View as Markdown (for AI/KB) Whole knowledge base (llms.txt)
A Probe review isn’t a button you press once. It is driven card by card: Probe splits the board into a coverage map of review cards (playbook cards), and you — through your AI — work through them one at a time until nothing is left open. Probe doesn’t reason on its own; it hands out cards, gathers evidence and keeps the count.
Why the review is driven by cards
Each card is a bounded question about one part of the design, with its own goal, its own material and its own acceptance criterion. That makes the review traceable: every conclusion can be tied to a card and its evidence, instead of a loose “this looks fine”.
Start and scope an analysis
You (or your assistant) start an analysis on a board. Probe creates a run, splits the work into cards, and hands back the first ones to work on. You can scope to a focus area or re-review everything.
What a card holds
A card carries four things: goal (what is to be decided), collection (what material is needed — schematics, datasheets), questions to answer, and acceptance (what it takes for the card to count as done).
Four answers per card
Each card resolves to exactly one of four outcomes — never an assumed pass:
- Clean — checked and without remark.
- Finding — something to fix, pinned at the design with evidence.
- Abstained — couldn’t be decided; states explicitly what’s missing.
- Boundary — lies outside what Probe owns (e.g. pure manufacturing geometry).
It abstains rather than guesses: a clean report means reviewed, not assumed.
“No more cards” doesn’t mean “done”
Read the status right. The run’s percentage says how far the work has got; a node’s headline % measures evidence over the active parts — the ones that can do harm — so a card stays shallow as long as it holds any unproven active part. You can’t average your way to a green board by clearing the easy 90 percent. The list running out means the cards have been worked through, not that every active part is evidence-backed.
Revise and re-review — without losing work
An answer can be revised, and a run can be re-scoped when the material changes, without losing what’s already been worked out (see boards and takes). Decisions are also invalidated on their own when the material they rested on changes.
Next: read and export the findings.
Probe is E-Sharp’s design-review engine. Probe doesn’t reason on its own — the AI in your own tool does the reasoning and you make the technical call.