GoLearning

What it shipped

• 01Per-engineer learning paths driven by actual codebase activity
• 02Spaced-repetition for internal systems and conventions
• 03Mentor matching by knowledge gap rather than by seniority
• 04Time-to-first-PR measurement and benchmarking

Why we built it

Static onboarding produces engineers who can recite the docs but not navigate the codebase. The gap between deck-completion and productive contribution is where most onboarding programs fail. We built GoLearning to close that gap. The curriculum adapts to where the engineer actually struggles, not to where the curriculum-author assumed they would.

What we learned

1. 01

   Time-to-first-PR is the only onboarding metric that matters.

   Time-to-orientation, time-to-deck-completion, time-to-quiz-pass: all noise. The first merged PR is the signal. Every other metric in onboarding is a leading indicator at best and a distraction at worst.

2. 02

   Mentor matching by gap, not by seniority.

   Pairing a struggling engineer with the most-senior available person is the default and the wrong answer. Matching by the specific knowledge gap (even when the matched mentor is a peer) produced dramatically better outcomes.

3. 03

   Adaptive systems beat static ones when the activity data is rich.

   GoLearning worked because we had access to real codebase activity. The principle generalizes: adaptive systems require dense behavioral signal as a precondition. Without it, prescriptive systems are honest; with it, adaptive systems dominate.

What we'd do differently

1. 01

   Lower-friction codebase activity ingest.

   Self-hosted Git was easy; complex monorepo or polyrepo setups required engagement-specific adapters that didn't scale. An open-spec ingest contract would have served us better than three bespoke integrations.

2. 02

   Adapt the existing deck, don't replace it.

   We treated GoLearning as a replacement for static onboarding decks. The teams who got the most value were ones that kept their deck and let GoLearning sit on top. Lesson: meet operators inside their existing artifact.

3. 03

   Surface struggle earlier than week one.

   We started surfacing engineer-struggle signals at the one-week mark. In retrospect, day-three signal would have been dramatically more valuable. Adaptive systems compound when they intervene early.

Why it's retired

Studio focus consolidated to flagship products. GoLearning's adaptive-system thinking was absorbed into how we approach operator onboarding in our flagship products. Every pilot operator gets a curriculum adapted to their setup and trade history rather than a static deck. The product was retired with existing engagements wound down; learning-path data was returned to operators on request.

Where the lessons live now

• 01
  Time-to-first-PR-grade discipline

  Applied to operator onboarding in Verum and Auspex pilots: only metric that matters is time-to-first-real-decision.

• 02
  Activity-driven adaptation

  Now the default principle for every personalization decision the studio ships.

• 03
  Mentor-by-gap matching

  Permanent practice in studio hiring and engineer onboarding; abandoned the seniority-default pairing entirely.