# SuperStack Final-Polish Report

**Candidate:** JPL_AUTONOMY_EDL_05
**Dissertation:** EDL Heritage and Landing-Success Hazard: Does Reuse of Flight-Proven Entry-Descent-Landing Architecture Reduce Landing-Failure Risk?
**Mode:** ACADEMIC | **Branding:** NEUTRAL (no MITRE markings, no vendor-AI names)
**Method/stage:** Firth-penalized discrete-outcome logistic hazard; DESIGN-STAGE (no coefficient fitted)
**Date:** 2026-06-16
**Files revised in place:** `dissertation.md`, `deck.md`


## Headline result

| Metric | Value |
|--------|-------|
| Em dashes before (this session) | 0 (dissertation), 0 (deck) |
| Em dashes after | **0** (dissertation), **0** (deck) |
| En dashes / horizontal bars | 0 / 0 in both files |
| Body word count (dissertation) | 63,178 (from 63,932; net change is the removal of two redundant per-chapter reference lists, not lost content) |
| Chapters / sections / subsections preserved | 8 / 73 / 62 |
| Appendices preserved | A–F (all 6) |
| Reference entries | 149 (92 grade-A, 57 grade-B; 99 DOI, 50 NTRS/URL) |
| Clickable body citations after | 792, all well-formed `[\[N\]](#ref-N)` |
| Unresolved citation markers | **0** |
| Leftover raw corpus-key brackets | **0** |
| Banned phrases remaining (non-load-bearing) | **0** |

The dissertation was authored em-dash-free (the source chapter files carried zero em dashes), so the em-dash transformation was a confirmation, not a removal. The substantive work this session was **citation unification** (a prior pass had left it incomplete and internally inconsistent across three competing schemes, with three mis-numbered table citations), a **banned-phrase purge**, and the **King-Charles-register framing** touches.


## Passes run (canonical SuperStack order)

1. **SPACE-SCIENTIFIC-WRITING (domain rigor).** Verified domain fluency and source traceability. Every in-text citation resolves to a real corpus entry (`research/corpus.jsonl`, 149 verified references) with a registered DOI or a resolvable NTRS/URL. No fabricated sources, numbers, or results. The falsifiable contribution (`beta_1` sign-and-significance on the EDL-heritage-reuse index) and the design-stage posture are intact and untouched.

2. **HAAF LOGIC BAND (mandatory tier; Toulmin wrapped at Layer 4).** Verified the nine-layer flow holds end to end:
   - L1 answer-first executive thesis present (Abstract and §1.0 state the contribution as `beta_1` / H0–H1 before any background).
   - L2 problem frame (current state → desired state → gap → consequence) present in §1.1 and §2.0.
   - L4 Toulmin units intact for each major claim (Claim / Grounds / Warrant / Backing / Qualifier / Rebuttal), most explicitly through Ch2 and Ch3.
   - L5 claim-evidence-reasoning prose (evidence interpreted, not listed) confirmed in Ch3.
   - L6 named causal mechanism (driver → mechanism → observable effect → operational consequence → strategic implication) present in §1.4, §2.5, and the Assurance Case Summary; correlation kept distinct from causation; the design downgrades to correlation explicitly where heritage proves inseparable from program strength.
   - L8 confidence calibrated to the design-stage evidence grade; minimum-detectable-effect reported alongside every non-rejection; what-would-change-confidence stated.
   - L7 assurance-case summary present and consistent (top goal, five subclaims with chapter+evidence map, assumptions, context, residual risk).
   - L9 architecture traceability correctly **declared out of scope** per `research/expansion_plan.md`; the one permitted objective-to-decision endpoint is stated in plain prose only (the fitted `beta_1` as an input a program weighs against the cost of qualifying a novel element); no architecture table is populated. Not forced anywhere.

3. **SUPER-HUMANIZER (anti-slop, authentic voice).** No surviving AI tells ("it is important to note", "in conclusion", "delve", "tapestry", "landscape", "realm", "moreover"/"furthermore" overuse). Sentence-length variation and named abstractions confirmed across chapters.

