docs
How the contract model works.
candlelit is opinionated about one thing: a change is not finished when it ships, it is finished when it has been shown to work. Everything else follows from that.
the loop
From idea to verdict.
Write the contract. Before work starts, agree the metric, the window, and what comes out if it fails. Anything at 13 points or above comes to you for approval.
Build it. Your coding agent claims the item over MCP, opens a branch, and works. Progress is recorded as it goes rather than reconstructed afterwards.
Ship it. The deployment is recorded against the work item with the commit, the environment, and how that fact was established.
Wait. The measurement window runs. Nobody argues about the result while it is still being collected.
Evaluate. The metric is read from the source you named — Sentry, PostHog, Stripe, Datadog, Grafana. Met, or failed. If it failed, the removal scope you agreed is what comes out.
a contract
What one actually looks like.
metric checkout.conversion (posthog) threshold >= 3.4% window 14 days removal revert #482, drop the express-checkout flag points 8
The metric has to be something already measured. "The team feels good about it" is not a threshold, and a contract that cannot fail is not a contract.
what's here now
The current surface.
- MCP server — the tools your agent drives the workflow through
- Strategy: goals, objectives, and key results with real metric sources
- Backlog, sprints, and contract lifecycle
- Escalations, with approvals in Slack
- Client reporting: a read-only view with evidence attached
- Session capture, for what the work actually took
next
candlelit is opening gradually. If the model fits how you want to work, join the waitlist and we'll be in touch as places open.