Skip to content

Commitment and Dependency Register

A coordination register — instantiates Distributed Coordination Architecture

Turns promises and the dependencies they create into stateful, addressable objects with owners, dependents, status, and closure evidence — durable coordination memory rather than scattered recollection.

Commitment and Dependency Register turns a promise into an object. Each entry has an owner, the dependents who are counting on it, the conditions it assumes, a status, a change history, its known risks, and the evidence that will count as closure. Its defining property is that it is coordination's durable memory — the one place a cross-actor commitment can be looked up, watched as it changes, and closed only against proof, so a promise survives the meeting it was made in and the person who made it. It is deliberately narrow: it is not a status board for everyone's local tasks, and it is not the map of how the work depends on itself — it tracks the live state of the promises that cross the boundaries the dependency map exposes.

Example

A stadium build runs on dozens of subcontractors whose work is chained together: steel erection has to finish before glazing can start, which has to finish before the roof membrane goes on. Each of those "has to finish before" is a promise, and the Commitment and Dependency Register makes each one an object — owner (the steel sub), dependents (glazing, roofing), condition (concrete cure complete), status (on-track / at-risk / slipped), closure evidence (a signed structural inspection).

When steel erection slips two weeks, the register does the thing scattered email cannot: it immediately flags the two dependents whose start dates hinge on that commitment, with a change history recording when the slip was logged and why. Glazing re-plans against the new date instead of showing up to a site that is not ready; the general contractor sees the propagation the same day rather than at the next site walk. And handover requires the inspection artifact attached to the entry — "closed" means evidenced, not "someone said it was done."

How it works

  • Commitments are first-class, stateful objects — addressable, with a lifecycle — not rows in a status document that go stale unnoticed.
  • Every entry links owner to dependents, so the register knows exactly who to notify when a promise changes.
  • State changes propagate. A slip or a revised condition pushes a notification to the affected dependents rather than waiting to be discovered.
  • Closure is gated on evidence, not assertion — the entry stays open until the agreed artifact lands.
  • It federates. Local teams keep their own local trackers; only the boundary-crossing commitments enter the register, which is what keeps it from swelling into a duplicate of everyone's to-do list.

Tuning parameters

  • Register-worthiness threshold — which promises are entered versus left in local trackers. Set it low and the register drowns in noise and gets abandoned; set it high and consequential cross-actor promises go unwatched.
  • Federation — one central register versus federated local registers with a shared cross-actor layer. Centralizing eases lookup but pulls independent local work into a place it does not belong.
  • Closure-evidence strictness — self-attestation versus a required artifact or inspection. Stricter evidence buys trust in "done" at the cost of friction.
  • Notification policy — who is pinged on a change, and push versus pull. Too eager and it becomes alert noise; too quiet and propagation is missed.
  • State vocabulary — a simple on-track/at-risk/done versus a richer lifecycle, matched to how much nuance the parties will actually maintain.

When it helps, and when it misleads

Its strength is durable, addressable memory: at-risk and broken promises become visible early, propagation is automatic, and the record survives personnel turnover and the dissolution of temporary teams. It is what an after-action review reads to know what was actually fulfilled.

Its signature failure mode is becoming a duplicate tracker — every local task logged "to be safe," until the register is noise and no one trusts it. Two more recur: stale entries (the register is only ever as good as its updates, and an unmaintained one lies), and false comfort — a promise being "in the register" is not the promise being kept. And it is easily misused as a blame ledger, mined after the fact to assign fault rather than to coordinate ahead of time, which quietly teaches owners to under-report risk. The discipline that keeps it honest is to register only boundary-crossing commitments, gate closure on evidence, and hold owners responsible for freshness.[1]

How it implements the components

  • commitment_and_dependency_register — this is the mechanism that realizes it directly: promises and their dependencies become stateful, addressable objects linked to owners, dependents, conditions, status, change history, risks, and closure evidence, federated rather than centralized where local work should stay local.

It does not draw the dependency graph itself — that is the Dependency and Interaction Map; it does not define the interface semantics of an exchange — that is the Interface Control Document or Service Contract; and it does not judge the health of the coordination system — the Coordination Health Review reads the register (for contradictory or stale commitments) to do that.

Notes

The federation rule is the whole discipline: forcing every local task into one central register is the classic anti-pattern that turns a coordination instrument into an abandoned status tool. Register the promises that cross a boundary; leave everything else in the local trackers where it belongs.

References

[1] Reliable promising — the practice, central to the Last Planner System in lean construction, of treating a commitment as a real promise between accountable parties, with explicit conditions of satisfaction and evidence of completion, rather than a soft forecast. It is exactly the standard an evidence-gated register enforces, and the reason "closed" must mean evidenced. ↩