4. **MITRE-HUMANIZER (discipline only; neutral non-MITRE work).** Zero em dashes enforced. Banned-phrase purge completed (see below). Academic null-hypothesis / limitations / threats-to-validity language deliberately preserved (the zero-negative-words rule was not applied, per the brief). No MITRE branding present.

5. **NOVEL-POLISH (academic).** Opening and closing strengthened; transitions and lexical variety reviewed at the Kepler/Vox bar. Length and all data points preserved.

6. **EXECUTIVE-STORYTELLING (light, academic).** Abstract and Introduction lead stakes-first; process-label scaffolding absent; payoff arc clear under both branches of the disjunction.

7. **KING CHARLES REGISTER (framing only).** Ran `query_style_exemplars.py --mode ceremonial --limit 6` and inferred the dignified, measured, stewardship-and-service public-address register (service, duty, constancy, stewardship). Applied **only** to three executive-facing loci: the first sentence of the Abstract, the opening framing sentence of §1.0 Introduction, and the closing sentence of §8.6 Conclusion. The analytic body register is unchanged. Guardrails honored: inspired register only, never impersonation; no royal signature forms; no copied transcript phrasing; no claim of Royal Household approval.

8. **LITERARY-METHODS (Deictic Shift + Rhetoric of Science).** Temporal references anchor to calendar context; "this/that/it" referents are unambiguous; entities are introduced before pronoun reference. Epistemic modality is calibrated to the design stage: estimates are stated conditionally ("if the assembled population yields…", "a coefficient of the form…", "would estimate"), never as executed findings. The illustrative 40%→15% figure is labeled a reporting-format placeholder everywhere it appears.

9. **HAAF RECHECK (closing logic pass).** Confirmed no layer broke after the prose, voice, and register passes: every major claim retains its warrant and qualifier; the causal mechanism was not flattened to a correlation; confidence statements survived; the Layer 9 boundary still resolves to the single plain-prose objective-to-decision endpoint; the assurance-case summary remains consistent with the body.


## Citation hygiene (final deliverable requirement)

The prior pass had left the dissertation with **three competing, inconsistent citation schemes**:

- Ch1 and Ch8 used per-chapter local numbering that did **not** match the global reference numbers (e.g., Ch1 local `[15]` = Fogel, but global `[121]` = Fogel).
- Ch2–Ch6 still carried raw `[CorpusKey]` brackets (e.g., `[Railroadsnd]`, `[Puhr2017]`), about 548 of them, never converted.
- Ch7 used parenthetical author-key citations (e.g., `(Railroadsnd; Leunig2010)`), 44 of them, which a bracket-only converter had silently skipped.
- The global References list `[1]…[149]` (alphabetical by title, with `<span id="ref-N"></span>` anchors and clickable DOI/URL links) was already correct and was adopted as the single canonical scheme.

All in-text citations were unified to that canonical scheme:

- Every `[CorpusKey]` mapped to its global number by title match against the corpus (147 exact, 2 resolved by DOI: `Yu2014`→[18], `Blette2016`→[138]) and rendered as `[\[N\]](#ref-N)`.
- Ch1 and Ch8 local bare numbers were remapped to global numbers via each chapter's local reference list and prose context (validated against adjacent spelled-out keys), then rendered clickable.
- The 44 parenthetical author-key citations were converted to the same bracket-clickable form.
- The two now-redundant per-chapter "References cited in this chapter" lists (Ch1, Ch2) were removed; the canonical global list carries every reference.
- Adjacent citation runs were de-duplicated (e.g., a cluster that named the same reference twice was collapsed).
- Every reference-list DOI is a clickable `https://doi.org/...` link; every DOI-less entry resolves through a clickable NTRS/repository URL.

**Verification (post-conversion):** 792 clickable body citations, 0 malformed, 0 out of range; every distinct in-text number (142 distinct, range 1–149) resolves to a real reference entry; 149 `ref-N` anchors present; 0 leftover bare numbers; 0 leftover corpus keys.


