Verified agentic research and execution
Answers you can defend.
For decisions where being wrong is expensive. Agents do the discovery, the drafting, and the execution. A named expert owns the verification and the conclusion.
Built for investment, commodity, and technical teams.
Does a common factor drive long-end yield co-movement?
- Claim
- Common factor share: 20 to 80%
- Evidence
- Primary sources, locator per claim
- Verification
- Independent pass, dropped claims named
- Owner
- Named macro expert
- Decision
- One testable block registered
Every conclusion carries a kill criterion. The record shows what was refused, not only what survived.
Every claim carries a source.
A locator, not a memory. Claims without one are hypotheses or removed.
An independent pass re-checks it.
The verifier is not the agent that proposed the finding. Unsupported claims are dropped.
A named expert signs the conclusion.
Accountability is not delegated to a model or a committee. One person owns the answer.
The practice
Plausible research is cheap. Verified judgment is not.
Agents can produce a thousand pages in an afternoon. Your risk function still asks who checked them, what was killed, and who is accountable.
We run the programme with you. Frame the decision, gather the evidence, verify it claim by claim, and put a named expert behind the conclusion. Then we hand the method to your team, so the next question does not need us.
- A decision you can defend in front of the people who can stop it.
- A record your risk function can inspect, claim by claim.
- A team that can run the next question without us in the room.
The method
Five gates. Every claim, every time.
The method is the product. Each gate produces an artefact the client can inspect, and every public field note shows the same five.
Frame
Turn the decision into a falsifiable question. Name the owner. Write the kill criteria before the work starts.
Decision frame
Discover
Bounded agents gather candidates, evidence, and counter-evidence. Every claim carries a locator.
Claim table
Verify
An independent pass re-checks every retained claim against its source. Unsupported claims are dropped.
Verification record
Decide
The expert writes the conclusion, the confidence, the conditions that would change it, and the cost of being wrong.
Decision memo
Execute and transfer
The decision enters the workflow. The artefacts, prompts, and method are handed over, so the client can run the next one.
Handover pack
Field notes
What a verified decision looks like.
Macro and rates
The engine we refused to build
A proposed factor-attribution engine was narrowed to one testable claim. The rest was refused with a reason, and the record says why.
Read the recordMarket infrastructure
120x from a better question
A measured optimization programme replaced a scan with a constant-time test, and reverted a 5.2x regression instead of shipping it.
Read the recordCommodity flows
What a flow can and cannot tell you
Physical links to price, registered as predictions before the outcome. A work in progress, stated as one.
Read the recordWhere it fits
Built for decisions like these.
Investment research
Macro, rates, and equity questions that need a defensible answer, not a faster draft.
Trading infrastructure
Latency, throughput, and market-data architecture, measured and recorded rather than asserted.
Commodity trading
Physical flows, basis, and policy transmission tied to a testable price view.
Corporate strategy
Market entry, competitive position, and capital allocation under uncertainty.
Risk and compliance
Evidence trails, approval records, and data boundaries your risk function can inspect.
Data and engineering
Agent workflows, evaluation harnesses, and the plumbing that makes them repeatable.
Ways to work
Start with one decision.
Most teams start with a sprint. The ladder is there when the work becomes continuous.
See how engagements work- 01
Decision sprint
Start hereA live decision, 2 to 4 weeks. You get a decision record and a verification trail.
- 02
Operator programme
The harness, support, and a bi-weekly expert cadence for named operators.
- 03
Configured framework
Your priors, data, evaluations, and governance encoded in the workspace.
- 04
Bespoke research and execution
Specialist knowledge or tooling on a decision your team cannot staff.
- 05
Organizational development
Process, roles, and governance redesigned around the method.
Security and governance
The evidence trail is part of the product.
Written for the people who can stop the deal. Short, checkable claims, and no certification we cannot show.
Read the security posture →Approved data only
Bloomberg, FactSet, internal files, git. If it is not approved, the job cannot reach it.
Identity on every request
Every job resolves to the person who started it. Agents inherit that identity.
Audit trail to sign-off
Which model, which sources, what was retrieved, what was written, who reviewed.
No training on your data
Client material is not used to train models without written agreement.
The engineering
The systems behind the practice.
The products are evidence that we build and run real systems. They are not the first thing a client buys.
AI traffic security
mGateway
Network-layer security and policy for AI traffic. Every request resolves to a person, every model and tool call follows the allow-list.
Visit site →AI coaching
MindCoachLabs
Decision coaching grounded in research-backed frameworks. A separate product line for individual and team development.
Visit site →Agent work loop
Harness
The loop behind the practice: research, plan, build, verify, and the evidence trail that makes the work inspectable.
Visit site →Nine methods in development, from red-teaming to market discovery. See Labs.
Start here
Bring one decision.
A 2 to 4 week sprint on a live decision. You get the decision record, the verification trail, and the method your team can run next.