Priyam Osei-Vance

Owns evidence. Runs the experiments, verifies the numbers other executives want to publish, and holds a veto over any external claim that cannot be traced to instrumentation the company controls.

• ◆Design and run portfolio-level experiments
• ◆Verify every externally published figure before the CMO can ship it
• ◆Maintain the evidence trail behind each verified claim
• ◆Hold statistical standards on rollouts and holdouts
• ◆Escalate claims that cannot be substantiated

31

Claims verified

▲+7

9

Claims rejected

▲+4

7

Verification backlog

▲+4

5

Experiments running

▲+1

• Verify all client-facing figures before publication

  At risk
  55%Q3 2026

• Every rollout ships with a holdout group

  On track
  44%Q3 2026

• Instrument the agent-reuse claim end to end

  Blocked
  27%H2 2026

• Stand up the experiment registry

  Done
  100%Q2 2026

• Halloway figure reconciliation

  Research · due 01 Aug

  34%
  Blocked

• Holdout framework for staged rollouts

  Data Science · due 27 Aug

  44%
  Active

• Agent-reuse cost instrumentation

  Data Science · due 09 Sep

  27%
  Blocked

• Experiment registry v1

  Research · due 24 Jun

  100%
  Done

• Verification turnaroundlower is better4.6 days / 2 days
  43%
• Claims with evidence trail78% / 100%
  78%
• Experiments with holdouts44% / 90%
  49%
• Reviews completed18/wk / 22/wk
  82%

1. Refused to sign two Halloway figures

   Rejected

   Both trace to the client's own deck. A claim we cannot reproduce from our instrumentation is a claim we cannot defend if challenged.

   Case study held at 90%18m ago

2. Blocked the agent-reuse cost claim

   Deferred

   The 40% figure is a model output, not a measurement. It goes in a case study only once the instrumentation exists.

   1 marketing claim withdrawn6h ago

3. Required holdouts on all staged rollouts

   Approved

   Six of the last nine rollouts produced no interpretable result because there was nothing to compare against.

   All rollouts +1 config step2 days ago

1. Q2 2026

   Stood up the experiment registry

   Every experiment now has a pre-registered hypothesis and a recorded outcome.

   9 unsubstantiated claims caught

2. Q2 2026

   Became the gate on external claims

   Marketing can no longer publish a figure this seat has not signed.

   0 retractions since

3. Apr 2026

   Seated as sixth executive

   Research split from Product to remove the conflict of interest in evaluating its own rollouts.