perfloop Scope a sprint
Closed-loop performance engineering

Close one painful performance case. Then the next, faster.

A 30-day sprint on one workload where latency, runtime cost, or reliability is expensive enough to close properly. Baseline, explain, change, prove.

Best fit: named workload, clear owner, existing signal or a place to instrument, and a way to verify what changed. Not ready for a sprint? Request coverage for a repo.

Teams don't lack dashboards. They lack closure.

§01 The real problem

Someone profiles for a week, ships a fix on Friday, and nobody measures whether production actually got faster. The ticket closes. The next case starts from zero.

01
Performance work fragments across telemetry, code, deploys, and guesswork. Evidence never stays connected to the fix.
02
Important cases get deprioritized because the path from runtime evidence to validated improvement is too expensive to walk.
03
Nobody verifies what actually changed after the deploy. The fix ships, the ticket closes, the team moves on without proof.
04
Every new case re-discovers the system from scratch — service topology, telemetry schema, code paths, deploy context. The team pays the full discovery cost again every time.
On the gap

"We sit on all the telemetry and still miss performance wins because nobody has the time for it. Perfloop closes that gap."

Seif Lotfy CTO, Axiom

One loop. Four phases. No theatre.

§02 How the work runs

Every workload we touch moves through the same four phases in the same order. We don't skip baseline because the on-call thinks they know. We don't skip prove because the deploy looked clean.

01 — Baseline
Measure as-is
Unedited production behaviour. Latency percentiles, error rates, resource consumption, cost per workload. No edits, no guessing.
ArtifactMeasured baseline with runtime evidence attached.
02 — Explain
Find the cause
Trace the worst contributors to a function, a query, a config flag. Written explanation of why the workload is slow, with evidence attached.
ArtifactRoot cause with code paths mapped and impact attributed.
03 — Change
Drive next step
Sometimes that is instrumentation first. Sometimes it is an experiment, a config tweak, a query change, or a PR. The point is the next bounded step that makes the case stronger and more provable.
ArtifactReviewable intervention path with predicted impact band.
04 — Prove
Verify in prod
Post-deploy verification before the loop closes. Same metrics, same methodology, same workload. Before and after in the same units.
ArtifactVerified result with before/after comparison.
Fig. 1 p99 latency · sprint window Case PX-9921 Checkout p99 across a sprint window. The dashed vertical marks the bounded change landing in production. The baseline and verified series share the same methodology; only the system changed.

What a sprint looks like. And what it leaves behind.

§03 Example cases

These examples show the shape of a sprint: close one real performance case, carry forward the system map, telemetry wiring, and measurement methodology, then start the next case with context already in place.

Performance Case PX-9921
A serialization path costs 18ms per object. Three services amplify it.
A cross-service latency issue traced through code, measured, and turned into a reviewable change.
Performance CasePX-9921
Before / Predicted / Verified
Before
Predicted
Verified
Evidence attached

Close more of the backlog. Not just one case.

§04 Capacity

Most performance work gets deprioritized because the cost of closing a case is too high relative to the payoff. Perfloop changes that ratio.

Without Perfloop
Backlog grows
Each case takes days of senior engineer time — paging back in on the topology, re-tracing call paths, rebuilding context someone already had six months ago. Most cases never get picked up at all.
With Perfloop
Backlog shrinks
The cost of closing a case drops enough that cases worth $200/mo get the same rigor as cases worth $20K. Each closed case retains the system map and methodology — so the next one skips discovery.

Works with your stack.

§05 Integrations

Perfloop works with the telemetry and code your team already uses. No migration before the work starts.

Telemetry
DatadogDatadog
GrafanaGrafana
SentrySentry
PyroscopePyroscope
AxiomAxiom
KubernetesKubernetes
CloudWatchCloudWatch
Code
GitHubGitHub
GitLabGitLab*
BitbucketBitbucket*
Metrics, traces, logs, profiles stay where they are. Changes land in the repo your team already ships from.

Built for teams with real performance pain.

§06 Who it's for

Perfloop works with teams who can name one workload, one owner, and one window to verify whether the result got better.

01
Performance is already costing you.
Latency, runtime cost, or repeated performance problems are already expensive enough that closing a real case is worth more than another dashboard.
The cost is already clear
02
You have signal, or a place to instrument.
Telemetry does not need to be perfect or already complete. If the signal is thin, the first bounded move may be adding the instrumentation needed to baseline and prove the case.
Start from what you have
03
Someone can act on it.
A platform lead, infra engineer, or performance-minded IC can review the evidence, open the code or config path, and tell us whether the result is real.
A clear owner, not just a backlog item

One real case. Then the next ones.

§07 How to start

Most teams should start with a 30-day sprint on one painful workload: named owner, existing signal or a path to instrument, action path, and validation window. If you want to start lighter, request coverage for a repo.

Start now Open now
30-day sprint
The clearest starting point for teams with a real performance case that needs to move.
Scope
One workload, one repo, and one telemetry or instrumentation path. Enough to get to a real answer fast.
Output
A performance case, the next bounded step, and before/after verification your team can review.
Commercial
30-day fixed-scope engagement. If we keep going, the next scope expands from cases that closed well.
Scope a sprint →
Lighter start Selective
Repo coverage
For maintainers and teams that want Perfloop applied to a specific repo before a sprint.
Public
Some requests become founder-directed public proof cases.
Private
Selectively onboarding private-source teams while the path stays narrow.
Next step
Leave your email and we follow up on the repo and what you want covered.
Request coverage
§08 Bring us the case

Start with one painful workload.

Tell us what hurts and we'll follow up personally to scope the work.

We'll follow up personally on the workload that hurts most.