Show the work. Keep the proof.

Keep one private evidence folder per project

1. Follow that module’s setup; copy the supplied defense template into your own folder before editing. Keep earlier kits and original files untouched.
2. Implement the named learner boundaries. Record any helper, reference, hint or tutor assistance; a wrapper is not independent implementation of its helper.
3. Run checks against practice, then produce actual changed-input artifacts with your own code. Keep exact commands, environment/revision, outputs and exit statuses.
4. Retain a failed attempt, root-cause explanation and repair. Demonstrate one normal, boundary and failure case; do not erase failures to make a clean story.
5. Ask another person to reproduce a run and observe an unfamiliar transfer. Record what they actually saw, including stalls and unresolved gaps. If nobody observed it, say self-review.
6. Keep source, outputs, tests and completed defense together. Backup & records covers saved browser notes/progress, not your computer’s project files. Back up those files separately.

Use invented fixtures. Do not upload private source, quarantine, full tracebacks, credentials or identifying file paths. Nothing on this page uploads files or sends evidence to a tutor.

Portfolio I — Asset Intelligence CLI

An owned fixture inventory, declared metadata validation, dry-run CLI and actual NEW-folder export. It does not decode or certify media.

What you own

Your inspect_batch, validate_ingest, audit_event and practice_cli.py integration. Disclose reviewed inventory, naming, QC and pipeline helpers.

Your actual artifact set

Your code and tests, a changed mixed fixture, the dry-run report, manifest.json, manifest.csv, audit.jsonl, report.json and your completed PORTFOLIO.md.

Changed inputs and visible failures

Reconcile every inventory path to accepted plus failed metadata. Explain mixed status 1 versus malformed/fatal status 2. Show an existing-output refusal without changing its files.

Run your practice checks

python -X utf8 -B -m pytest -q --target practice

Here python means the exact environment interpreter in the linked setup, not a system fallback. Portfolio III also requires the explicit extracted Module 11 PYTHONPATH and separate Module 10 environment. Read the kit README before running commands. Unfinished practice is expected to fail; passing reviewed support alone does not complete your boundary.

Fresh defense and assistance

Explain a normal, boundary and failure test. Add a new requirement test before the repair, rerun regressions and ask another person to reproduce one run.

Copy and complete: PORTFOLIO.md. Use the detailed rubric in the template; keep missing/unobserved criteria pending.

Portfolio II — Evaluation Dataset Toolkit

Typed evaluation layers, explainable rubrics and an actual import/validate/audit dataset CLI. Valid schema or matching hashes do not establish training eligibility.

What you own

Your model construction, observation_bundle, rubric strategy/composition, ship_dataset and practice_cli.py. Disclose the reviewed Module 6/7 support snapshots and I/O helpers.

Your actual artifact set

Your code/tests and changed CSV; exact source.csv, preferences.jsonl, conversion.json, audit.json, notes.json, data-card.json, manifest.json and completed PORTFOLIO.md. Source/quarantine can contain private text.

Changed inputs and visible failures

Reconcile converted and quarantined rows; preserve a tie as needs_review, name lost fields and demonstrate changed/missing provenance on an owned output copy. Keep a machine/human disagreement without inventing a verdict.

Run your practice checks

python -X utf8 -B -m pytest -q --target practice

Here python means the exact environment interpreter in the linked setup, not a system fallback. Portfolio III also requires the explicit extracted Module 11 PYTHONPATH and separate Module 10 environment. Read the kit README before running commands. Unfinished practice is expected to fail; passing reviewed support alone does not complete your boundary.

Fresh defense and assistance

Add a criterion without editing the first strategy, predict weighted results and show missing reviews remain unscored. Reproduce actual import/validate/audit outcomes and explain remaining privacy/rights/judgment limits.

Copy and complete: PORTFOLIO.md. Use the detailed rubric in the template; keep missing/unobserved criteria pending.

Portfolio III — Local Evaluation API

A specified owned loopback workflow, actual quarantine inspection, persistence, review and readiness observations. This classroom system is not a public production deployment.

What you own

Your SPEC.md, capstone_practice.run_workflow, changed-input tests, DEFENSE.md and CASE_STUDY.md. Select your completed Module 10/11 implementations where claimed; identify every reviewed transport/model/path/inspection helper.

Your actual artifact set

Your specification, code/tests and selected implementation revision; actual workflow.json, quarantine/payload.bin, quarantine/receipt.json, persisted finding/audit observations, incident evidence, DEFENSE.md and CASE_STUDY.md.

Changed inputs and visible failures

Use returned IDs/revisions, a new owned output and changed media/evidence. Demonstrate strict-schema, actor and stale-state refusals; stop/restart your own service and read retained finding/audit. Never reinitialize the existing database for restart evidence.

Run your practice checks

python -X utf8 -B -m pytest -q test_capstone.py --target practice

Here python means the exact environment interpreter in the linked setup, not a system fallback. Portfolio III also requires the explicit extracted Module 11 PYTHONPATH and separate Module 10 environment. Read the kit README before running commands. Unfinished practice is expected to fail; passing reviewed support alone does not complete your boundary.

Fresh defense and assistance

Have a reviewer observe fresh transfer: predict first, show a refused self-review/stale update, make an owned dependency not-ready then recover it, and add a regression before a new requirement. Explain measured versus supplied metadata and retained partial effects.

Copy and complete: SPEC.md · DEFENSE.md · CASE_STUDY.md. Use the detailed rubric in the template; keep missing/unobserved criteria pending.

Judge each claim separately

Missing

No actual artifact or observation supports this claim yet. Record the next action.

Attempted

You tried it and retained the result, including failure. A command transcript alone may not prove the requested behavior.

Guided

The result used examples, hints, reviewed helpers or tutor assistance. State the precise assistance and what you implemented.

Independently demonstrated

A reviewer actually observed you predict, change, test and explain fresh inputs. Record the observed scope and remaining gaps; a self-entered label is not independent verification.

No average converts a missing critical requirement into a pass. An automated check tests declared behavior, not authorship, creative judgment, learning efficacy or production security. Do not present unobserved work as defended.

Continue learning while evidence is incomplete; there is no checkbox lock. Return to the course map.