Methodology

How we think.

Six lenses we put on every engagement. They are not slogans. They are the questions we keep asking until the work is honest.

Reverse Math

Start with the failure you fear most.

In plain terms. Most projects start with a goal. We start with the worst-case outcome and work backward to the assumptions that get you there.

Every workflow has a quiet set of assumptions holding it up. Most teams cannot see them because they are inside the system. Reverse Math drags those assumptions into the open by asking one question: what would have to be true for this to fail in a way that matters?

The answers are usually small. A retry that does not retry. A timeout that fires too late. A prompt that asks the model to confirm rather than verify. Once the assumption is visible, the fix is usually obvious. The work is making it visible.

How we use it

Day-one workshops on a new client engagement
Pre-mortems before any AI workflow goes live
Tradeoff maps inside strategy sessions
The first hour of every Reverse-Math Audit

See also: the Reverse-Math Audit service.

Sideways Thinking

The angle nobody asked.

In plain terms. If the obvious answer worked, you would not need us. We find the orthogonal angle.

Most consulting tells you what you already suspected, dressed in slides. Sideways Thinking is the opposite move. We take a problem and turn it ninety degrees. We ask which industry already solved this under a different name. We ask which constraint is doing the most quiet damage. We ask which assumption everyone treats as fixed but is actually a knob.

The output is rarely the answer you walked in expecting. That is the point.

How we use it

Cross-domain problem solving sessions
Founder strategy reviews when growth has plateaued
Sideways Sessions: ninety minutes, flat fee, find a non-obvious framing or you do not pay

Intent. Integrity. Intuition.

In plain terms. Three gates every recommendation passes through before we deliver it.

Intent is the customer's actual goal, not the literal request. Most AI failures happen because the model honored the words and ignored the goal.

Integrity is the structural check. Does the recommendation rest on assumptions that can be named and defended? Are the claims in our deliverable atomic enough to be tested?

Intuition is the human override. After Intent and Integrity have run, a person who has seen the pattern before gets the last word. Pattern recognition is the part of the work nobody automates well.

How we use it

Final review on every deliverable before it leaves the building
The structure inside every strategy session
The internal rubric our consultants use when they disagree with each other

Anti-Slop Check

Sounds right is not the same as is right.

In plain terms. AI output that reads confidently and is wrong is the most expensive failure mode in business today. We catch it before it ships.

Slop is the polite name for AI output that is fluent, plausible, and incorrect. It costs more than obvious errors because it is harder to notice. The Anti-Slop Check is a small set of structural tests we run on any AI deliverable before it goes out: does the claim cite a real source, does the recommendation survive its own logic, does the response actually answer the question that was asked.

You can try a single-pass version of this on your own prompts using the Don't Kick Test.

How we use it

Every artifact we deliver passes through this check
Anti-Slop Audit-as-a-Service: send us ten of your live workflows, get a remediation report
The public Don't Kick Test as a free first taste

Triangulation

One model is an opinion. Three is a signal.

In plain terms. Critical recommendations get run through more than one AI engine. When they disagree, that is where we look hardest.

Single-model output is a confident guess. Triangulation runs the same question through two or three engines from different families and compares where they converge and where they diverge. Convergence is signal. Divergence is the place a human has to think.

Most consultancies do this by accident, with one model and one person. We do it on purpose, with infrastructure that makes the comparison visible.

How we use it

Any client-facing recommendation that carries financial or reputational risk
Our internal weekly Triangulation Snapshot newsletter
The verification step inside Anti-Slop audits

Fingerprint Portability

Your doctrine moves with you.

In plain terms. When you switch AI vendors next year, the structure you paid us to build should come with you. Not theirs. Yours.

Most AI rollouts quietly bind a company to whichever vendor was hot when the project started. The prompts are tuned to one model. The integrations live in one platform. The knowledge is locked behind one set of credentials. We design the opposite way.

The doctrine we help you write should be a portable artifact: prompts, evaluation rubrics, guardrails, and escalation rules captured in a form you own and can move across at least three major model families. The work belongs to the customer. The substrate is allowed to change.

How we use it

Half-day Fingerprint Portability workshops for enterprise teams
Lock-in risk reviews for legal and procurement
The structure inside every Platform / API Access engagement

Want this applied to your work?

Tell us what is stuck. We will tell you which of the six lenses fits and what the first hour looks like.

Start a conversation Try the Don't Kick Test