Depth by AcadableEarly access
Test coverage says the code works. Depth says someone can fix it.
Depth measures and raises comprehension coverage — how much of your codebase your current engineers have proven they understand.
Understanding Map
Payments platform
Comprehension coverage
67%4 of 6 critical modules healthy
Pick someone under “What if someone left?” to see what their resignation would cost.
The problem
Software that runs, but nobody can explain.
AI writes a growing share of every codebase. The understanding that used to form as a side effect of writing code no longer forms on its own — and nothing your team measures today can see it.
A payments feature is written mostly by AI. Review takes two minutes. Its author resigns. Six months later customers are charged twice, and nobody understands the retry logic. A one-hour fix takes three days.
How it works
Two minutes for engineers. A clear map for managers.
- 01
Connect
Install the GitHub App with read-only access, add Depth to Slack, pick your repositories. Five minutes.
- 02
Map
Depth groups your code into the modules your managers think in — payments, login, search — and drafts a first map from history within the hour.
- 03
Explain back
After a merge, the engineer answers two or three questions written from that exact change. Under two minutes, at most three times a week.
- 04
Close the gaps
Red modules get a 20-minute lesson built from your real code. Handover mode captures a leaver's knowledge before their last day.
Explain-back
Questions about why and what-if. Never trivia.
Each question is written from the change that was just merged, checked against five quality rules before it's sent, and graded with feedback that teaches.
New joiners start with Locate and Trace. Experienced owners get Justify and Spot.
DepthApp4:05 PM
Predict · Refunds · 1 of 2
If the bank's API times out twice in a row, what does the customer see?
Priya4:06 PM
'Refund processing'. We retry at 30 seconds and 2 minutes, then mark it for manual review and email the customer.
DepthApp4:06 PM
Mostly right. After the second timeout the email only goes out for orders above ₹500 — see notify.ts:88.
The metric
Comprehension coverage, precisely defined.
Built from evidence per person, per module. Writing code earns almost nothing; explaining it earns the most. We'll publish the method as an open specification.
- Understanding score
- 0 to 100, for one person on one module. 60 or more counts as understanding.
- Module status
- Healthy with two or more people. At risk with one. Red with nobody.
- Comprehension coverage
- The share of critical modules that are healthy — one number for the CTO and the board.
- Real bus factor
- How many people verifiably understand a module, not a guess from who wrote it.
Understanding fades. Depth notices.
A score halves every 90 days without new evidence. Just before it would fall below 60, a short refresh check arrives — and each one you pass makes the next come later.
Privacy by design
Built for engineers, not for surveillance.
Depth only works if engineers trust it, so the privacy rules aren't a settings page. They're how the product is built.
- You
- Your own scores and answers. Nobody else sees them.
- Your team
- Module colours and counts — never individual scores.
- Your company
- Coverage trends over time.
- Never used for performance reviews — it's in the contract
- Read-only access to your code
- Code being processed is deleted within 24 hours
- Written answers deleted after 90 days; only the score remains
- No AI training on your data; zero-retention terms with model providers
- Designed for India's DPDP Act and European works councils
Plans
Priced per engineer — for less than your AI coding tool.
Plans open after the design-partner programme. Design partners get founding pricing.
- FreeOne team trying it
- 1 repository · up to 5 engineers · the map and explain-back checks
- Team10–200 engineers
- Unlimited repositories · lessons for red zones · on-call checks
- Business200+ engineers
- Handover mode · merge guard · single sign-on · audit export
- Enterprise1,000+ engineers and GCCs
- Data kept in-country · private deployment · security reviews
Handover Pack. When a key engineer resigns, a one-off: handover mode for one person, and a report showing how complete the handover really is.
Questions
The hard ones, answered plainly.
Won't engineers resent being quizzed?
It's the first thing our pilots measure. Checks take under two minutes, arrive at most three times a week, come with free skips, and every one teaches something. Scores are private to the engineer. If fewer than half the questions get answered, we've built it wrong — and we'll tell you.
We already do code review.
Review checks the code before it ships. Depth checks that someone understands it afterwards — and for months after that. As AI makes changes larger and reviews faster, that second check is the one quietly disappearing.
Can't people just paste the question into an AI?
They could, but scores are private and never used in reviews, so there's little reason to. Questions are time-boxed and ask about context the code doesn't contain — team decisions, past incidents. Depth treats every score as evidence with a confidence level, never a verdict.
Isn't this surveillance?
Not by design. Individual scores are visible only to the engineer. Managers see module colours and, for handovers and on-call planning only, whether a person is covered on a module — never the score.
Tools like CodeScene already show knowledge loss.
They infer it from who wrote the code. When AI writes a large share of it, authorship stops meaning understanding. Depth verifies with evidence from the people themselves.
Design partner programme
Help us build Depth with your team.
A four-week pilot with one or two teams, success numbers agreed before we start, a day-30 report for your CTO, and founding pricing.
Apply as a design partnerA good fit: 30–300 engineers, AI coding tools across the team, GitHub and Slack, and a recent outage, resignation or AI rollout that made the gap obvious.