Role, situation, evidence, forecast, decision and outcome: the record is what turns individual calls into a method that can be scored and improved.
Why it matters when the plan changes
Most advice about people is delivered and never scored. The forecast is filed, the decision is made, and nobody later compares what happened with what was expected, so the next forecast starts from the same beliefs. Research on forecasting shows what tracking changes: Barbara Mellers and colleagues found that training, teaming and tracking forecasters improved both calibration and resolution over a 2-year geopolitical tournament.
The tension is between learning and privacy. A record that links people, roles and outcomes is valuable precisely because it is detailed, and that detail is what makes it sensitive. Learning from it across customers needs a defined purpose, the necessary rights, an appropriate lawful basis, data minimisation and sufficient aggregation, and without those the record stays within the organisation it came from. The value and the risk grow together, so the controls have to.
In practice
A company has run forecasts on several senior appointments. For each, the record holds the role, the situation the appointee walked into, the evidence used, the forecast, the decision taken and, a year later, what happened. When a new appointment comes up, the team can see which kinds of calls held and which did not, instead of starting from impressions.
Evidence
In a 2-year forecasting tournament, training, teaming and tracking forecasters improved both calibration and resolution.
Barbara Mellers and colleagues, Psychological Strategies for Winning a Geopolitical Forecasting Tournament, Psychological Science (2014)Model-risk guidance defines outcomes analysis as comparing model outputs with corresponding real-world outcomes.
Board of Governors of the Federal Reserve System, OCC and FDIC, Revised Guidance on Model Risk Management (SR 26-2) (2026)
What it cannot tell you
A learning record improves a method only once enough outcomes have been recorded, and outcomes have many causes besides the forecast. It shows whether calls tracked reality over time; it cannot prove any single call right, and cross-customer learning depends on rights and aggregation that must be established first.
How Atlas reads it
The useful data unit is role, situation, evidence, forecast, decision and outcome. Over time it builds two assets: measurement IP and outcome-labelled organisational evidence. Customer data remains separated by default, and cross-customer benchmarks are created only with a defined purpose, the necessary rights, an appropriate lawful basis, data minimisation and sufficient aggregation. Outcome capture is one of the next workflows being made repeatable, not a live feature.
Questions
Six elements: the role, the situation, the evidence used, the forecast made, the decision taken and the outcome that followed. The link between them matters more than any one part. Without the outcome, the record describes what was believed; with it, the record shows whether the belief held.
Because forecasts that are never scored do not improve. Barbara Mellers and colleagues found in a 2-year forecasting tournament that tracking forecasters, along with training and teaming, improved both calibration and resolution. An organisation that records its people decisions and their outcomes can apply the same discipline to its own calls.
Not by default. Customer data stays separated. Cross-customer benchmarks are created only where there is a defined purpose, the necessary rights, an appropriate lawful basis, data minimisation and enough aggregation that no individual or customer is exposed. Consent in employment is not assumed to be freely given.
Outcome capture and audit records are among the next workflows being made repeatable, not live product features. Learning across contexts is further out and described as direction. The record describes the intended data unit; how much of it is captured in a given piece of work depends on what has been agreed and built.
Training data teaches a model patterns in general. A learning record documents specific judgements and what came of them, so it can be used to score a method, calibrate its confidence and find where it fails. Model-risk guidance from the Federal Reserve in 2026 calls the equivalent comparison outcomes analysis.