Documentation

Coding Visual Learning Studio

After Final Submission on a coding arena run, WinJob can open the Visual Learning Studio at /app/sprint/mock/coding/[runId]/studio. The Studio is a calm technical cinema for comparing your execution with the Reference Solution — not a second AI judge.

What you see

ViewPurpose
Twin Trace CinemaYour execution and the reference side by side, with a shared scrubber
Flow / Sequence / Code compareThe same Trace IR projected onto CFG, call sequence and source lines
Complexity hotspotWhere work concentrates relative to the Optimal Reference
First DivergenceThe first behavioural fork that explains a failing case
Help ladderWhat’s wrong → hint → strategy → Reference Solution
RetryGuided repair modes that seed a fresh arena attempt

Practice modes control when visualisation is allowed: Learn may show viz during the solve; Interview, Timed and Blind unlock the Studio only after Final Submission.

Trace IR versus the LLM

Three layers stay separate on purpose:

  1. Sandbox execution runs your code (and the reference, when available) under hard time, memory and output limits. Pass / fail and variable deltas come from that run.
  2. Trace IR is the deterministic event stream (LINE, CALL, RETURN, EXCEPTION, …) recorded by the Python or JavaScript tracer. Every playback frame maps to a source line in that IR.
  3. The LLM narrates the scorecard, strategy labels and coaching copy. It must cite traceEventId / source ranges when it talks about what ran. If narration conflicts with runtime, runtime wins and the conflicting claim is dropped.

Languages without a local tracer (for example Java, C++ or Go without an optional Judge0 adapter) still get Final Submission evaluation. The Studio shows a clear empty state when cinematic Trace IR is unavailable — evaluation is never blocked by a missing trace.

Honesty rules that do not change

  • Citations must quote your submitted lines before a hire signal is shown.
  • Paste-heavy (unobserved) runs earn no Prep XP and keep the Reference Solution locked.
  • One Final Submission review per run; Rematch and retry create a new run.
  • Quiz, flashcards, system design and the live AI mock interviewer are separate hub modes and are not rewritten by the Studio.