The queue includes questions your team has already answered.

The first week

What day one actually looks like.

You bring one policy your team keeps re-explaining — travel gifts, vendor data sharing, social-media mentions of the company, whichever question interrupts your week most often. We do the structuring work with you: what the policy decides, what it leaves open, and which answers your team has already given in mail and memos that deserve to outlive the thread they are buried in.

Your employees get one place to ask. Your lawyers approve every reusable answer and the conditions under which it applies — the fallback positions, the audiences, the facts that would change the call. Adding a document and saving a review do not yet put its answers into use. No answer goes out that counsel has not approved.

When counsel asks for a draft, Newcomb reads the question against the policies on file. Where a policy has already been read, a plain word-matching pass narrows what the model reads; the draft quotes the policy, or says it found nothing quotable — and “nothing quotable” is never concluded from a narrowed read. Counsel may accept or edit the draft, or write her own; only the wording she approves may be delivered.

A policy is written for a category and leaves words open — “reasonably necessary,” “material change.” When a question turns on one of those, counsel reads it once: what she thinks it means, and what would change her mind. The record keeps that reading beside her answer. A model may find the open term; it never decides what it means.

Supervised means exactly what it says.

At the start of the pilot, every submitted question is available to Legal, and no automatic answer is delivered. This is not a limitation we are waiting to remove quietly; it is how the pilot is designed to run. You see the real questions, in their real words, before any answer is reused.

After matching is separately approved and enabled, a model may nominate a candidate from the closed registry of counsel's current guidance — nominate, not deliver. Automatic delivery remains a later, independently qualified and counsel-activated capability. Until counsel activates it for a tested family of questions, it does not exist for your tenant. And a question the system cannot match goes to Legal with a plain referral, not a scolding and not an invented instruction.

The weekly report counts everything, including the misses.

The founder-delivered weekly report starts from the full denominator: every question employees submitted, not just the ones that went well. It separates the questions Legal answered, the questions still waiting on facts, and the questions kept for review. If half the questions had no applicable guidance, the report says so — that is the list of answers worth writing next, which is more useful to you than a favorable percentage.

What the record contains.

When an answer is delivered, the record keeps the exact approved text, who approved it, and when. Reusable guidance also records the source versions and conditions counsel approved. If one of those sources is updated in Newcomb or counsel turns that answer off, it stops being available for reuse.

What the pilot asks of you.

The pilot is founder-assisted, which means we do the heavy work. Your side is one policy, the counsel hours to approve the first reusable answers, and a standing weekly review of the report. There is no integration project and no rollout committee — employees reach the front door by a link. If the pilot shows your team nothing worth keeping, the record of what your employees asked is still yours.