v2.7.0 — Predict-Before-Reveal

The Junior profile moves from type every line to predict-before-reveal: ownership becomes the ability to evaluate code, not transcribe it.

Highlights

• Predict-before-reveal loop — on each implementation task the junior commits a prediction (approach, data structure, control flow, edge cases) before the AI writes the code, then the AI grades it against a rubric and names the specific gap.
• Inline hard gate — the loop runs as a new /own:feature Phase 6 (Junior profile), so it can't be skipped; only judgment-carrying Implementation tasks gate.
• Prediction Scorecard — per-dimension MATCH/PARTIAL/MISS logged to the global learning registry, turning "is judgment improving?" into a measurable curve.
• Adaptive fading — the gate relaxes per-dimension as the junior proves the skill (4 consecutive matches), with a deterministic staleness re-check so it never decays into a rubber-stamp.
• Self-explanation OWN step — juniors explain why the actual approach is better and where their prediction broke, feeding /own:done Gate 1 with real ownership evidence.
• Evidence-based — design validated against learning-science literature and real-world industry practice (docs/research/junior-profile-predict-before-reveal-validation.md).

See CHANGELOG.md for the full list.