GitHub delivery evidence
Build and verify the change with GitHub context alongside it.
GitHubCycle timeCI insightsBranch risk
See where a pull request actually waits.
Cycle time, code churn, CI reliability and branch risk, read from the GitHub activity already connected to lube. Delivery flow you can drill into, without turning your team into a leaderboard.
pull request · cycle time
2d 4h- Open to first review6h
- Review to approved1d 2h
- Approved to merged3h
- Merged to deployed5h
Insights read the work happening in Engineering.
One connected loop, held on the stage this capability serves. The other stages stay as context so you can see what feeds in and what comes next.
BD
Discovery
Feedback
Product
Engineering
Release
Reliability
Evidence
Engineering
Build and verify the change.
Available now
VCS insights
Inspect delivery workflow activity in the connected release workspace.
Test intelligence
Investigate test outcomes and flakiness evidence across runs.
API Explorer
Keep API contracts inspectable next to the release that changed them.
Compute pricing
Compare public cloud compute pricing as planning evidence.
Flow, not a leaderboard
Insights read GitHub activity and describe how work moves: where a pull request waits, which pipeline keeps failing, which branch is risky. They describe the system, not the people in it, and lube will never rank your engineers.
What we read from GitHub
One OAuth connection and lube ingests the repos you pick — commits, pull requests and reviews, CI runs and jobs, and branches — over webhooks, then mirrors each repo to compute line-level churn and lines-of-code. Every metric is derived from that real history, so what you read is the activity you already have, not an estimate.
Watch merged catch up to opened.
Opened and merged pull requests per week, drawn from the GitHub history lube already indexes. A widening gap means review is falling behind; a closing one means the queue is clearing.
pull requests · per week
opened merged
W1W2W3W4W5
Churn, split four ways.
A git mirror classifies every line into new work, rework of your own recent code, help on a teammate's, or legacy refactor — then derives a survival rate and the AI-vs-human split. It measures the codebase, never a person.
lines written · this window
- New work48%
- Rework27%
- Help others16%
- Legacy refactor9%
- Survival rate
- 64%
- AI-assisted lines
- 38%
A red pipeline, made diagnosable.
Runs, checks and jobs feed a failure rate per workflow, colored by threshold, plus duration percentiles and recovery time. The flakiest workflow surfaces itself instead of hiding in the noise.
workflow · failure rate
runs- 214e2e19%
- 892unit6%
- 640lint2%
- 431build4%
- p50 dur
- 3m40s
- p90 dur
- 11m
- recovery
- 18m
Break it down, then drill in
Scope every read by repository, monorepo path, date window up to 400 days, and humans-only or with bots. Commits filter by provenance — human, AI-assisted, or agent — and group by PR; PRs filter by state and author; CI runs by conclusion and workflow. Open any pull request or commit for its full lifecycle.
Connect once, then it flows.
One OAuth, four reads.
Authorize GitHub, pick the repositories to ingest, and lube starts reading the four surfaces a change touches — over webhooks, keyed so replays never double-count.
commitspull requestsCI runsbranches
- 01
Connect GitHub
OAuth, then pick repos
- 02
Ingest
webhooks + signature verify
- 03
Project
commits, PRs, CI, branches
- 04
Read
cycle time, churn, CI health
Where a PR actually waits.
The stages from open to deployed, drawn to scale from merged pull-request history. A long review wait shows up as the widest bar, not a hunch.
real review threadsper stagedrillable
pull request · cycle time
22h- Opened to first review6h
- Review to approved14h
- Approved to merged2h
One flag where it's needed.
Activity per contributor fills to its share. Only a blocked or stalled reviewer is flagged, so attention lands on the one thing to unblock, not the whole team.
activity shareat-risk onlyno leaderboard
- dana14 PRs
- sam11 PRs
- rae3 open · stalled reviewat risk
- kai7 PRs
Bigger PRs, longer waits.
Reviews are bucketed by change size, xs to xl, with the median time to review each. The larger the pull request, the longer it sits — a pattern you can act on when you split the next one.
by size bucketmedian review timemerge rate
- xs2h
- s5h
- m11h
- l1d 4h
- xl2d 6h
Backfilled, then kept honest.
On connect, lube backfills history so the charts are populated from day one, then reconciles on request to heal any missed webhook — and retention runners prune raw detail on schedule.
historical backfillon-demand reconcileretention
- 01
Backfill
history on connect
- 02
Sync
live over webhooks
- 03
Reconcile
on request, heal gaps
- 04
Retain
prune raw on schedule
GitHub today. More on the way.
The entire insights engine is native — every chart is computed by lube, not the provider. GitHub is connected now; GitLab and Bitbucket are coming, and we mark them coming soon rather than pretend they are live.
- GitHubOAuth · commits, PRs, CI, brancheslive
- GitLabingestion in progresscoming soon
- Bitbucketon the roadmapcoming soon
The details most dashboards skip.
From CI conclusions to how AI-assisted work is detected, the raw material is specific and the method is written down.
CI conclusions
How a run ended.
successfailurecancelledskippedtimed_outaction_required
Branch risk
Computed level.
lowmediumhighblocked
Branch ops
With dry-run.
deleterenamerefreshprotectunprotect
AI detection
How it is inferred.
co_author_trailermessage_markerauthor_identitycommitter_identity
Governance
Events tracked.
releasessecurity alertsrule eventsbranch protection
Search
Over commits.
shamessageauthortrigram full-text
GitHub
connect it and cycle time, churn, and CI insights fill in
Per-PR
review latency read from real review threads, not estimates
CI MTTR
recovery time tracked over runs so a red pipeline is diagnosable
Risk score
computed per branch with structured reasons before any cleanup
Follow the change further.
Ground your delivery conversations in real activity.
Start free
Connect GitHub to get started.