The labor bridge

Where agent capability meets human work.

The Runtime Atlas maps how agent runtimes work. This is the seam to the other map: the human work each capability augments. Read the Atlas from a primitive and you see the skill it touches; read it from here and you see, for a piece of human work, the agent capabilities now reaching into it. 8 edges across 4 skills.

Tier: reasoned hypotheses

The Atlas primitives these edges link from are runtime-verified. These mappings are not — they are reasoned, ed25519-signed for provenance, and held to a deliberately lower bar. The signature proves each edge is mine, unaltered, and grounded in a real source; it does not claim the mapping is tested or objectively correct.

← The Atlas Singulariki: the work map ↗
Active Learning ↗ human work · O*NET via Singulariki
The self-improver invented its own safety limits Finding The whole behavior is an agent learning from its own runs and acting on what it learned.
Skill hot-reload Primitive A skill that hot-reloads and rewrites itself from each run is machine active-learning - the O*NET work requirement it most directly augments.
Skill Runtime Class Self-modifying skills learn from their own execution - active learning as a runtime property.
Judgment and Decision Making ↗ human work · O*NET via Singulariki
The cost frontier: pay for intelligence only where it bends the outcome Finding Deciding which model tier each piece of work belongs to is a routing judgment under a cost-quality tradeoff.
The self-improver invented its own safety limits Finding Setting its own stopping rule at ten iterations is the skill exercising judgment under uncertainty - the human capability it mirrors.
Learning Strategies ↗ human work · O*NET via Singulariki
Skill hot-reload Primitive Choosing how to revise itself across ten iterations is selecting learning strategies under uncertainty.
Skill Runtime Class A runtime whose units of behavior rewrite themselves is the machine analog of selecting and revising learning strategies.
Systems Evaluation ↗ human work · O*NET via Singulariki
The cost frontier: pay for intelligence only where it bends the outcome Finding Measuring whether a cheaper model still clears the bar for a class of work is systems evaluation.

Each edge is a signed cross-link emitted through @f3/attest: it cannot exist without citing the source that grounds it, and it carries an ed25519 receipt. "Verified derivation" means something precise and narrow — the engine refuses to emit a bridge entry whose signature does not check, so every edge here is genuinely *signed and unaltered*. It does not mean the mapping is correct; that is a separate, untested claim (above).

One edge, fully resolved
  1. atlas primitive The self-improver invented its own safety limits
  2. → signed cross-link cross-link_f93a1b4b60f6210add3e4134
  3. → ed25519 receipt verifiable on the primitive's page (the "verify edges" button recomputes it in your browser)
  4. → augments Active Learning (O*NET, via Singulariki)
public key 9b87705613b1e2fd064d57fa75a6b679d2856ceafad6b1daa8f982493871b6dd
← The Atlas Work with me