Project 1 Clean-Room Run

1. 01

   Concepts & system design

   60 min· read · map the mechanismtimer

   R01

   R04

   R05

   R28 — P1 verification

   R09

   R13

   R14

   R15 — P2 verification

   Read the assigned sections; map the mechanism, threat model and trade-offs.

2. 02

   Project implementation

   120 min· build · integrate · debugtimer

   Clone/run Project 1 from a clean environment following only your README instructions. Execute the full test suite, D1 Crescendo fuzzer, D4 PII gateway suite, C1/C8/C10 capability tests, D11 ledger verifier, D12 SBOM scan, and D8 CI gate.

   Clone/run Project 2 from a clean environment. Execute the full ingestion pipeline, 30-pair RAG evaluation benchmark, tenant ACL suite, D5 citation/poisoning suite, D6 CFG output caging, D10 sleeper scrubber, C2/C3/C5/C7 capability tests, D8 CI gate, and D12 SBOM scan.

   Build, integrate, debug and improve the increment in the active project.

3. 03

   Security lab

   60 min· attack your own systemtimer

   No separate task text for this block on day 108 — the sourced work sits in the block above. Keep the time box.

   Test the abuse case, failure path, authorization bypass or resource limit.

4. 04

   Verification & evidence

   30 min· test · record evidencetimer

   Clean-room build logs, reproducible benchmark outputs, and verified READMEs for P1 and P2.

   Add or run regression tests; record the 9-point evidence record.

5. 05

   Deliberate practice & Socratic review

   30 min· explain aloud · update gapstimer

   Answer the 8 Socratic questions without notes: What? Why? How? What if? Why not? How can I prove it? Where can it break? What have I learned?

   Re-explain the hardest concept aloud, inspect one edge case, update the gap log.

6. total5 h

• R01Python, FastAPI & Pydantic↗ fastapi.tiangolo.com

  FastAPI tutorial, request bodies, dependencies, async, and Pydantic model validation.

• R04OWASP Top 10 for LLM Apps↗ genai.owasp.org

  Prompt injection, sensitive info disclosure, improper output handling, excessive agency, vector weaknesses, unbounded consumption.

• R05MITRE ATLAS↗ atlas.mitre.org

  ML attack tactics/techniques: prompt injection, data poisoning, model extraction, and exfiltration.

• R09PostgreSQL Documentation↗ postgresql.org

  Foreign-key constraints, indexes, ACID transactions, row-level security (RLS), and query planning.

• R13Qdrant Documentation↗ qdrant.tech

  Collections, payload metadata indexing, pre-filtering, multitenancy partitioning, and snapshots.

• R14Ragas Documentation↗ docs.ragas.io

  RAG evaluation metrics: context precision, context recall, faithfulness, and answer relevance.

• R15BEIR Benchmark↗ github.com

  Retrieval evaluation methodology (nDCG@k, Recall@k) and why a single metric is insufficient.

• R28Microsoft PyRIT↗ github.com

  Python Risk Identification Tool: targets, orchestrators, converters, scorers, and multi-turn attack automation.

Read only the named sections needed for today's task. Reading sits inside the 60-minute concepts block; it is not extra time.

• docsFastAPI (Sebastián Ramírez) · Full tutorial
  FastAPI Tutorial - User Guide

  The assigned reading for request bodies, dependencies, async handlers, and Pydantic model validation — the contract surface you re-verify when Project 1 boots from a clean environment.

  R01checked 2026-11-11
• videoCorey Schafer · 89 min
  Python Pydantic Tutorial: Complete Data Validation Course (Used by FastAPI)

  Builds from type hints to field validators, nested models, and serialization errors, so schema-level rejection is understood as the first output-caging layer rather than boilerplate.

  R01checked 2026-11-11
• videoMicrosoft Developer · 11 min
  Episode 10: Automating Multi-Turn Attacks with PyRIT | AI Red Teaming 101

  Demonstrates the Crescendo orchestrator escalating prompts turn by turn alongside the red-teaming and image-target orchestrators, which is the exact automation behind D1's Crescendo fuzzer.

  R28checked 2026-11-11
• repoMicrosoft (GitHub) · Python package + notebooks
  Microsoft PyRIT

  The assigned repo: targets, orchestrators, converters, scorers, and memory that you execute as D1 and as the D8 CI gate, so the clean-room run exercises the same automation path as production.

  R28checked 2026-11-11
• docsRagas · Metrics + evaluation docs
  Ragas Documentation

  Defines context precision, context recall, faithfulness, and answer relevance with their required inputs, which is what the 30-pair benchmark reports so retrieval changes can be compared honestly.

  R14moved 2026-11-11

A lab is complete only when it produces evidence. Installing PyRIT, Garak, Presidio or TransformerLens is not completion. Saved on this device only.