REMORSE Evaluate
SUBLATE / SCOPED PILOT

Prepare your solver diagnosis

Describe the problem and the evidence you have. Build a review brief locally, then choose what to share.

No file upload. Keep private code, data and credentials on your machine.

Your problem

Required fields help identify the next useful review.

Keep private details out of this brief.

Baseline available?

Deterministic scorer available?

A runnable baseline and repeatable scorer make results comparable.

Data permission

Can you use the intended instances for this review?

Choose Yes only when you have permission. Describe blockers in the brief.

Promotion rules

Optional · confirm with your reviewer

Works in this browser. Nothing is sent from this form.

Run on your machine

Download the pilot kit. Connect your own local Python adapter, baseline and scorer.

Download pilot kit

Local quickstart

Commands shown with placeholder paths.

node sublate/cli.mjs doctor --project ./project
node sublate/cli.mjs run --project ./project --out ./runs/baseline
node sublate/cli.mjs propose --project ./project --out ./runs/proposal

Put project.json in a prepared project directory. See the kit README for setup and macOS platform requirements. The runner uses the Python standard library.

External AI: off by default

Baseline runs locally. Proposal generation needs an explicit opt-in in your local config. Your Claude key stays on your machine; this page uploads no key or project.

Inspect the requested disclosure before enabling AI: provider, model, task text and permitted files. The optional Anthropic connector requires your key and a budget. No live paid API call has been verified.

You control the next step

The intake brief is a planning record. Local execution and any external AI request require separate choices on your machine. Both approvals stay false in exported JSON.