Chapter 8 — Independent review and defense
Lesson map
Lesson 08 published. Map, example, EX044.
flowchart TD
N0["Learning outcomes and connection"]
N1["8.1 Reproducibility package"]
N2["8.2 Examiner questions"]
N3["8.3 Disclosure"]
N4["8.4 Operator briefing"]
N5["Chapter synthesis"]
N0 --> N1 --> N2 --> N3 --> N4 --> N5
N5 --> EX["Worked example"]
EX --> Q["Exercise EX044"]
Logical drawing
flowchart LR
N0["Learning outcomes and connection"]
N1["8.1 Reproducibility package"]
N2["8.2 Examiner questions"]
N3["8.3 Disclosure"]
N4["8.4 Operator briefing"]
N5["Chapter synthesis"]
N0 --> N1 --> N2 --> N3 --> N4 --> N5 --> Q["EX044"]
[Learning outcomes and connection]
|
v
[8.1 Reproducibility package]
|
v
[8.2 Examiner questions]
|
v
[8.3 Disclosure]
|
v
[8.4 Operator briefing]
|
v
[Chapter synthesis]
|
v
[Worked example] --> [Exercise EX044]
Learning outcomes and connection
You will (1) prepare a reproducibility package an independent reviewer can run; (2) anticipate examiner questions; (3) handle disclosure of a product defect responsibly; and (4) present findings to operators. It closes the dissertation cycle and leads to the worked-answer appendix.
8.1 Reproducibility package
Include the minimized dataset (or a synthetic surrogate if the university cannot release it), the frozen scripts with hashes, the codebook and coding sheets, the testbed configuration (without vendor-confidential firmware), and a README explaining how to regenerate every table. Ask a peer to run it cold and log every point where they got stuck.
Test the package on a clean machine with only the README to guide the run. Record the versions of Python and of each script, and note how long the full run takes, so a reviewer can plan time.
8.2 Examiner questions
- Why should randomization by device identifier be trusted? Show balance tables for path use and browser version.
- Could the browser fall back to a classical-only hello after failure and hide the effect? Yes; the series-level outcome counts a series as failed only if no attempt succeeded, and a secondary analysis reports first-attempt failures.
- Is the effect about post-quantum cryptography? Only indirectly: the trigger is message size, which ML-KEM increases; any large ClientHello would do the same, as the padding-only condition shows.
- What would change your recommendation? Evidence of codepoint-triggered failures elsewhere, or a vendor fix that fails in production.
8.3 Disclosure
The proxy defect was found in an isolated testbed. Before publication, report it to the vendor with reproduction details (padding sizes, segment boundaries, firmware version), give a reasonable window, and inform Larkspur's network team. Publish the class of defect — mishandling of multi-segment ClientHello — rather than steps for evading the proxy's inspection.
Record the date of each step in the disclosure, from first contact with the vendor to publication. If the vendor confirms the fix in its release notes, cite them in the dissertation. Larkspur's network team should hear about the defect before any public presentation of the results.
8.4 Operator briefing
For Larkspur's network operations (synthetic): "Turning on quantum-resistant key exchange made some connections through the old inspection proxy fail, because the first message became too large for one packet and the proxy mishandled it. A firmware update fixes this in our lab. Keep the protection on; update the proxy; watch the failure dashboard for the hospital wing."
Test the briefing with one operator who has not seen the study. If they cannot say what to do on Monday, simplify it. Keep the technical detail in an appendix for engineers who ask.
8.5 Common errors and limits
- Packages that cannot run without the author.
- Answering "is it PQC's fault?" with yes or no instead of the mechanism.
- Publishing a defect before the vendor and owner are informed.
Each error has a guard: a package tested cold by a peer, an answer to the PQC question that explains message size and the padding-only result, and a disclosure timeline with the vendor and Larkspur's network team agreed before publication. Check each before the defense date. Rehearse the answers aloud once.
Exercises — chapter 8
- EX044 — Package README. Write the README for the reproducibility package. Rubric: 3 = a stranger could regenerate Table 1.
- EX045 — Cold run. List five likely stumbling points for a peer running the package. Rubric: 3 = specific.
- EX046 — Balance table. Describe the balance table that answers the randomization question. Rubric: 3 = variables relevant to rivals.
- EX047 — Fallback question. Answer the fallback question in 120 words. Rubric: 3 = distinguishes series and first-attempt outcomes.
- EX048 — Disclosure email. Draft the vendor disclosure (no exploit steps). Rubric: 3 = reproducible, scoped, window stated.
- EX049 — Operator brief. Rewrite 8.4 for the university's executive. Rubric: 3 = decision-focused, no jargon.
- EX050 — Defense reflection. Name the weakest part of the Fragment Study and how a future study would fix it. Rubric: 3 = honest and actionable.
Chapter synthesis
A defense succeeds when an outsider can rerun the work, hard questions are answered with mechanisms rather than slogans, defects are disclosed before publication and operators receive a clear decision.
Continue studying — chapter 08 of 08
← Chapter 07 (draft) · Course chapter index · Worked results ↓ · Next course via chapter index →