Activity Feed
Unified chronological feed combining:- Comments with @mention support, edit/delete
- System events from the knowledge graph (experiment created, status changed, decision logged, PR opened)
Sidebar
Logging a Decision
The most important step in the ExperimentOps workflow. After reviewing results:- Scroll to the Decision section in the Analysis card
- Write what the team decided and why
- The decision is timestamped and attributed to the author
“Ship to 100%. The re-ranker specifically helps with multi-topic tickets where the old retriever returned tangentially related articles. Three retrieval experiments now, all positive. This is our best direction.”
Related
Insights
See cross-experiment patterns and recommended next steps
Projects
Leadership portfolio across all projects
Connectors
Link GitHub, Linear, Jira for bidirectional sync
Project Settings
Configure metrics, repos, and integrations per project