A second opinion is only worth as much as the seat it came from.

On telling a second opinion you can check from one you can only hope is right.

Any assistant will give you another take on request. Few give you the one thing that makes a take worth acting on: a way to check who is answering.

A second opinion is easy to offer. You can have one from any assistant in the time it takes to ask. What is hard is making one checkable — giving yourself a way, after the fact, to see where it came from and decide whether to trust it. Last week the point was tracing a confident instruction back to whoever actually gave it; a second opinion raises the same question one level up — not who ordered this, but who is answering, and are they the right one to ask.

The seat, not the score

Here is the uncomfortable part. A confident wrong opinion reads exactly like a confident right one. When an assistant answers outside what it actually knows, nothing in the phrasing marks the difference — the sentences are just as smooth, the reasoning just as tidy. Fluency is not evidence that the judgment underneath was sound. So a second opinion, on its own, adds a voice without adding a way to weigh it.

What makes an opinion weighable is knowing the seat it speaks from. In CRAFT, the personas that review your work are not one voice wearing hats; each answers from a named role with a written remit — what it is responsible for, and just as importantly what it is not. A strategy role answers strategy. An editorial role answers coverage. A stand-in for your customer reacts as that customer — and is not entitled to make the call that belongs to strategy. When an opinion arrives, the seat travels with it. That is the difference between a number that says how sure the AI is and a record that says who was talking.

What a receipt actually is

A receipt is not the score. It is the answer to “who said this, and were they the one to ask?” The practical shape of it: route a question to several seats at once, cold and independent, and have each return its take with the seat it answers from attached. Now the second opinions are not a chorus — they are a panel, and you can see the standing of each voice on it.

Illustrative — a constructed example of seats you might define for your own work. The first two opinions come from roles entitled to weigh the price. The third is confident, plausible, and out of its lane — a brand voice ruling on a call that belongs to sales and finance. Without the receipt, all three read alike; with it, you can see which one to set aside.

This is why the receipt does more work than the confidence number. A high number on that third line would not have made it safe — a score is only as good as the seat that produced it. The receipt is what lets you catch the plausible answer that came from the wrong place, which is exactly the answer a number alone will wave through.

There is a second layer for the moments that matter most. CRAFT can have one role’s work checked by another in a fresh, independent pass — a reviewer handed the work and the standards to judge it against, but never the first author’s reasoning. It starts cold, so it cannot quietly nod along with a conclusion it was told to reach; it has to arrive at its own. A second opinion from a voice that already agreed with itself is not a second opinion, and that isolation is what keeps this one honest.

And defining a seat is not a project. A seat is a few lines of plain English — a name, what it answers, and what it does not — written the way you would brief a new colleague, not configured like software. You might keep a “compliance” voice, a “brand” voice, a “numbers” voice, each answering only what it is entitled to on your own work:

# Seat: Finance
ANSWERS   Margin, pricing floors, whether a deal covers its cost.
VOICE     Plain, cautious with numbers, says when data is thin.
NOT MINE  Brand, positioning, or whether to take the deal — flag, don’t rule.

A seat is a short plain-English brief, not code — this one is a made-up example; a real one reads the same way. The “not mine” line is what keeps a confident voice in its lane.

Where it earns its keep

This matters most exactly where it is hardest to notice — deep in a long session, when an assistant’s takes quietly drift toward the plausible and away from the grounded, and your own attention is the least reliable check. A receipt that names the seat is the part that still works when you are moving fast.