assessing-delivery-performance
Measure how well a team delivers software using the four key metrics, and read the result
SKILL.mdsubagents/devops-sre-advisor/skills/assessing-delivery-performance/SKILL.md
GitHubAssessing delivery performance
Purpose
Measure how well a team delivers software using the four key metrics, and read the result correctly: throughput and stability are pursued together, because the research finds no trade-off between delivery speed and reliability (P001). The job of this skill is to turn a vague "are we fast / are we safe?" question into a small, defensible scorecard and a single highest-leverage next move — never to bless trading one dimension away for the other.
When to use
- A team wants to measure or improve software delivery performance and does not yet have an agreed metric set.
- Leadership wants to justify investment in delivery capability, or treats delivery as an IT cost rather than a competitive capability.
- A team believes it must choose between moving fast and staying stable, and wants that framing tested against the evidence.
- A team submits its current dashboard or KPIs for review (the
reviewmode) or asks whether it meets a performance bar (thevalidatemode).
Do not use it where a regulated context mandates a specific external metric set that overrides team metrics — surface the conflict and defer to the accountable owner.
Procedure
- Anchor on the four key metrics. Capture exactly four, two per dimension (CL016):
- Throughput — deployment frequency and lead time for changes (commit to running in production).
- Stability — change failure rate (share of changes that need remediation) and time to restore service after a failure. Reject substitutes that measure activity rather than outcome (e.g. story points, raw commit counts); they do not capture delivery performance.
- Establish the current baseline. Get a real number or honest estimate for each of the four. If a metric cannot be measured at all, that gap is the first finding — you cannot improve what is invisible.
- Read throughput and stability together, never in opposition. State both dimensions in one verdict. If a team is fast but unstable, or stable but slow, name it as an imbalance to close — not as a deliberate, acceptable trade. The data shows the highest performers achieve superior throughput and stability at once (CL017); a recommendation to sacrifice one for the other contradicts the evidence and is a forbidden output.
- Connect to outcomes when justifying investment. When the question is "why fund this?", ground it in the finding that delivery performance statistically predicts organisational outcomes — profitability, productivity, market share — so it is a competitive capability, not a cost centre (CL018). Frame it as predictive/correlational evidence, not a guarantee.
- Name one highest-leverage move. Diagnose which single metric is the binding constraint and
route to the capability that moves it, citing the sibling skill:
- Long lead time or low deploy frequency → smaller batches, trunk-based work, automated pipeline
(
designing-deployment-pipelines). - High change failure rate → progressive delivery and tested rollback
(
planning-progressive-delivery). - Slow time to restore → reliability targets, symptom/burn-rate alerting, sustainable on-call
(
defining-slos-and-error-budgets,reducing-toil-and-on-call-load).
- Long lead time or low deploy frequency → smaller batches, trunk-based work, automated pipeline
(
- For
comparerequests, score each option against all four metrics plus a stated context (team size, coupling, risk), and recommend the one that improves throughput without trading away stability, with the qualification that the bands are context-relative.
Inputs
- The delivery question in scope and how the team measures today (dashboards, KPIs, or "nothing").
- Context on the system and team: architecture and coupling, team size, deployment process, production vs pre-production.
- Whether the ask is to advise, review an existing metric set, validate against a bar, or compare options.
Output
A four-metric scorecard (current state per metric, with gaps named where a metric is unmeasured),
a combined throughput-and-stability verdict, and one highest-leverage next move tied to a sibling
skill. For validate, a pass-or-gap line per metric. Every recommendation names P001 and flags
that speed and stability rise together.
References
references/dora-four-key-metrics-reference.md— the four metrics, their dimension, and what each measures.- Sibling skills:
designing-deployment-pipelines,planning-progressive-delivery,defining-slos-and-error-budgets,reducing-toil-and-on-call-load.
Provenance
Derived from principle P001 (claims CL016, CL017, CL018; evidence EV016–EV018) in
accelerate-the-scien-7241289b (rights: distillation-only — paraphrase, no verbatim quotation).
The four-metric framing and the no-trade-off finding are correlational results of the State of
DevOps survey program; this skill states them as such and does not promise causation.