# SuperStack Final-Polish Report

**Candidate:** JPL_MGMT_SMA_TECH_04
**Document:** *Does Flight Heritage Buy Reliability? A Cross-Mission Regression of Realized On-Orbit Failure Rates Against Heritage and Parts-Class*
**Mode:** ACADEMIC | **Branding:** NEUTRAL (no MITRE markings, no vendor-AI names)
**Date:** 2026-06-16
**Method:** targeted edits in place; no from-scratch rewrite. Every chapter, data point, citation, and the full length preserved.


## Headline metrics

| Metric | Before | After |
|--------|--------|-------|
| Em dashes (U+2014) in dissertation | 0 | 0 |
| En dashes (U+2013) / horizontal bars (U+2015) | 0 / 0 | 0 / 0 |
| Em dashes in deck | 0 | 0 |
| Word count (dissertation) | 61,807 | 61,807 |
| Reference-list entries | 116 | 116 |
| In-text citations (body) | 836 | 836 |
| Citation scheme | string keys `[\[key\]](#ref-key)` | numbered `[\[N\]](#ref-N)` |
| Unresolved citation markers | 0 | 0 |
| Clickable DOI links | 104 | 104 |

The document arrived already em-dash-free and substantively compliant with the HAAF logic band. The polish was therefore light-touch on prose and concentrated on the citation-hygiene transformation.


## Passes run (canonical stack order)

1. **SPACE-SCIENTIFIC-WRITING (domain rigor).** Audited. Domain fluency and scientific rigor already strong; every load-bearing claim traces to a real retrieved source with a DOI or resolvable URL. No fabricated content. No change required.

2. **HAAF LOGIC BAND (enforced, mandatory tier; HAAF wraps Toulmin at Layer 4).** Verified the nine-layer flow holds across all eight chapters and the appendices:
   - Answer-first executive thesis present per chapter (each opens with "the chapter's answer, stated first"); contribution stated as H0/H1.
   - Problem frame (current state, desired state, gap, consequence) present in every chapter.
   - Major claims carry full Toulmin. Chapter 5 uses explicit **Claim / Grounds / Warrant / Backing / Qualifier / Rebuttal** headers (3 full blocks); Chapters 1-4, 6-8 carry the same structure in prose. No bare Toulmin run separately (HAAF wrap respected).
   - Evidence is interpreted as claim-evidence-reasoning, never listed.
   - Every causal claim names a mechanism (driver -> mechanism -> effect -> consequence -> implication). 18 explicit driver-chains; 104 mechanism references.
   - Confidence stated with what would change it, calibrated to DESIGN-STAGE grade. 97 confidence statements.
   - Assurance-case summary (top goal G0, five subclaims S1-S5, assumptions A1-A5, context, residual risk) present and consistent.
   - Layer 9 architecture traceability correctly OUT OF SCOPE per research/expansion_plan.md (unit of analysis is a statistical cell, not a system/capability/data exchange). Not forced anywhere; the out-of-scope decision is stated explicitly in Chapters 1, 4, 7, 8, and the Assurance Case Summary.
   - No missing HAAF element found; nothing fabricated to fill a gap.

3. **SUPER-HUMANIZER (anti-slop / authentic voice).** Scanned for AI tells. Zero occurrences of "delve," "tapestry," "landscape," "realm," "moreover," "it is important to note," "in conclusion," "firstly/secondly," "furthermore." Sentence-length variation (burstiness) and named abstractions already strong. One vague intensifier adverb removed ("establishes, robustly, that..." -> "establishes, across independent datasets, that...", Ch. 5 problem frame), which also sharpens the empirical claim.

4. **MITRE-HUMANIZER (discipline only; neutral work).** Em dash purge: zero present, zero introduced. Banned-phrase sweep clean: no "cutting-edge," "leveraging" (the two "leverag" hits are inside real citation titles and a citation key, untouched), "seamless," "navigate the complexities," "testament to," "at the end of the day." "Robust" (34) retained where statistical ("robustness check," "robustness battery," "survivorship-robust," and the FIXED verbatim H0 hypothesis wording "largest and most robust association"). Null hypotheses, limitations, and threats-to-validity left stated plainly (zero-negative-words rule correctly NOT applied to academic prose).

5. **NOVEL-POLISH (academic).** Sentence-length distribution, lexical variety, and paragraph transitions held to the Kepler (precision) and Vox (argument clarity) bar. Openings and closings already strong; the abstract, introduction, and conclusion framings were sharpened (see Layer 7).

6. **EXECUTIVE-STORYTELLING (light, academic).** Abstract and introduction are stakes-first; the conclusion has a clear payoff arc across the three readings (H1 supported / H0 supported / unidentified). No process-label scaffolding ("this section will") present.

