The Current Problem

The submission arrives like they all do: a polished deck with crisp satellite renders, confident projections, and a narrative that holds together. The operator's track record looks solid. The technology seems proven. The underwriter starts forming a judgment almost immediately.

This is the inside view doing what it does best: anchoring attention on the vivid particulars of this specific case. The outside view (the statistical base rate for missions like this, the reference class of similar operators, the historical frequency of failures in this orbital regime) requires effort to retrieve. By the time an underwriter pulls these numbers, the narrative has already done its work.

This is the default cognitive sequence in high-stakes underwriting, and it explains why even experienced professionals systematically misprice unfamiliar risks. The governing question is simpler than the solution: how do you price a risk you've never seen before without letting the story write the forecast?

The Governing Insight (The Principle)

Every insurance underwriting decision is, in truth, a forecast. Underwriting and geopolitical forecasting are fundamentally the same challenge: making probabilistic judgments under deep uncertainty. We are asked to predict outcomes we have never encountered before, with real financial stakes.

Philip Tetlock and Dan Gardner's Superforecasting gives the clearest map yet for how to do this well. Over four years, thousands of ordinary volunteers competed in a massive forecasting tournament against professional intelligence analysts with access to classified information. A small group of "superforecasters" won decisively. The difference was not superior intelligence, credentials, or secret data. It was superior process: better training in debiasing techniques, better teaming structures that aggregated diverse perspectives, and better tracking through rigorous scorekeeping.

The single most transferable principle from that process is the disciplined ordering of thought. Superforecasters begin with the outside view (the base rate, the reference class, what usually happens in situations like this) before turning to the inside view of what makes this particular case different. Most of us do the reverse. We answer the easy, narrative-friendly question ("Does this operator seem competent?") instead of the actual question being asked ("What is the historical failure rate for operators like this?").

In space insurance, this reversal is especially costly. Because every mission feels genuinely novel, the temptation to treat it as unique is almost irresistible. But uniqueness does not exempt a risk from belonging to a reference class. The antidote is not more data; it is a better ordering of the questions we ask, and a system that makes the right sequence the easiest one to follow.

What follows is a practical translation: how to build that outside view first discipline into underwriting software, what other forecasting tools transfer cleanly to risk judgment, and how to measure whether the shift actually improves calibration over time.

Hedgehogs see everything through one dominant framework. Foxes synthesize many perspectives. In forecasting, the fox wins

The Translation (Pattern to Proof)

The challenge was not building better models. It was changing the default path of human attention. Most underwriting errors are not failures of analysis; they are failures of sequence. The mind commits to a position based on what arrives first, and everything that arrives later gets interpreted through that initial commitment. Under time pressure, narrative becomes the default interface.

Translating the outside view first principle into daily practice required accepting a simple truth: you cannot train people out of cognitive bias, but you can design around it. So I designed the Evidence Tab as a structural intervention in that attention sequence. When a submission arrives, the first thing the underwriter encounters is not the operator's story. It is the statistical frame that story sits inside: the reference class performance, the historical frequency of similar outcomes, the base rate that should anchor any judgment about this particular risk. The system holds the inside view back until that outside view has been engaged with properly.

Once the base rate is visible and understood, the interface invites the inside view, but as a structured adjustment process. The underwriter notes what genuinely distinguishes this case, estimates how much that difference should move the probability, and records the reasoning. The adjustment is not a black box number pulled from intuition; the entire chain from statistical anchor to final judgment becomes a legible, traceable series of steps.

In practice, this small change in information architecture has a disproportionate effect. The work feels different. Underwriters are not fighting their instinct to trust the story; they are beginning from statistical humility and building toward a judgment they can defend, step by step. The Evidence Tab does not make anyone a superforecaster overnight, but it makes the superforecaster's discipline the path of least resistance in the daily workflow.

The workflow enforces sequence: statistical anchors appear before case-specific narratives, making the outside view the default starting point for judgment.

The Rest of the Toolkit (Pattern Expansion)

Fermi Decomposition
When a risk feels too complex to price in one mental move, the answer is to Fermi-ise it: break the question into smaller questions that can each be anchored to a reference class. Big, intractable risks are decomposed into component probabilities (launch reliability, on-orbit survivability, operator execution, orbital environment) before being reassembled into a coherent forecast. This shifts the work from narrative interpretation to structured estimation: fewer sweeping intuitions, more testable assumptions.

Granular Probabilities
Judgment is expressed in finely grained increments rather than coarse buckets. This shift alone improves calibration; forcing yourself to distinguish between 63% and 67% makes you interrogate your own confidence in ways that choosing between low and medium risk never does.

Update Often, Update Small
The workflow is designed for frequent, incremental revision. As fresh evidence arrives, the underwriter adjusts the probability in small, reasoned steps. The interface records each change with its justification, creating a visible trail of how judgment evolves. Over time, this creates a continuous learning loop rather than a static number frozen at the initial submission.

Dragonfly Vision and Aggregation
Diverse analytical lenses are deliberately brought into view simultaneously. The underwriter sees the engineering assessment alongside the operational history and market context, then synthesises them. This trains the same multi-perspective habit that distinguishes agile forecasting foxes from hedgehogs who see everything through one dominant framework.

The Bigger Shift (Practice → Philosophy)

The Evidence Tab represents a deeper shift in how we think about AI and human judgment in high-stakes domains. The real work is no longer to replace underwriters with machines. It is to create software that makes good judgment the easiest path rather than the hardest one.

This is the real promise of decision architecture. We are moving from building a better algorithm to building an environment where calibrated judgment is the default. From automation to augmentation. From trusting a single model to training a dragonfly-like synthesis of many perspectives. And from static, one-time predictions to a perpetual beta mindset.

The ultimate measure is not whether the software gets the answer right on day one. It is whether, over time, the people using it become demonstrably better at turning uncertainty into calibrated probability, and whether the organisation as a whole keeps getting less wrong.

Continuous improvement toward better calibration: each iteration moves closer to the ideal, but the process never claims to arrive

Closing on Calibration (Measurable Ambition)

A submission still arrives as a story. It always will. But the story no longer has to be the first thing that shapes the number. By letting the outside view enter the room early, we set a statistical anchor that forces the narrative to justify itself. It is one small piece of infrastructure for perpetual beta — a system that expects forecasts to evolve, judgements to be revised, and calibration to improve through deliberate practice.

The goal is not perfect foresight; it is measurably better calibration. Every insurance decision is a forecast. The best systems do not promise certainty; they help ordinary professionals become steadily less wrong. That is the quiet ambition behind the Evidence Tab: not to replace human judgement, but to make it visible, testable, and capable of improving, one submission at a time.

Every Insurance Decision is a Forecast