Groundwork

Intake

Four ways to score an idea

Scope block · scoped it in a Claude chat? paste the handoff block into Scope notes and the app sets sliders, idea, and runs itself
Auto-score · type freehand notes, hit the button, Claude sets the sliders (API key, or keyless in a Claude artifact)
Presets · one-tap starting shapes for recurring builds, then nudge
/groundwork in Claude Code · skips this panel entirely: the verdict logs straight to the dispatch log below, and the session hook fills in real cost when you exit

Drag any slider any time; the verdict always follows the sliders.

Idea

Scope notes (optional)

already scoped this in a Claude chat? paste this prompt there, then paste the scope block it returns here · the sliders set themselves

Presets

Reasoning depthModerate

Template fill and reformatting on the left. Novel multi-step problem solving on the right.

AmbiguityModerate

Fully specified spec on the left. Open-ended judgment calls on the right.

Context volumeModerate

A prompt and a paragraph on the left. Whole codebases or document sets on the right.

AutonomyModerate

One shot, one answer on the left. Long agentic loops with tool use on the right.

Cost of failureModerate

A retry is free on the left. Mistakes burn client trust or real money on the right.

Output volumeModerate

A short reply on the left. Long documents or large code files on the right.

Runs per month

Verdict

awaiting intake · run dispatch to score

HaikuSonnetOpusFable

Sonnet Claude Sonnet 5 score 00

Why

Escalate if

Est. cost / run

Monthly

Cheaper play

Auto-score · Anthropic API key

API key (optional)

Remember key in this browser. It is stored in plain local storage on this device only. Skip this on any shared computer.

Auto-score sends your idea and scope notes to the Anthropic API, which reads them against the rubric and sets the sliders for you. You review and adjust before running dispatch: the model proposes, you decide. Running inside a Claude artifact, no key is needed. Deployed or forked, add your own from platform.claude.com. Keys are used directly from your browser to Anthropic and never touch any other server.

Pricing table · USD per million tokens

Model	Input	Output

Rates are editable and saved with your log so this tool never quietly lies to you. Verify against claude.com/pricing. Last verified:

Plan & usage

Your plan Your usage ceiling ($ equivalent) Rolling window (days)

Anthropic does not publish an exact per-plan token quota, so picking a plan above seeds a rough, tier-scaled starting ceiling ($ equivalent, 30-day window) — higher tiers get meaningfully higher seeds, since that is the entire point of paying more. It is a starting guess, not a claim: the moment you edit either field by hand, it is yours and switching tiers to look around will not overwrite it again. If you have used the plan for a while and the seed reads low for what you actually get out of it, raise it. The usage panel tracks real cost across all your projects in the window against whatever number you land on.

Calibration

fill in "actually used" on logged entries and your drift readout appears here

Plan usage

Log

Date	Idea	Score	Recommended	Est. cost	Actually used

no dispatches logged yet · your first scored idea lands here