The CDI companion guide

What the guide covers

The companion guide is the long-form partner to the one-page CDI Canvas. The Canvas is for the assessment session. The guide is for what comes before and after it: understanding the dimensions, calibrating scoring, facilitating an honest conversation with the team, and turning the assessment into ninety days of focused work. It presents the Index as a measurement instrument, not a maturity model. It is silent on whether agents are good, fast or appropriate for the work, and asks one question of every dimension: can you demonstrate that the control is real?

Each of the six dimensions gets its own section, with a five-level scoring rubric, the signals that mark each end of the scale, and one concrete action to move forward. For Provenance Integrity, that action is to pick one defect from the last sprint, ask the team to reconstruct the full provenance of the change that caused it, and track how long it takes and what was missing.

Key points

• A control is real when it produces observable behaviour under pressure. A team at level four can pass an audit. A team at level five can demonstrate, with examples from the last quarter, that the control changed an outcome.
• The facilitator’s job is to challenge level four claims. If nobody can name an occasion in the last quarter where the control changed an outcome, the dimension is at three, not four.
• Four result patterns are described: the flat, audit-ready, split and brittle profiles. Fours across the board with no fives is the governance theatre profile.
• Six failure modes are named, including the optimism tax, the documentation reflex, the hero pattern and the big bang ambition, each with a symptom and a remedy.
• The ninety-day focus is a movement of one level on a single dimension: the largest gap relative to where the team needs to be, feasible given current capacity, and likely to unlock progress elsewhere.
• The Index is designed to run quarterly, with an annual half-day deep assessment and triggered assessments after any agent-related incident. It is meant for internal use, because external reporting tends to invite optimistic scoring.