Chapter 7 — Dissertation writing and limitations
Lesson map
Lesson 07 published. Map, example, EX038.
flowchart TD
N0["Learning outcomes and connection"]
N1["7.1 Structure"]
N2["7.2 Writing results"]
N3["7.3 Limitations with direction"]
N4["7.4 Recommendations"]
N5["Chapter synthesis"]
N0 --> N1 --> N2 --> N3 --> N4 --> N5
N5 --> EX["Worked example"]
EX --> Q["Exercise EX038"]
Logical drawing
flowchart LR
N0["Learning outcomes and connection"]
N1["7.1 Structure"]
N2["7.2 Writing results"]
N3["7.3 Limitations with direction"]
N4["7.4 Recommendations"]
N5["Chapter synthesis"]
N0 --> N1 --> N2 --> N3 --> N4 --> N5 --> Q["EX038"]
[Learning outcomes and connection]
|
v
[7.1 Structure]
|
v
[7.2 Writing results]
|
v
[7.3 Limitations with direction]
|
v
[7.4 Recommendations]
|
v
[Chapter synthesis]
|
v
[Worked example] --> [Exercise EX038]
Learning outcomes and connection
You will (1) structure a dissertation so that each chapter answers a question; (2) write results that separate estimate, uncertainty and interpretation; (3) write limitations with direction and consequence; and (4) make recommendations proportionate to evidence. It turns Chapter 6's results into a document examined in Chapter 8.
7.1 Structure
Introduction (question, why it matters, contribution); background (quantum threat, ML-KEM, hybrids, middlebox ossification); method (design, data, minimization, analysis plan with amendments); results (rates, DiD, coding, testbed, validation); discussion (mechanism, scope, recommendations); limitations; conclusion. Each chapter should open with the question it answers and close with what the next chapter needs.
Draft the results chapter first, then the methods that produced them, and write the introduction last, so it does not promise more than was delivered. Check that each chapter's opening question is answered by its last paragraph. The discussion should connect testbed and log evidence explicitly rather than presenting them separately.
7.2 Writing results
Separate three layers in every results paragraph. Estimate: "proxy-path failures rose by 20.4 per 10,000 series relative to control." Uncertainty: "95% CI 15.1–25.8; coding κ = 0.88." Interpretation: "consistent with the size mechanism, which the testbed isolates." Never put interpretation in the estimate sentence. Report the null result at Fenwick in the results chapter, not in a footnote.
Apply the same pattern to the testbed and to Fenwick: estimate, uncertainty, interpretation in separate sentences. Generate the numbers in each sentence from the analysis scripts, so that a revision of the data cannot leave stale figures in the text.
7.3 Limitations with direction
- Only managed browsers were observed → unmanaged devices on the guest network are excluded; the effect for them is unknown.
- Testbed proxy is the same model but a lab instance → configuration differences could make the production effect larger or smaller.
- Eight weeks → rare destinations may be under-represented, likely underestimating failures for unusual servers.
- Synthetic data in this teaching book → no claim about real products.
Add a column for magnitude where it can be estimated, and update the table if later evidence settles a direction. Unmanaged devices could be studied in a follow-up if the network team agrees, which would turn an unknown into a measured result.
7.4 Recommendations
Proportionate recommendation: update the legacy proxy's firmware (validated in testbed) and monitor proxy-path failures; do not roll back the hybrid group for the direct path, where no effect was found; ask the proxy vendor to confirm the fix in release notes. Disproportionate recommendation, to avoid: "inspecting proxies are incompatible with post-quantum cryptography".
Write each recommendation with its owner and the evidence that supports it. The firmware update needs production monitoring after rollout, because the testbed showed the fix only in a lab instance. If proxy-path failures do not fall after the update, the recommendation is revisited.
7.5 Worked example: abstract
"Enabling the post-quantum/traditional hybrid TLS group X25519MLKEM768 at a synthetic university increased failed handshakes on a legacy inspecting-proxy path by 20.4 per 10,000 connection series (95% CI 15.1–25.8) relative to a concurrent control, with no meaningful change on the direct path. Testbed experiments showed failures occurred only when the ClientHello spanned two TCP segments, independent of the new codepoint, and stopped after a firmware update. A second site with a different proxy showed no effect. Repairing the specific device preserves quantum protection; rollback is unnecessary."
7.6 Common errors and limits
- Interpretation mixed into estimates.
- Limitations listed without direction.
- Recommendations broader than the product and site studied.
Each error has a guard in this dissertation: three-layer results sentences, a limitations list where each item names its direction, and recommendations limited to the legacy proxy model at Larkspur. Read the abstract once more against the results tables before submission. Ask a peer to mark any sentence that mixes estimate and interpretation, any limitation without a direction, and any recommendation that names a category of products instead of the device studied.
Exercises — chapter 7
- EX038 — Three layers. Rewrite a paragraph that mixes estimate and interpretation. Rubric: 3 = layers separated.
- EX039 — Limitation directions. Add two limitations with directions. Rubric: 3 = directions argued.
- EX040 — Proportionate recommendation. Write recommendations for a site where the testbed showed codepoint-triggered failures instead. Rubric: 3 = matches mechanism.
- EX041 — Abstract critique. Identify any overclaim in the 7.5 abstract. Rubric: 3 = notes synthetic status and single-model scope.
- EX042 — Chapter openings. Write the first two sentences of each dissertation chapter. Rubric: 3 = each poses its question.
- EX043 — Null result paragraph. Write the Fenwick results paragraph. Rubric: 3 = reports CI and scope implication.
Chapter synthesis
A dissertation answers its question chapter by chapter, separates estimates from interpretation, gives limitations a direction, reports null results prominently and recommends no more than the evidence supports.
Continue studying — chapter 07 of 08
← Chapter 06 (draft) · Course chapter index · Chapter 08 (draft) →