# AI-Native Worksheet Pack

> Copyable operating artifacts for the six-volume AI-Native methodology. Use these files as working
> documents, not as essays: duplicate the relevant template, fill it during a real project, then feed
> the resulting artifact back into the module's context store.

## Templates

| File | Module | Use when |
|---|---|---|
| `workflow-graph.md` | Organization | A workflow is not getting faster after adding AI; map nodes, edges, human judgment anchors, and policy gates. |
| `engineering-eval-dashboard.md` | Engineering | Agent output is abundant but correctness is unclear; define verifier ownership, eval gates, and escalation rules. |
| `design-rule-matrix.md` | Design | A generated interface looks "fine" but generic; split hard rules from human taste criteria before generating again. |
| `research-evidence-ledger.md` | Research | Claims, sources, and paradigm distance are getting mixed together; keep evidence strength separate from novelty. |
| `learning-withdrawal-drill.md` | Learning | A learner or team is over-outsourcing cognition; plan what to keep, what to withdraw, and how much friction to add. |
| `innovation-bet-log.md` | Innovation | A direction looks feasible but may be a trap; write the falsifier, downside, owner, and next recognition signal. |
| `research-loop.md` | Research | You need the full RES loop: evidence base first, generation second, integration before more retrieval. |
| `spec-template.md` | Any surface | Before an agent generates anything non-trivial; turn intent into a boundary, machine-checkable acceptance assertions, and counter-examples that must fail. |
| `checker-template.md` | Any surface | Build the one load-bearing wall: a checker independent of the generator, with an assertion checklist, a run command, and a failure feedback loop. |
| `permissions-template.md` | Any surface | Before an agent touches anything world-changing; set read-only defaults, per-tool authorization, isolation, and an irreversible-action gate. |

## Operating Rule

Each filled worksheet should end with one of three statuses:

- **Ship / run**: the next action is reversible, bounded, and has a verifier.
- **Hold / learn**: the judgment is not yet externalized; gather evidence or write the missing criterion.
- **Stop / rewrite**: the falsifier fired, the downside is not affordable, or the human judgment anchor is missing.