Could someone who was not there reconstruct how a piece of AI-assisted client work was done, from our record alone?

The reconstruction test asks nine questions about one recent piece of AI-assisted client work: what the system saw, what it produced, what the person decided, and whether the record shows each without asking anyone. Each answer is scored, the result explains what the score means for supervision, records, and succession, and the whole thing prints as a one-page worksheet. It runs in your browser and stores nothing.

The reconstruction test

Key facts

When a system does part of advisory work, the firm's memory depends on three things captured at the time: what the system saw, what it produced, and what the person decided.
Rule 204-2(a)(7) requires an SEC-registered adviser to keep copies of written communications sent relating to advice; the sent version is the record, whoever drafted it.
Rule 206(4)-7 requires written compliance policies reasonably designed to prevent violations; a supervision policy that cannot show a person approved AI-drafted client work is hard to describe as reasonably designed.
A firm's own record, in its own systems, is the only version of "asking the person who was there" that works for machine-assisted work.

Pick one recent piece of AI-assisted client work

A meeting summary, a client email, a plan section, a review note. Something a system drafted and a person approved within the last month. Hold that one item in mind and answer from the record alone, not from memory. If you would have to ask the advisor, the answer is "no."

The item (optional, for the printed worksheet)

Part one: what the system saw

1. Does the record show which client information the system was given to produce this item?

Yes, the sources are named and dated Partly, the general source is known but not the version or date No, or only by asking the advisor

2. Does the record show which system, and which version or date of it, produced the draft?

Yes The system is known, the version is not No

3. If the information the system used was stale or wrong, would the record show that?

Yes, inputs are dated and checkable Only if someone noticed at the time No

Part two: what the system produced

4. Does the draft as the system produced it, before anyone edited it, still exist in the firm's own record?

Yes, in a system the firm controls Only inside the vendor's tool No

5. Can a reader tell which sentences were machine-drafted and which were written by the advisor?

Yes, provenance is marked or the two versions are kept Roughly, by comparing draft and final by hand No

6. Is the version that actually went to the client stored in the firm's own system, not only in the tool?

Yes It is in the email system but not filed to the client record No, or unsure

Part three: what the person decided

7. Does the record show that a named person reviewed the draft before it was sent or filed?

Yes, with who and when Implied by the send, not recorded as an event No

8. Does the record show what the person changed, or that they chose to change nothing?

Yes, a note or a diff exists Changes are visible only by comparing versions No

9. Is there a short note, in the advisor's own words, of what mattered and why, beside the AI output?

Yes, routinely Sometimes No, the AI summary replaced it

What the three parts are testing

Part one tests whether the output can be traced to its inputs; without that, emphasis and omissions can be observed but never explained. Part two tests whether the machine's contribution and the person's are still distinguishable, which is where professional judgment lives. Part three tests whether the human decision exists as a recorded event rather than as silence. A firm that scores well on all three can hand the work forward. A firm that scores well on none has a record it cannot retrace, however polished the finished artifact looks.

Questions

Why one piece of work rather than a policy review?

Because a policy states what the firm intends and a record shows what happened. Ten minutes with one real file teaches more about the firm's actual memory than an afternoon with the manual.

What counts as a good score?

A record that lets a stranger answer all three core questions, what the system saw, what it produced, and what the person decided, without asking anyone. The scoring bands on the page explain what each range means and what to change first.

Does this page keep my answers?

No. The answers are scored in your browser and shown on the page. Nothing is transmitted, saved, or measured. Print the result if you want to keep it.

Is this a Rule 204-2 assessment?

No. Rule 204-2 lists categories of required records and the firm decides which items fall into them. The test measures whether the record supports reconstruction, which matters for supervision and succession whether or not a given item is a required record.

Primary sources

General information from ValaisOS LLC, not legal, compliance, tax, or investment advice. Confirm requirements for your firm with counsel. See Terms of Use.