Pre-commitment review
The review runs before the room commits
The AGI²⁺ pre-commitment review is a structured pass over one strategic decision, run before the commitment is made. It does not produce the answer. It produces four things: a frame check, a crux list, a decision structure, and signposts you keep.
A pre-commitment review for high-stakes strategic decisions. It helps leadership see whether it is solving the right problem, what it does not yet know, and where its intended decision would be wrong. It is the first operational application of the AGI²⁺ architecture, the point where it meets one real decision.
Four outputs, and the question each one answers
01 — Frame check. Are you about to answer the right question? It names the actors, the objective the decision actually serves, and what has been left outside. Most organisations discover a frame error after paying for the analysis.
02 — Crux list. Which two or three uncertainties decide this? Open questions get ranked by whether the answer could flip the choice, which also identifies workstreams that cannot change the outcome.
03 — Decision structure. Is this a one-way door, and what would turn it into a two-way door? Stage the commitment, buy the information first, or write the exit trigger before you sign.
04 — Signposts. For each crux, which observable indicator would show the assumption degrading, and at what threshold do you act? Signposts read forward, warning you while there is still time to act.
Who is it for?
Two decision classes. The first is an international market entry or a partnership structure, where the counterparty sets the criteria and a timetable that does not move. The second is a stage gate — continue, change or stop — where sunk capital tempts a room into funding for another year what it has already funded for two.
It is not an approval workflow or an audit tool, and it earns nothing on a decision already taken or cheap to reverse.
How an engagement runs
-
Scope.
One decision, not a strategy refresh or a portfolio review.
-
What you bring.
Whatever already exists: a memo, a board paper, a deck, a partner proposal. There is no blank canvas, which killed the previous generation of this category.
-
Strawman back.
You receive a completed frame with the cruxes drafted and wrong in places, because correcting a draft is more honest than filling in a form.
-
One question at a time.
Each arrives with a short line on why that question and not another. No matrices on screen.
-
Readiness call.
The point at which another week of analysis can no longer change what you do, said out loud and recorded.
-
The package.
Frame check, crux list, decision structure and signposts, with what was known and what was assumed on the day.
-
Signposts handed over.
With thresholds attached, they keep the decision alive after the room has emptied.
Three to six weeks, fixed fee, agreed before the work starts.
The moment it pays for itself
The value tends to arrive early, as a change of frame. In the pilot, a Nordic company preparing a market entry arrived with a well-built case for one route to one customer, and the review moved the problem from the export chain to the partnership structure:
“We were about to pitch the wrong value proposition to the wrong customer.”
That sentence has a cost everyone in the room can calculate, and it landed before any new external evidence entered the process. The correction came from restructuring what the team already knew. One uncontrolled, self-assessed case, judged useful by the party whose own framing was corrected.
Measured against the obvious alternative
Could you not just do this with a chatbot? The comparison is written into every engagement: the identical brief goes in parallel to a capable person with one strong model and source-and-counterargument discipline, and the package reports where the review failed to beat that baseline. It has not yet been run on a completed engagement, which is the honest state of it.
Where this actually stands
The method has been demonstrated. The system has not. What has run so far is people and a language model following a discipline: no engines, no gates that could have failed, no ledger. The software is built engagement by engagement under one rule: anything the facilitator does twice gets automated before the third engagement. Every earlier attempt in this category added facilitators to cover what the software could not do, and all are consultancies today.
FAQ
What do you need from us?
Whatever already exists, plus your open workstreams and the deadline. Half-finished is fine, because the review reads the assumptions underneath a document rather than its prose.
Who is in the room?
The people who will make the call, plus anyone holding information a model cannot have: the counterparty relationship, the internal constraint, the thing nobody wrote down. Three to five active participants works well.
What happens to the record afterwards?
You keep it, and it records what was known, what was assumed and what was contradicted as of the decision date. Nobody is scored, and a signpost tripping in eight months leads back to its assumption. In the v26 architecture that record is the starting state for everything that follows the decision, which the Runtime page sets out; today it is handed over as a document.
Can it tell us to stop?
Yes, and often that is the output, because a gate that has never failed is not a gate. Where the decisive uncertainty cannot be resolved before the deadline, the package recommends staging or stopping.
Can we see it run first?
There is a demonstration session, and it is not self-serve. Tell us the decision you are facing and we agree a case, because a generic example shows only the format.