train your OS

You are talking to an orchestrator, not filling in a form. State what you want in plain language and it routes the request to whichever of the seven agents should handle it. You never have to know which one that is.

Ask for the outcome, not the tool. Say "write this account to the CRM" rather than naming the CRM product. The connector layer knows which tool you bound to which role.

Be specific about the account and the decision. The system does better with a request that names what you are trying to decide than with a broad one.

The pattern on the right side of that table is the same every time: name the account, name the output you want, and hand over any context that would sharpen it.

How to correct it when it gets something wrong. You do not hand-edit scores, and you do not argue with an output. You feed the source and let it recompute. The loop is five moves: Feed the source material, Extract the signals from it, Update the fields, Recompute the scores, Log what changed.

When the tone or the wording is wrong rather than the facts, that is a training problem, not a correction problem. Section 2 covers it.

Your OS runs off two things you own and control: a Company Brain and one or more Offer Profiles. Training the system means filling those in.

What the Company Brain actually is

The Company Brain is the seller model. It is your company, described well enough that an agent can act on your behalf without asking you first. Every agent reads it before it does anything.

What an Offer Profile actually is

An Offer Profile is a child of the Company Brain, one per product or service you sell. One complete Offer Profile is a hard requirement before any account can be scored.

Why a thorough answer beats a fast one

When you first launched, you answered four questions in one line each. That is the Provisional Pass, and it is deliberately easy so you can start using the system on day one. A one-line answer clears the gate. It does not make the output good. This is also why every field you filled with a one-liner is flagged as estimated, the system is telling you it is working from a guess you gave it.

How to deepen a Provisional Pass answer later

The deepen path is inside the product, available anytime, with no deadline. Highest-value things to add, roughly in order:

• Your ICP in full, including your anti-ICP, the accounts you should never chase. The most underrated field in the whole Brain.
• Your differentiation. Why you, against each named competitor.
• Your value. The quantified outcomes you move.
• Your messaging. What resonates, your objections, your never-say list.
• More Offer Profiles. One per additional thing you sell.
• Real samples of your own output. These are what teach the system to write like you rather than like a competent stranger.

Every answer you add moves your Readiness score, and the bands go Partial, then Operational, then Deployment-Ready.

The Quality Layer, and what it means for you

Every client-facing thing the system produces runs through three checks before you ever see it, always in the same order.

The Voice pass rewrites the draft to sound like your company. It changes wording and rhythm, never facts.

The Edit pass tightens the prose, cuts filler, and enforces house discipline, including length targets per output type. Being outside range is flagged, not blocked.

The Review pass is the one with teeth. It checks every factual assertion in the draft against what your data actually supports for that account. If anything does not check out, the draft is held, and it does not reach you or get written to a connected tool. A held draft is the system working, not failing. You get the exact sentence that failed and a pointer to which field would resolve it.

Reading your two account scores

Completeness is breadth, how much of the record is filled in, Sparse to Full. Intelligence is depth and winnability, L1 Cold up to L5 Wired, with hard gates: no L4 without a confirmed champion, no L5 without a confirmed economic decision-maker, a compelling event, and a known decision process.

Habits that separate teams who get value from teams who do not

• Feed it after every meaningful moment. Records that get fed stay accurate.
• Ground the data before you score anything. Run the pre-FIELD data pull first so the system can see who you already know.
• Read a FIGG score by its verb split, never by the total. The total tells you nothing you can act on.
• Do not force a deal through a stage gate. A deal that cannot honestly pass all five checks does not advance.

My Readiness score is low and it just launched

That is correct behavior. Four one-line answers typically land you in the Partial band. The score is honest rather than flattering, on purpose. Treat it as a progress bar, not a defect.

It says "Not Ready to Score" and will not run

The gate is doing its job. Four fields are required before any account scoring runs. Fill the one it lists and it runs.

The output looks thin or generic

Almost always a training gap. Check in order: your gate fields are still one-liners, your Offer Profile is missing disqualifiers, you have not written an anti-ICP, or the account has not been fed anything since the initial research.

The tone is wrong

Fill in your voice description and never-say list, then upload three or four real examples of your own writing. Highest-return fix if the facts are already right.

A draft came back held

The Review pass found an assertion your data does not support. Supply the missing fact or cut the claim and let it regenerate. Do not paste the claim back in manually.

Should I contact support or just add more training data?

Add training data for generic output, wrong tone, shallow scores, or a stuck Readiness score. Contact support for anything about your license, activation, or billing, an install or launch failure, or a connector that will not authenticate.

FiggBrain Deployed is self-serve and does not include white-glove support. Hands-on help is available at $150 per hour. Email support@revsparkai.com with your license key and what you were doing when it happened.