What Probe is, how to drive it, and how to run a review from an AI client. This page is readable without signing in — “what is this” and “how do I connect” are questions people have while they are still outside.
Probe is E-Sharp's design-review engine. It reads a board as built — the ODB++ netlist, the schematic, the BOM and the datasheets behind the parts — and interrogates it against recorded intent, holding up every gap for a human to resolve. It probes the design, never the engineer.
A probe is the instrument every test engineer reaches for: the scope probe, the flying-probe tester. To probe is to investigate.
Probe cites, or it abstains. A check with no evidence behind it is reported as a gap, never as a pass. That is why a review here has three outcomes and not two: a finding, a clean result with its citation, and an abstain that names what was unread. An abstain is not a pass, and reading it as one is the single most expensive mistake a reader can make with this product.
E-Sharp AB builds electronics and the tools that keep the record of them honest. The products are signal-themed and each owns one thing: Maestro conducts, Trace remembers, and Probe reads what is really there.
Probe and Trace federate, and never merge. Raise a thing where its subject lives:
A worry about a board, a net, a pin or a device is a Probe concern — it federates up to Trace and, once confirmed, gates Green. A worry about a requirement, a test or a purchase order is Trace's from the start. Probe holds the device-knowledge half of the component library; Trace holds the sourcing master.
A design line holds several takes. A take is one ingested copy of a board — one ODB++ export, with its own netlist, its own BOM and its own review. When a design is re-exported the new upload becomes a new take; the previous one is superseded, not deleted, and it stays a real board with a real review on it.
Boards are named PROJECT CODE · REVISION · PRODUCT — for
example ETT00000022 R4.1 A12 Control — plus the take where a line has several. If you
are told about a board any other way, ask which take it was.
Everything lives in one rail on the left. Board actions are at the top, because they act on the board you have open: layers, the connection tree, the schematic, the review, the mark-up tools. Global destinations are the icon strip at the bottom — Boards, Fleet, Interconnects, Design rules, Parts, this page — with your initials at the very bottom as the sign-out.
Probe speaks MCP, so an AI client — Claude Code, or anything else that speaks it — can read boards, nets, schematics and datasheets, and drive the review with the same tools a person has. Your key carries your roles, resolved on every call: it can never do more than you can, and it stops when you do.
…
…
On a deployment that authenticates, every call needs a credential — reads included.
Mint a personal key on the Settings page and put it on the
registration as X-Probe-Key. There is no shared service key and no way to name
yourself in a parameter: who did something always comes from the credential.
401 ANONYMOUSclaude mcp remove probe --scope local.A refusal for lack of a seat reads differently: it names the seat. That is the licence authority's answer, not an authentication failure, and minting another key will not change it.
A review is a walk, not a single call. The engine scopes the board into cards, an AI or a person works them one at a time against the documents, and each card is reported back with an outcome and its evidence.
analysis_scope turns the board into a coverage DAG of cards and
returns the first of them. Re-scoping later is safe and is the intended remedy when guides
activate after a run was scoped — a card that reached a terminal state keeps its state,
its outcome and the reviewer's note.analysis_next. Each card names one entity and asks one
question, with what to gather and how to resolve it.analysis_report with the outcome: a finding, clean with the
citation, or abstain naming what could not be established. Abstaining is a correct answer
and is what keeps coverage honest.analysis_status shows nothing open, then read the report. A
human adjudicates the findings; a confirmed concern federates to Trace and gates Green.done is the only statement Probe makes about a board. Not
“passed”, not “clean”. It means every card was walked and every gap was
either resolved or named. A verdict is not quotable until the product says it is usable —
until then a report prints INTERIM on its face, and so should anything written from it.
Anything you write from a review that outlives the chat — a document, a mail, a slide — carries a source line: Probe, the board as code, revision and product, the take, and the date. A finding with no source reads as an opinion, and an opinion cannot be re-run or disagreed with.
Two things in the product are better documentation than this page, and both stay current because something depends on them: the design rulebook, which says what each rule looks for and why, and the MCP tool descriptions, which an AI client reads when it connects.
Found a defect, or want something Probe cannot do? An AI client with a key can file it directly
— bug_report and feature_request — and it lands on the same
backlog the product is built from.
A public Probe, such as the CERN showcase, counts its visits. It sets no cookie and stores nothing in your browser. Your IP address is not stored: it is used once to look up which organisation the network is registered to (a university, a laboratory, a company, or an internet provider), and only that name is kept, together with the pages, boards and views opened during the visit. A visit is identified by a code that changes every day and cannot be traced back to you. The number of visits is shown publicly; the rest is seen only by E-Sharp AB. A Probe you sign in to does not count visits at all.