Deep 1:1 strategy coverage for every founder in the cohort.
Each founder gets 3 evidence-graded venture routes. Every score is computed in code from verified sources — not LLM opinion. No extra mentor hours, no extra headcount.
Vision → Wedge → Bridge.
The moonshot
A venture-scale endgame, anchored to the assets the founder actually owns.
The entry
One beachhead, bottom-up: who buys, why, ACV, frequency — the real factors.
The route
A de-risking sequence from the best venture playbooks, verified on failure stats.
Underneath all three sits the claim │ reality firewall. The model only proposes — routes, and the structured factors behind them. The verdict comes from non-model reality particles: reference classes of funded rounds, demand signals, and point-in-time evidence that must verify against a real source or be dropped. Claim and reality are combined only in code. A displayed quote is a checked substring of a supplied document; a market size is summed from extracted factors. No total is ever taken from the model's word.
What we can honestly claim so far.
A retrospective venture backtest on a frozen, survivorship-safe cohort of binary-outcome cases. Scores were frozen before outcomes were revealed.
The method is published, not pitched.
From Checker to Forecaster: Code-Owned Evaluation of Model-Generated Strategic Routes Under Delayed Ground Truth
Many evaluations of model outputs rely either on contracts checkable at evaluation time or on feedback that arrives within the operating loop.
arxiv.org/abs/2607.10972 →Win by Silence: Deletion Non-Monotonicity, Autonomous Exploitation, and Typed-State Gating in LLM Plan Evaluation
Plan evaluators can reward a strategic plan for becoming less explicit.
arxiv.org/abs/2607.12986 →
Win by Silence.
The 15-minute version of the argument: how a scorer can reward a plan for omitting work, how we measured it on a frozen cohort, and what the gate does about it.
Two places it fits.
Screening
Applicants run their own intake before you ever meet them, so each one arrives at the interview with three costed routes instead of a single rehearsed narrative. You read what the founder owns, what each route would cost to de-risk, and which assumption is load-bearing — and you spend the interview on the decision rather than on extracting the basics.
Coaching
Portfolio founders re-run their routes at milestones. Because every score is recomputed in code from current evidence, a route that stops being defensible says so — and the change is legible to the founder and to you at the same time. Partner time goes to the founders whose route actually moved, rather than to a uniform check-in cadence.
Worth a 15 minutes?
We'll walk one of your founders' routes through the system live, and you can judge the output rather than the pitch.