7. **KING CHARLES REGISTER (framing only).** Evidence step run: `query_style_exemplars.py --mode ceremonial --limit 6` (Coldstream Guards, Royal Marines, Gurkha Artillery, QEII centenary). Inferred a dignified, measured, stewardship-and-service register and applied it ONLY to:
   - Abstract, first sentence: now opens on the quiet weight and scant scrutiny of the heritage appeal.
   - Introduction Section 1.0, framing clause: the discipline "this work owes to the missions and the people who fly them."
   - Conclusion Section 8.5, closing sentences: "the obligation that the assurance of a mission owes to the people who depend on it is to settle the question on evidence... the service this dissertation is built to render."
   - Deck "The problem" lead bullet mirrored.
   The analytic body was left in its analytic register. GUARDRAILS honored: register only, never impersonation; no royal signature forms; no copied transcript phrasing; no claim of Royal Household approval.

8. **LITERARY-METHODS (deictic + epistemic).** Deictic Shift cold-reader audit: temporal references anchor to calendar context (design-stage status stated repeatedly; dates carried in references); "this/that/it" referents resolve unambiguously; entities are introduced before pronoun reference. Rhetoric of Science: epistemic modality calibrated to the design-stage evidence grade throughout. Expected results are stated in conditional/future terms ("the design would expect," "expected direction," "illustrative, not executed"), never as executed findings. No change required.

9. **HAAF RECHECK (closing logic pass).** After the prose, voice, and citation passes, re-confirmed against haaf_quality_gate.md that no HAAF layer broke: every Toulmin element retained (3x6 explicit headers intact, prose Toulmin intact), every qualifier and rebuttal survived (33 qualifiers, 21 rebuttals in prose plus the labeled blocks), causal mechanisms not flattened to correlations (driver-chains and "this is a mechanism, not a bare correlation" guards intact), confidence statements survived (97), and the architecture-out-of-scope decision still resolves to a policy-carried decision rather than a forced trace.


## Citation hygiene (final deliverable requirement)

- Unified every in-text citation to a single numbered scheme. Numbering follows reference-list order (1-116).
- Body citations rendered as clickable `[\[N\]](#ref-N)` (836 total).
- Reference-list entries anchored with pandoc-safe `<span id="ref-N"></span>[N] ...` (NOT `<a id>`).
- Every DOI is a clickable `https://doi.org/...` link (104 DOI links). The 12 URL-only sources (NTRS handles, arXiv, USU DigitalCommons, NASA HDL, AMOS, CNKI) already carry clickable URLs and have no DOI by nature.
- Verified: zero unresolved citation markers; every in-text number resolves to a real entry; anchor numbers are contiguous 1-116; every body-cite label equals its target; zero string-key markers remain.
- No citation was repointed to a different source. Only the style was normalized and the links added.


## Flagged for the author (no action taken; author's call)

1. **FIXED verbatim wording retained.** The H0 hypothesis text contains "largest and most robust association" (the word "robust"). This is carried verbatim from the FIXED bible in `research/expansion_plan.md` and appears identically in the Abstract, Section 1.4, and Section 8.1.4. It was deliberately NOT changed to preserve the locked hypothesis wording. If the author wants "robust" purged everywhere, the bible's hypothesis block must be edited first so all instances stay synchronized.

2. **URL-only references (12 of 116).** These resolve but carry no DOI: NTRS citation handles (`nasa_2018_nasa`, `radiation_2016_radiation`, `why_2014_why`), NASA HDL handles (`hamlin_2018_large`, `hirshorn_2016_final`), arXiv (`atz_2023_on`, `huang_2025_relative`), USU DigitalCommons (`langer_bouwmeester_2016`, `venturini_2018_improving`, `devaraj_2019_planet`), AMOS (`chandra_2018_leveraging`), CNKI (`yilin_2009_a`, grade C, non-English). All are flagged in Appendix F's source-provenance log. No DOI exists to add; left as clickable URLs.

3. **JPL-internal archival sources are uncitable.** The heritage-assessment matrices, parts-control records, and LLIS/PRACA outcome data carry no DOI and are documented by access path and provenance in Chapter 4 and Appendix F rather than as bibliographic entries. This is correctly handled as a data-access dependency, not a citation gap.

4. **Deck references slide.** The Marp deck (`deck.md`) lists anchor authors in prose ("Castet and Saleh...", "Angrist and Pischke...") and points to the dissertation's full IEEE bibliography. It carries no inline numbered cites, so the numbering transform did not touch it (correct for a defense deck). If a numbered bibliography is wanted on the deck, that is a separate request.


## Verification evidence

- `grep` for U+2014 / U+2013 / U+2015 across dissertation.md and deck.md: 0 / 0 / 0.
- Word count before vs after: 61,807 -> 61,807 (length preserved; character delta is solely from shortening citation keys to numbers).
- Citation integrity script: 836 numbered body cites, 116 contiguous anchors, 0 unresolved, 0 label/target mismatches, 0 string-key remnants, 104 clickable DOI links, 0 bare DOIs.
- HAAF element counts after all passes: Claim/Grounds/Warrant/Backing/Qualifier/Rebuttal headers 3 each; 97 confidence statements; 104 mechanism references; 18 driver-chains.