Evals for Your Harness

Build the eval suite that tells you whether harness changes help: task sets, graded rubrics, and regression discipline.

0 runs0 forks0

Chain

Inputs

Step 1

flow-evals-for-your-harness-1-design-the-task-set

Step 2

flow-evals-for-your-harness-2-implement-graders

Step 3

flow-evals-for-your-harness-3-run-honestly-variance-and-the-results-store

Step 4

flow-evals-for-your-harness-4-gate-changes-and-close-the-loop

Final output

No inputs required — hit Run to stream each step in order.

Each step streams from its model and feeds the next. BYOK — you pay your provider directly. Add a key in Settings if a step needs one.

Step 1 · Representative, tiered, and partially held out.

flow-evals-for-your-harness-1-design-the-task-set

Claude Sonnet 4.6→ {{step_1}}

Awaiting upstream steps…

Step 2 · Automate what can be checked; rubric what must be judged.

flow-evals-for-your-harness-2-implement-graders

Claude Sonnet 4.6→ {{step_2}}

Awaiting upstream steps…

Step 3 · One run is an anecdote; the distribution is the datum.

flow-evals-for-your-harness-3-run-honestly-variance-and-the-results-store

Claude Sonnet 4.6→ {{step_3}}

Awaiting upstream steps…

Step 4 · The suite earns its cost when it starts saying no.

flow-evals-for-your-harness-4-gate-changes-and-close-the-loop

Claude Sonnet 4.6→ {{step_4}}

Awaiting upstream steps…

Build the eval suite that tells you whether harness changes help: task sets, graded rubrics, and regression discipline.

1
Representative, tiered, and partially held out.
Runs flow-evals-for-your-harness-1-design-the-task-set on Claude Sonnet 4.6
2
Automate what can be checked; rubric what must be judged.
Runs flow-evals-for-your-harness-2-implement-graders on Claude Sonnet 4.6
3
One run is an anecdote; the distribution is the datum.
Runs flow-evals-for-your-harness-3-run-honestly-variance-and-the-results-store on Claude Sonnet 4.6
4
The suite earns its cost when it starts saying no.
Runs flow-evals-for-your-harness-4-gate-changes-and-close-the-loop on Claude Sonnet 4.6

•Recurring tasks you currently copy-paste between tools.
•When a single prompt isn't enough — you need a chain where output(n) feeds prompt(n+1).
•Fork and adapt if your inputs differ materially from the author's use case.

Claude Sonnet 4.6