contract-driven product development
Product development that has to prove it worked.
candlelit runs your backlog, your sprints and your coding agents around one rule: every change carries a contract. A success metric, a time window, and a removal scope agreed before the work starts.
When the window closes, the metric is checked. If the change did what it promised, it stays. If it did not, it comes out. Nothing accumulates because someone forgot why it was there.
the problem
Agents made it cheap to add code. Nothing made it cheap to remove it.
Output went up and legibility went down. More gets shipped, less is understood, and the reason any particular thing exists is increasingly nobody's memory. The usual answer is more review. The better answer is deciding what "worked" means before the work starts.
the mechanism
Every change earns its place.
A success metric. Something already measured — an error rate, a conversion, a latency figure. Not a feeling about whether it went well.
A time window. How long the change gets to prove itself before anyone looks. Agreed up front, so the verdict is not negotiated afterwards.
A removal scope. Exactly what comes out if it fails, written while everyone still remembers what was added.
An evaluation. When the window closes the metric is read and the contract is met or failed. Irreversible calls come to you first.
how it works
Your agent drives it. You approve what matters.
Claude Code works through an MCP server, claiming items, opening branches, recording deployments. You keep the decisions that are actually decisions: signing a contract, approving anything irreversible, judging whether a result counts.
Approvals arrive in Slack. Everything else stays out of your way.
for your clients
A page they can read in twenty seconds.
Invite a client and they get their own view: what is verified done, what the numbers moved, what shipped with the evidence attached, and what is waiting on a decision from them. Read-only, scoped to their engagement, and honest about how each fact was established.
waitlist
what it costs
Pricing isn't set yet.
candlelit is opening gradually, and we'd rather work the price out with the first teams using it than guess at it now. Everyone on the waitlist hears it before anything is charged, and nobody is billed without asking to be.
You bring your own Anthropic plan — candlelit provides the methodology and the infrastructure, your Claude Code subscription provides the model usage.