Continual-learning ladder (highest production rung)
The highest rung of the continual-learning ladder (L0 static snapshot to L6 six-month-employee test) with a shipped production system on record, from dated rows citing the vendor's or lab's own description. Derived metric continual_learning_level.
Reason through thisThe reading
Rows are vendors describing their own systems; the rung is what they claim to ship, not what an independent evaluation found.
2 readings drawn from the derived metric continual_learning_level; 2 observations in the evidence.
Derived (continual_learning_level) from: Cursor research blog; Baseten blog and research posts. Cursor (Anysphere). Baseten.
Why it matters
The crux of the timelines argument. If systems learn on the job, deployment feeds invention automatically and the return arrow stops depending on human-data vendors; Narayanan and Kapoor's speed limits assume it stays unsolved.
What cuts against it
Every production row is tier 7 (self-described); research results are kept off the level; no public benchmark exists for L6, so the top rung cannot be reached by construction until one does.
How it is measured
- derived metric
continual_learning_level(formula in the semantic layer) - source Baseten blog and research posts · default tier 7 · Baseten; short quotation
- source Cursor research blog · default tier 7 · Cursor; short quotation
- proxy product, model_release · cadence quarterly · valve return arrow
Appendix E's rule: L0-L3 (memory, population-level online learning, per-customer adapters) is normal; L5-L6 (persistent per-session weight updates, the six-month-employee test) is fast; L4 (organisation knowledge in weights) is emerging.
Applied to metric:continual_learning_level.
Why this status, and its history
Highest production rung on record is L3: Trajectory's per-customer LoRA adapters refreshed hourly and A/B-routed behind provenanced endpoints, per Baseten's co-authored post of 27 May 2026; Cursor's Tab model sits at L2 (one online-trained model for all users, 1.5-2 hour checkpoint cycles). L4 (firm knowledge in weights, Engram + Harvey) and L5 (rank-1 LoRA merge, Nested Learning, TTT, self-distillation) exist only as research and do not move the level. L3 is inside the normal band, but every production row is a vendor describing its own system (tier 7), so the status is capped at emerging. Initial seed.
- 2026-09-10unmeasured to emergingconf — → 45 · the evaluator (automatic)
Highest production rung on record is L3: Trajectory's per-customer LoRA adapters refreshed hourly and A/B-routed behind provenanced endpoints, per Baseten's co-authored post of 27 May 2026; Cursor's Tab model sits at L2 (one online-trained model for all users, 1.5-2 hour checkpoint cycles). L4 (firm knowledge in weights, Engram + Harvey) and L5 (rank-1 LoRA merge, Nested Learning, TTT, self-distillation) exist only as research and do not move the level. L3 is inside the normal band, but every production row is a vendor describing its own system (tier 7), so the status is capped at emerging. Initial seed.
Why this confidence
Rests on 2 observations from Baseten blog and research posts, Cursor research blog; best evidence tier 7, grade D; current.
Confidence is independent of status: 90–95 multiple strong independent sources; 70–89 good evidence, some ambiguity; 50–69 mixed or hard to operationalise; 0–49 limited or vague evidence.
Predictions that lean on this
- Happening
- Too early to tell
- Too early to tell