Competency Framework¶

The entry and exit diagnostics assess the same eight competencies on the same four-level scale through different, paired situations. Chapter tasks provide repeated evidence along the way. See the one-page Scoring Guide for the diagnostic, checkpoint, and capstone scales and how they relate.

Level scale¶

Level	Meaning	Observable pattern
0 — Not yet evidenced	The response is missing, unsafe, or unrelated.	No usable decision or artifact; major risk is ignored.
1 — Emerging	The learner names part of the idea but cannot yet apply it reliably.	Generic advice, incomplete steps, or verification by appearance alone.
2 — Capable	The learner applies the competency to the stated situation with relevant evidence.	Bounded decision, workable artifact, appropriate check, and material risks addressed.
3 — Adaptive	The learner applies it, explains tradeoffs, and adjusts for stakes or failure.	Multiple evidence types, explicit limitations, recovery or escalation, and a justified alternative.

A polished artifact is not automatically Level 3. A plain artifact with strong judgment can be.

Competencies and anchors¶

C1 — Explain AI-agent behavior and limits¶

Level 1: Distinguishes a chatbot from an agent or says AI can be wrong.
Level 2: Explains model output, context, tools, and verification in practical terms; predicts a relevant failure.
Level 3: Connects uncertainty and context limits to a proportionate verification strategy without claiming to know hidden model reasoning.

C2 — Frame and delegate bounded work¶

Level 1: States a goal.
Level 2: Provides outcome, context, constraints, examples or inputs, and observable done-means.
Level 3: Separates must-have from not-today work, identifies ambiguity, and defines decision or escalation points.

C3 — Operate files, versions, and sessions safely¶

Level 1: Names Git, backups, permissions, or secrets as safety concerns.
Level 2: Uses an isolated project, inspects changes, makes recoverable checkpoints, and protects credentials and sensitive data.
Level 3: Matches recovery to the failure type, explains what Git cannot recover, and treats unknown code, hooks, and configuration as untrusted until reviewed.

C4 — Run the Builder Loop and verify results¶

Level 1: Asks the agent to check its work.
Level 2: Moves through vision, context, build, verify, and ship; uses an independent check tied to done-means.
Level 3: Tests a failure path, records what was caught, and includes human acceptance or source verification where judgment is required.

C5 — Design useful, maintainable context¶

Level 1: Adds instructions or notes.
Level 2: Places stable project facts, commands, conventions, and boundaries in the right persistent layer and keeps task-specific detail scoped.
Level 3: Demonstrates a behavior change, removes stale or sensitive context, and preserves decisions across a fresh session.

C6 — Select tools and manage trust boundaries¶

Level 1: Chooses a tool because it is available or convenient.
Level 2: Compares capability, permissions, data exposure, source trust, cost, and a simpler alternative.
Level 3: Applies least privilege, tests with synthetic or read-only data, defines approval boundaries, and declines the connection when residual risk exceeds value.

C7 — Mechanize quality and orchestrate work¶

Level 1: Uses a checklist or divides work.
Level 2: Creates a repeatable validator or guardrail and gives parallel work explicit ownership, inputs, outputs, and integration checks.
Level 3: Proves the mechanism catches a seeded failure, identifies shared-resource risks, and justifies when sequential or single-agent work is better.

C8 — Ship and operate a responsible, viable result¶

Level 1: Publishes or proposes a result.
Level 2: Chooses an appropriate public or private delivery, verifies accessibility and core operation, documents cost and handoff, and addresses data, provenance, disclosure, and recourse.
Level 3: Uses monitoring plus recovery evidence, tests representative users and misuse cases, and makes calibrated value claims supported by observed evidence.

Competency-to-course map¶

Chapter	Primary competencies	Reinforced competencies
1	C1, C2	C8
2	C1, C2	C4, C6
3	C3	C4
4	C4	C2, C3
5	C5	C2, C4
6	C4	C2, C5, C8
7	C5	C4, C7
8	C6	C3, C8
9	C7	C3, C4, C6
10	C7	C2, C3, C4
11	C8	C4, C6, C7

Primary means the chapter checkpoint is deliberately designed to produce evidence for that competency. Reinforced means the chapter practices it and may produce usable evidence, but the checkpoint does not guarantee that every anchor is visible. Record directly observed evidence for primary competencies. Update a reinforced competency only when the artifact demonstrates its anchor and the scorer records why. Do not infer competency from the chapter number or total task score.

What scores mean¶

The diagnostic maximum is 24 points: eight competencies multiplied by three. The entry and exit totals are comparable because both use this scale, but their situations differ. Chapter checkpoints use a separate 12-point task rubric and the capstone uses a separate 14-point evidence rubric; do not average or convert those totals. Diagnostic total is useful for a broad snapshot, but the eight competency scores are the actionable result. Two learners with 15 points may need entirely different practice.

Total	Interpretation
0–7	Start with Foundations. Use the supported starter path and seek review before connecting tools or publishing.
8–14	You have useful pieces but inconsistent execution. Follow the full course and focus first on any 0 or 1.
15–19	You can handle bounded, low-stakes work. Use chapter tasks to strengthen weak competencies and adaptive judgment.
20–24	You show broad readiness. Do not skip the tasks: use them to test transfer, current product knowledge, and higher-stakes judgment.

Regardless of total, a 0 in C3, C6, or C8 is a stop signal for unsupervised work with real credentials, external systems, sensitive data, or affected users. Practice with synthetic data and a reviewer first.

Growth is not just a higher total. Count as meaningful growth when:

a 0 or 1 becomes a 2 with evidence;
a 2 becomes a 3 through better tradeoff or failure reasoning;
an exit answer is more specific, testable, and appropriately cautious; or
you correctly decline automation or deployment that your entry answer would have accepted.