A team runs a self-assessment across five areas: intellect, personality, motivation, behavior, experience. The problem-solving numbers are generous. The resilience numbers are generous. The drive numbers are, remarkably, generous.
Meanwhile, in a less inspected corner of the company, the evidence sits quietly. The handoff logs show which requests arrived with context and which arrived as vibes. The repository shows whose wins became systems and whose evaporated. None of it was asked its opinion. None of it could perform.
The assessment and the artifacts disagree, and the review trusts the assessment, because the assessment was commissioned and the artifacts were merely produced.
Which of the five areas can you actually measure?
Two of the five, honestly.
Behavior, how work crosses people, deposits evidence every day: handoffs that arrived with context or without it, escalations that carried the right amount of detail, requests that could be acted on the first time. Experience, the conversion of wins into repeatable systems, is even harder to hide from: the systems either exist and run without their author, or they do not, as Ten Years Is Not Ten Years of Experience argues. Both areas leave artifacts. The artifacts are legible, dated, and indifferent to the self-image of their authors.
The other three areas do not leave artifacts. Intellect, personality in the operational sense of resilience and adaptability, and motivation are observed in one of two ways: through self-report, or through live observation. Self-report has a well-documented upward drift, because a questionnaire is a small performance and everyone knows their lines. Live observation fixes the performance problem and installs a worse one: people behave differently while observed, so the measurement moves the thing it measures. Neither instrument reports ground truth, because for these three areas there is no ground truth to report, only capacity under whatever load was present that day.
The consequence is structural, not moral. An assessment anchored on the self-reported three is anchored on the flattering three. An assessment anchored on the auditable two is anchored on whatever actually happened. Validation at Machine Speed, Evidence Still Decides makes the same point about ventures: at any speed of generation, the claim remains a claim until evidence adjudicates it. Team assessment is the same adjudication at lower clock speed.
Start the audit where the evidence lives
The order of operations matters more than the instrument. Most reviews gather the self-reports first, then, if time permits, glance at reality. Reverse it.
- Sample ten handoffs from the last month, unannounced and unattributed to individuals. Count how many arrived with enough context to act on the first time. That number is the behavior score, and it does not round up.
- List the systems that run without their author present: defaults, playbooks, escalation rules, anything extracted from past wins. That list is the experience score, in the only currency that compounds.
- Only then read the self-reports, with the artifacts beside them as a calibration weight. Where the two agree, believe both. Where they diverge, believe the artifacts and get curious about the gap, because the gap itself is diagnostic. It usually marks either an area running below its self-image, or a genuine strength the current work never gives room to display.
This inversion also protects the assessment from the contamination described in A Demotivated Quarter Looks Like a Capability Problem. Self-reported scores respond to morale; artifacts respond to what actually occurred. When the questionnaire says everything is fine and the handoff logs say otherwise, the disagreement is the finding.
What this is not
Not a claim that the unobservable three do not matter. Intellect, resilience, and drive decide most of what makes a team worth building. The claim is about epistemics, not importance: they cannot be read off a form, so they must be inferred from what they produce under real load. The artifacts are how the three invisible areas show themselves. A hard problem attempted is intellect made visible. A setback absorbed in public is resilience made visible. Pushed-through work no one asked for is drive made visible. Audit the outputs, infer the capacities.
Not an invitation to surveillance. The audit samples work products in aggregate: handoffs and systems, not messages and keystrokes. The moment artifact-reading becomes person-watching, the instruments start performing again, and you are back to measuring a stage.
And not a promise that artifacts are unbiased. Handoff logs reward teams whose work crosses many boundaries and understate teams whose work runs deep inside one. Every artifact carries the bias of its substrate. That is a calibration problem, and it is small next to the flattery problem it replaces.
Take-home
- Rerun your last team assessment starting from the artifacts: ten handoffs, one systems list. Read the self-reports only after, as claims to be checked rather than data to be averaged.
- Treat any divergence between the two as a finding, not an error. The gap is where the real scorecard lives.
- Keep the artifact audit aggregate and unattributed. Evidence works only while it is not watching anyone.
Related reading:
- Ten Years Is Not Ten Years of Experience
- A Demotivated Quarter Looks Like a Capability Problem
- Teams Ship at Their Weakest Area, Not Their Average
- Validation at Machine Speed, Evidence Still Decides
- Confident Wrong Output Marked Done Is the Agentic Failure Mode
If your team reviews run on questionnaires and the quarter runs on artifacts, the two are already disagreeing somewhere. If you want help building the artifact-first audit before the next review season, book a discovery call.
FAQ
Which team capabilities can actually be measured?
Behavior and experience. Behavior deposits daily evidence in handoffs, escalations, and requests that either carry context or do not. Experience shows up as systems that run without their author. Intellect, resilience, and motivation leave no artifacts; they are observed through self-report or live observation, both of which distort what they touch.
Why do self-assessments inflate?
Because a questionnaire is a performance, and everyone knows their lines. Self-reports of intellect, resilience, and drive drift upward without anyone lying deliberately. The fix is not better questions. It is anchoring the assessment on the areas that leave artifacts and using them to calibrate the self-reported claims.
How do you audit a team using artifacts?
Sample ten handoffs from the last month in aggregate and count how many were actionable on first contact. Then list the repeatable systems that survive their author's absence. Read those two numbers before any self-report, and treat divergence between artifacts and questionnaires as the finding rather than noise.
Doesn't artifact-based assessment become surveillance?
Only if the artifacts are attributed to individuals. The audit works on aggregate work products, handoffs and systems, never on messages or monitoring. The moment measurement turns into watching, people begin performing for it, and the instruments lose the ground truth they existed to provide.