## Banned phrases removed

Statistical "robustness" and the regression term "leverage" (high-leverage observation) are load-bearing academic vocabulary and were **preserved**. The following non-load-bearing banned phrases were rewritten:

| Phrase | Location | Replacement |
|--------|----------|-------------|
| "pivotal" | §3.2 (SLIM case) | "decisive" |
| "leveraging MSL heritage" (paraphrase) | §3.3 synthesis | "reusing MSL heritage" |
| "holistic judgment" | §5.5 coding-endogeneity | "whole-mission impression" |
| "crucial addition" | §6.5 | "load-bearing addition" |

One occurrence of "leveraged" was **kept verbatim** because it is inside a direct quotation from the Mars 2020 software-implementation report ("largely leveraged heritage from the Mars Science Laboratory EDL System…"); altering a cited quotation would violate source fidelity. The statistical "high-leverage attempt" in §5.4 (leave-one-out / influence diagnostics) was kept as a technical term.


## Citation-error corrections (introduced by the prior pass)

The prior partial conversion had mis-numbered three citations in the Assurance Case Summary table. All three were corrected against the corpus:

| Evidence label | Was (wrong) | Now (correct) |
|----------------|-------------|---------------|
| retropropulsion survey (Korzun2008) | [58] human Mars EDL | **[9]** survey of supersonic retropropulsion |
| Rangita simulation (Rangita2025) | [117] telemetry failure prediction | **[19]** Firth/ridge bias-calibration simulation |
| Faghih bias correction (Faghih2020) | [4] high-mass Mars concept | **[3]** comparative bias-correction study |

No citation's intended target was otherwise changed; only style was normalized and links added.


## Items flagged for the author

1. **Ch8 originally carried an unlabeled local numbering scheme.** The bare numbers in Chapter 8 (e.g., `[23]`=Fogel, `[24]`=Mokyr/Perez, `[22]`=Lunine ocean-worlds) had no corresponding "References cited in this chapter" list and were reconstructed from prose context plus adjacent spelled-out corpus keys. Each mapping was cross-checked so adjacent clusters do not duplicate a reference, but the author should spot-check Chapter 8's converted citations against intent, since the original local key was inferred rather than read from a list.

2. **Reference [22] in the corpus carries an "et al." author placeholder** ("Comparison of autorotation and propulsive landing for planetary exploration," Acta Astronautica, 2025). The reference list renders the information that exists rather than inventing an author. The author may wish to complete the author field before final submission.

3. **The illustrative 40%→15% failure-probability figure** appears in the Abstract-adjacent framing, Ch5, Ch6, and Ch8, and is labeled a reporting-format placeholder at each appearance. This is correct design-stage practice; flagged only so the author confirms the label survives any future rendering to DOCX/PDF where emphasis formatting can be lost.

4. **Deck citation numbers match the global reference scheme** (Fogel=121, MSL=88, etc.) but are not rendered as clickable anchors, because Marp slide references point the reader to the dissertation back matter rather than to in-deck anchors. No change made; flagged for awareness.

5. **Layer 9 remains correctly out of scope.** If the dissertation is ever extended toward an executed system or data-service exchange, the architecture-traceability layer would need to be activated; at the design stage it is properly absent and the single plain-prose objective-to-decision endpoint is the only architecture-adjacent statement.


## Convergence

- HAAF quality gate: every in-scope box checked; Layer 9 excluded from the rubric average per the out-of-scope decision; no failure-mode pattern surviving.
- MITRE-humanizer discipline: zero em dashes, zero non-load-bearing banned phrases, neutral branding.
- Citation hygiene: zero unresolved markers; every in-text number resolves to a real entry; all DOIs clickable.
- HAAF recheck: no layer broken by the prose, voice, register, or citation passes.

Both `dissertation.md` and `deck.md` are publication-ready at the design stage, preserved at full length, with every chapter, data point, citation, and the falsifiable contribution intact.