Proof of work for strategy

Yours. The engagement, the record and the Decision Assurance Pack carry your firm’s name; the method is attributed underneath it. Your client is buying your judgement, delivered to a standard they can check — not a software product with your logo on it.

No, and the difference is testable. The core is an evidence graph with deterministic belief propagation: confidence is computed from evidence tiers and the rules on each edge, so the same inputs always produce the same number and every movement traces to the evidence that caused it. Language models draft, summarise and pattern-match inside that structure — they are never asked how confident they are. That mechanism is three years of engineering.

A chat assistant speeds up drafting; it cannot testify to how the work was done. The assurance lives in the structure around the models, not in the models: go/kill criteria fixed before the evidence arrives, confidence computed rather than asked for, kills and overrides logged at the moment they happen. A template asks the team to write the record up afterwards — which is exactly the discipline that slips when the deadline is close. Here the record is a by-product of doing the work, not a second job after it.

One live engagement at a time, not a firm-wide rollout. A partner runs a real piece of work on the platform with us alongside them, and the argument is settled by whether the engagement went better. Rollouts that start with a training programme and a mandate are how software becomes shelfware in a partnership.

The analyst layer’s grind — data pulls, model rebuilds, deck cycles, storyline assembly — becomes platform work, so a partner and a small team carry engagements that used to need a pyramid. The gain lands hardest where fees are fixed or value-based; if you bill the pyramid by the hour, the conversation is about what replaces that leverage. Either way, anyone putting a number on your P&L on a public page is guessing — it is a conversation with your numbers in front of us.

By interrogating it rather than trusting a summary. Every claim opens: the evidence underneath it, the tests it survived, the routes that were killed and why. Forming a view takes minutes of targeted checking instead of the old choice between trusting the team or re-deriving the analysis. The final gate is a named human approval — your partner’s name, not a model’s.

The record is the protection, not the exposure. When a decision is challenged later, the firm without one has a deck and some memories; the firm with the pack can show what was known at the time, what was considered, what was rejected and why, and who decided — the same way an audit file has defended professional judgement for a century. The dangerous position is not a documented process; it is an undocumentable one. And nothing in it ships without a named partner choosing to ship it.

It stays in your infrastructure. The platform deploys into your own VPC, where we have no access; data is encrypted in transit and at rest, and access runs through your single sign-on. For the strictest environments it runs entirely on open-weight models you host yourself — no new subprocessor for your clients to approve, and no plaintext leaving your perimeter.

No. Hypothesis3 is a software company — we license the platform and we do not run engagements or hold client relationships. Firms deliver; we build the layer they deliver on.

The Decision Assurance Pack stays live in your client’s hands. When the world moves — new data, a revised forecast — the evidence re-propagates and the answer updates. That is a reason for them to come back to the firm that built it, rather than a study that ages on a shelf.