Skip to main content

Engineering health brief

Northstar Labs, a fictional sample organisation · Last 90 days · 1,284 merged pull requests

Review pickup is the constraint: 31 hours waiting, 20 hours reviewing

Northstar Labs cut cycle time 14% without weakening reliability, so the bottleneck has moved. Work now waits longer for attention than it spends in active review, and 18 pull requests sat unclaimed for more than 48 hours.

Read-only GitHub · Your code stays yours · Team-level by default · No mandatory call · No credit card

Engineering health score

B76 / 100 · Stable
Cycle time3.3 days · C
Review coverage91% · B
Merge without approval4.1% · B
Test failure rate7.2% · C
Deployment frequency1.6 / day · A

These are the five dimensions CodePulse grades, against published thresholds. Each bar is that dimension's score out of 100; the figure beside it is the measurement behind the score. Every dimension stays visible, so a single letter never hides a weak one.

Median cycle time

3.3 days

↓ 14%

Improved versus the previous 90 days

Review pickup

31 hours

↑ 22%

The largest and fastest-growing wait state

Review coverage

91%

↑ 4 pts

Most merged work received human review

Merge without approval

4.1%

↓ 1.2 pts

Merged PRs with no approving review

Every reading below describes the system - queues, handoffs, review load, ownership - and never individual productivity. Each finding is followed by the experiment it suggests and the measure that would show it worked.

Decision layer

Evidence-backed action plan

Included with GitHub alone

The report turns observed constraints into bounded experiments. It does not invent revenue, headcount, or causality that the connected data cannot establish.

1

Evidence

Review pickup now consumes 39% of cycle time and 18 PRs waited longer than 48 hours.

Experiment

Route unclaimed reviews after 12 working hours and set a 24-hour first-review service level.

Success measure

Bring median review pickup below 18 hours without reducing review integrity.

2

Evidence

Seven frequently changed files have a bus factor of one; two sit in the payments path.

Experiment

Assign a secondary owner and require paired review on the highest-risk modules.

Success measure

Reduce critical single-owner files from seven to three this quarter.

3

Evidence

Bug-fix work rose to 17% while feature investment fell four points.

Experiment

Reserve one capacity lane for the two defect clusters creating the most rework.

Success measure

Return bug-fix allocation below 12% while keeping change failure under 8%.

Your plan will name your constraints, not Northstar's. The first version is ready once your initial sync completes.

The same constraint, in a real organisation

“I'd been telling the CTO we were slow because of tech debt. Turns out almost a third of our PRs were just sitting there waiting on the same two reviewers. We redistributed reviews and the difference was noticeable within a couple weeks.”
Robert StrikerVP of Engineering at Article Fiesta

Review concentration is the most common constraint we find, and it is invisible in a sprint board. It shows up in the review network further down this report as two people carrying the connections everyone else runs through.

Northstar’s figures are illustrative. This one is not.

Delivery flow

Where the 3.3-day cycle time goes

Included with GitHub alone

Phase timing separates active work from queue time, so a faster team is not confused with a less-blocked system.

Coding

18h

23% of total cycle time

Waiting for review

31h

39% of total cycle time

Active review

20h

25% of total cycle time

Approval to merge

11h

13% of total cycle time

Delivery forecast

For the current 23-PR queue: 50% likelihood of clearing within 4 days; 85% within 8 days, from a Monte Carlo simulation over historical throughput and PR size.

Production outcomes

DORA metrics from deployment events

Requires deployment events

With GitHub deployment data connected, the report distinguishes shipped software from merged pull requests, and grades each metric against the published DORA thresholds.

Deployment frequency

11 / week

Elite performance tier

Measured from deployment data

Lead time to deploy

1.8 days

High performance tier

Measured from deployment data

Change failure rate

7.2%

Medium performance tier

Measured from deployment data

Time to restore

3.4 hours

High performance tier

Measured from deployment data

Tiers follow the published DORA thresholds, including the unflattering ones: 7.2% change failure rate is Medium, not High. Without deployment data, CodePulse marks merge-based proxies with an amber indicator instead of presenting them as production outcomes.

Investment

Where engineering capacity went

Included with GitHub alone

PRs are classified into system-level work types so leadership can discuss portfolio balance without counting individual activity.

Features
44%
Bug fixes
17%
Maintenance
16%
Technical debt
15%
Infrastructure
8%
Bug-fix allocation rose five points while feature work fell four. The report flags the shift; leadership decides whether it is planned stabilization or unplanned rework.

End-to-end flow

From work item to merged change

Requires an issue tracker or GitHub Issues

With Jira, Linear, or GitHub Issues connected, the report includes backlog dwell instead of starting the clock only when coding begins. Link coverage is reported beside it, because every issue-derived figure depends on it.

