Write a short reproducibility report

1. 01

   Concepts & system design

   60 min· read · map the mechanismtimer

   R34 — TransformerLens docs

   R42 — NIST GenAI evaluation limits

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

2. 02

   Project implementation

   120 min· build · integrate · debugtimer

   Write a concise, rigorous Mechanistic Interpretability Reproducibility Report: experimental setup, model/seed, logit-lens and activation-patching results, failure cases, and explicit statement of what the experiment does *not* prove about safety or alignment.

   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 70 — 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

   Completed S10 Interpretability Report ready for the final portfolio.

   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

• R34TransformerLens Docs↗ transformerlensorg.github.io

  HookedTransformer, residual stream, ActivationCache, Logit Lens, and Activation Patching tutorials.

• R42NIST AI RMF GenAI Profile↗ doi.org

  NIST AI 600-1: generative AI risks and suggested actions for confabulation, privacy, information integrity, and value chain.

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

• docsNIST (NIST AI 600-1) · PDF, ~50 pages
  Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile

  Provides the documentation and risk-communication expectations that shape the report's explicit limits section on what the experiment does not prove about safety or alignment.

  R42checked 2026-11-11
• videoAAIF Live · 2 min
  Provenance and Reproducibility in Machine Learning; what is it and why you need it?

  Defines provenance as the record of data, code and parameters behind a result, the minimum metadata spine for the report's setup, model revision and seed section.

  R42checked 2026-11-11
• videoThe Artificial Intelligence Channel · 38 min
  Reproducibility in Deep Reinforcement Learning - Prof. Pineau - NIPS2017

  Explains seed variance, environment sensitivity and reporting practices that make results reproducible, which is how you justify single-run versus multi-seed claims and document failure cases.

  R42checked 2026-11-11
• articleNeurIPS · checklist
  PaperInformation / PaperChecklist

  Gives the field-standard reproducibility checklist (code, data, seeds, compute, limitations), a ready scaffold for the setup, results and limitations sections of the report.

  R42checked 2026-11-11
• repoGitHub · checklist repo
  GitHub - craig-willis/reproducibility-checklist: Proposed checklist for the computational reproducibility review and verification

  Turns reproducibility into pass/fail verification items, so you can state precisely which parts of the logit-lens and patching runs another person could and could not reproduce.

  R42R34checked 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.