Responsibility Without Authority Is the Oldest PM Trap.

Two Sprints Below Baseline. Same Cause Both Times.

The microservices platform was the unsung star of Phase 2 until sprint 11. Sprints 7 through 10 had averaged 50 points, slightly above baseline. Sprint 11 closed at 32 points. Sprint 12 closed at 32. Two data points are not yet a trend. They are also not yet a coincidence.

The PM booked forty-five minutes with Priya Raman (Technical Lead) and Jin-ho Park (Development Lead) on 4 August. The conversation did not need to be longer than that. Within fifteen minutes, both engineers had described the same root cause from their two different vantage points.

The Falcon program had no RACI matrix. The charter (Post 02) had named decision authority at program level. The Stakeholder Register (Post 03) had tiered governance. But no artifact spelled out, at the activity level, who is Responsible for executing each piece of cross-cutting work, who is Accountable for the outcome, who must be Consulted, and who must be Informed. For the first ten sprints this had not mattered. By sprint 11 it mattered enough to cost 33 percent of velocity.

One Activity Was Actually Four.

The first instinct was to draft a RACI with one line called “Observability” and assign it to a single team. That was the wrong instinct, and the conversation with Raman and Park made it visible quickly. Observability on a 19-microservice platform is not one activity. It is at least four, and they have different natural ownership.

Logs are service-owned. Each microservice writes its own logs and decides what to log, what to redact, and how long to retain. The platform team can publish logging conventions, but the service team executes.

APM dashboards are platform-owned. Cross-service views require a single source of definitions for SLIs and SLOs. Asking each service team to define its own dashboards produces 19 incompatible dashboards. Platform owns this; services consume it.

Distributed tracing is shared. The platform owns the instrumentation library and the trace collector. Each service is responsible for propagating trace IDs correctly across its calls. Neither owns it alone.

Quality gates are platform-defined, service-enforced. Platform sets the gate thresholds. Each service team is responsible for keeping their service compliant. Platform cannot enforce; services cannot define.

The decomposition was the unlock. Once observability was four lines instead of one, R/A/C/I assignments became obvious. The next morning the PM scheduled a working session with Raman, Park, Hassan, and Volkov to author the matrix.

Volkov’s Team Was Also Writing Microservices.

The integrator team was building several of the 19 services. The question that surfaced in the working session: do they own observability for their own services, or is it a platform-applied responsibility uniform across all services?

Volkov argued for the first. His position was contractual: the integrator scope is what is in the contract; observability for vendor-built services is what was scoped. Alemu argued for the second. His position was operational: Atlas Bank carries the readiness consequence at go-live; uniform observability is a precondition for production support, regardless of which team wrote which service.

Both were correct from where they sat. The resolution was procedural, not commercial. Vendor team gets a named role on the RACI matrix, distinct from Park’s development team. Where the contract scoped a vendor responsibility, the cell carries an explicit contract clause reference. Where Atlas Bank applied responsibility on top, the cell shows it as a platform-uniform requirement with the vendor as Consulted, not Responsible. The RACI did not relitigate the contract; it documented where contract met operational reality.

Drafting the RACI

RACI Draft, Extract

The AI produced a 12-row matrix. Three flagged items are shown below; the human edits resolved each before the matrix went to working session.

The third flag was the one the PM had not seen coming. The instinct in regulated environments is to over-Consult; it feels safer. In practice, over-Consulting produces meetings where the Consulted parties expect to drive decisions they were never going to drive, and the velocity penalty is identical to no RACI at all. Tightening C versus I was the single edit that mattered most for sprint velocity recovery.

RACI Matrix v1.0, Extract

Published 15 August 2025. Eight rows shown across the 12-row matrix, focused on the observability stack and the three other cross-cutting activities where ownership ambiguity had cost velocity.