Backlog dwell

1.6 days

Issue created → first commit

Build and review

3.3 days

First commit → merged PR

Total issue lead time

7.8 days

Issue created → merged PR

Issue linkage coverage

76%

Merged PRs connected to work items

Issue keys are read from branch names, PR titles, and GitHub closing references, so link coverage is a data-quality figure worth watching rather than a target.

Review system

Review load and review depth in the same view

Included with GitHub alone

Volume alone cannot tell you whether review is healthy. CodePulse combines team-level workload concentration with the tone and content of the review discussion itself.

Who reviews whom, weighted by volume

PPriyaMMarcusDDanSSofiaTTomasAAna

Reviews submitted

1,046

Top-two concentration

42%

Waiting over 48h

18 PRs

Actionable comments

74%

Share of review comments classified as actionable

Constructive tone

88%

Constructive or positive rather than critical

Bugs and security issues flagged

14

Comments categorised as bug or security findings

Average comment length

214 chars

Depth signal across 1,278 review comments

Knowledge and change risk

Hotspots that need a second pair of hands

Included with GitHub alone

Changed-file paths, churn, ownership, and review history expose operational concentration without ranking developers.

payments/settlement.ts

46 changes · 1 primary owner · churn up 31%

High risk

Add a secondary owner before the next settlement release.

auth/session-manager.ts

38 changes · 2 owners · 6 high-risk PRs

High risk

Require two reviewers when authentication behaviour changes.

reporting/export-worker.py

29 changes · 1 primary owner · no review backup

High risk

Pair on the next change and document the operational runbook.

AI investment

AI spend, output, acceleration, and reliability

Requires an AI provider connection

When AI providers are connected, the report compares spend with shipped work and quality signals. Before-and-after movement is labelled as association, not proof of causality.

Observed seats

42 / 58

72% roster coverage

AI spend

$4,860

Across connected providers

Merged PRs

312

$15.58 per merged PR

Acceleration

+12%

Medium confidence versus pre-adoption

Throughput improved after adoption, while revert rate remained at 3.8%. The correct conclusion is “promising, keep measuring”, not “AI caused a 12% gain”.

Scope

What you get on day one, and what each integration adds

This sample shows a fully connected organisation. Most of it needs nothing but read-only GitHub. The rest is labelled, and stays absent rather than estimated until you connect the source.

From GitHub alone, in about five minutes

  • Engineering health score across all five graded dimensions
  • Cycle time split into coding, review pickup, active review, and merge
  • Review coverage, merge-without-approval, and review load distribution
  • Review comment depth, tone, and bug or security findings
  • File hotspots, churn, and single-owner knowledge risk
  • Work-type investment mix from PR and branch labels
  • Delivery forecast for the current queue

Added by optional integrations

  • GitHub deployment events

    DORA measured from real deploys: deployment frequency, lead time to production, change failure rate, time to restore

  • Jira, Linear, or GitHub Issues

    Backlog dwell, full issue-to-merge lead time, issue link coverage, and issue-derived work types

  • AI provider (Anthropic, OpenAI, Copilot)

    AI spend against merged output, seat roster coverage, and pre- versus post-adoption movement

None of these are required to start, and none are billed separately.

Confidence and trust

Know what the report can, and cannot, conclude

A credible report exposes its data boundary. Missing integrations reduce confidence instead of being filled with invented estimates.

Coverage in this sample

  • 1,284 merged pull requests across 8 repositories
  • 93% work-type classification coverage
  • 1,278 review comments categorised for tone and content
  • 76% of merged PRs linked to an issue
  • 72% of the AI-tool seat roster observed

Coverage is reported, not assumed. Where it is low, the figures that depend on it are labelled rather than quietly presented as complete.

Connection boundary

  • Read-only GitHub permissions; CodePulse cannot push, merge, or modify repositories.
  • Your code stays yours: repositories are never cloned and files are never read.
  • The only code stored is the short snippet GitHub attaches to an inline review comment, retained so review depth and quality can be assessed.
  • Results are team-level by default; individual views require an explicit organisation setting.
Review the full security boundary

You leave with the artifact, not just the dashboard

This brief is a view in the product, and it is also something you can put in front of a board without rebuilding it in slides the night before.

PowerPoint export

The executive summary as a .pptx deck, with the health grade, the phase breakdown, and the action plan already laid out.

PDF and CSV

The report as a PDF for circulation, and the underlying figures as CSV when someone wants to check the arithmetic.

Scheduled digests

A recurring summary to you or your leadership group, so the next board pack starts written rather than blank.

Every plan is one flat price per organisation, whatever your headcount. We never bill per seat, so this report does not get more expensive as your team grows.

See plans and pricing

Your data next

Replace the fictional evidence with your own delivery history

Connect in about five minutes. Your initial analysis runs after sync, and your fourteen usable trial days begin when the first repository is ready.

Start with selected repositories · No mandatory call · No credit card