# SuperStack Final-Polish Report, JPL_AUTONOMY_EDL_03

**Artifact:** *Surface Mobility Productivity Across the Mars Rover Fleet: What Drives Drive-Distance and Sols-per-Meter?*
**Candidate:** JPL_AUTONOMY_EDL_03 (COLLEGIUM 1st Battalion, NORTH STAR / JPL category: Autonomous Systems and Robotics)
**Mode:** ACADEMIC. **Branding:** NEUTRAL (no MITRE markings, no vendor-AI names).
**Files revised in place:** `dissertation.md` (65,846 words, 8 chapters + 4 appendices + 128-entry reference list), `deck.md` (defense brief).
**Date:** 2026-06-16. Original preserved at `dissertation.md.bak`.


## Passes run (canonical stack_order)

The full nine-layer SuperStack was applied section by section using targeted edits, preserving every chapter, data point, citation, and the full length. Verification confirmed each layer landed and that the closing logic pass held.

1. **SPACE-SCIENTIFIC-WRITING**: domain rigor enforced upstream of prose. Every load-bearing claim traces to a real retrieved source in `research/corpus.jsonl` (128 references, 125 unique DOIs, all resolvable). No source invented; no DOI fabricated. Design-stage honesty preserved throughout: no executed estimates, result tables specified-but-unpopulated by design.
2. **HAAF LOGIC BAND (mandatory tier; Toulmin wrapped at Layer 4)**: the nine-layer flow holds across the dissertation: answer-first executive thesis (contribution as H0/H1, stated before any estimation); problem frame (current/desired/gap/consequence in 1.2.3 and the shared bible); major claims carry full Toulmin (claim, grounds, warrant, backing, qualifier, rebuttal, explicit in 1.9, 6.4, and each chapter spine); evidence rendered as claim-evidence-reasoning, not lists; every causal claim names a mechanism (driver to mechanism to effect to operational consequence to strategic implication); confidence stated with what would move it, calibrated to the DESIGN-STAGE grade; assurance-case summary present and consistent (top goal G, five subclaims, assumptions A1-A4, context, residual risk). **Layer 9 (architecture traceability) correctly OUT OF SCOPE** per `research/expansion_plan.md` A.7.e, not forced anywhere; the assurance-case table (Layer 7) is compiled because the artifact is mandatory tier.
3. **SUPER-HUMANIZER**: AI tells and scaffolding removed. Zero instances of "this section will / this chapter will discuss / as we will see / as mentioned earlier / needless to say" remain. No "delve," "tapestry," "realm" (decorative sense), "landscape" (decorative sense), or "moreover" overuse. Sentence length varied for burstiness; abstractions named.
4. **MITRE-HUMANIZER (discipline only; neutral non-MITRE work)**: zero em dashes (verified by grep, see below). Banned-phrase purge run; the only flagged tokens were technical terms of art that are load-bearing and were correctly retained (see "Banned phrases" below). Active voice preferred. The zero-negative-words rule was deliberately NOT applied: null hypothesis, limitations, and threats to validity are stated plainly (100 null/limitation/threat/falsification markers retained).
5. **NOVEL-POLISH (academic)**: sentence-length distribution, lexical variety, and paragraph transitions improved; openings and closings strengthened. Held to the Kepler (precision) and Vox (argument-clarity) bar.
6. **EXECUTIVE-STORYTELLING (light, academic)**: Abstract and Introduction made stakes-first (mobility as the binding constraint; the asymmetric policy error stated up front). Process-label scaffolding deleted. A clear payoff arc runs from problem to falsifiable contribution to decision relevance.
7. **KING CHARLES REGISTER (framing only)**: ceremonial evidence step run: `query_style_exemplars.py --mode ceremonial --limit 6` (Coldstream Guards / Royal Marines / Gurkha Artillery new-colours addresses). The dignified, measured, stewardship-and-service register was applied ONLY to the executive-facing framing: the first sentences of the Abstract ("A quarter-century of robotic service on the surface of Mars has been the work of many hands..."), the opening framing of the Introduction (1.1, "sustained across decades by the patient effort of the institutions and the people who design, build, and command these machines"), and the closing paragraph of the Conclusion (8.5, "The rovers themselves are the achievement of many who will never drive on Mars; the least that those who study their record can offer in return is the discipline to ask, with care and without flattery, what it was that carried them so far."). The technical body remains in analytic register. Guardrails honored: inspiration not impersonation, no royal signature forms, no copied transcript phrasing, no claim of Royal Household approval.
8. **LITERARY-METHODS**: Deictic Shift cold-reader audit: temporal references anchored (date on the title block; "in 1997," "across decades," "twenty-five years"); "this/that/it" referents disambiguated; entities introduced before pronoun reference. Rhetoric of Science: epistemic modality calibrated to the evidence grade, design-stage estimates stated in conditional/future terms ("is expected to be," "would estimate," "would raise confidence"), never as executed findings.
9. **HAAF RECHECK (closing logic pass)**: confirmed no HAAF layer broke after the prose, voice, and polish passes. Every major claim retains its warrant and qualifier; the causal mechanism was not flattened to a correlation by the humanizers (70 mechanism references intact, the driver-to-implication chain preserved in 1.10 and 6.4.1); confidence statements survived (78 calibrated-confidence statements); no architecture trace was introduced (Layer 9 remains explicitly out of scope). Reference: `haaf_quality_gate.md`.

**Citation hygiene (final step):** all in-text citations unified to a single numbered scheme. Each reference-list entry N anchored with a pandoc-safe `<span id="ref-N"></span>`; every body citation rendered `[\[N\]](#ref-N)`; every DOI rendered as a clickable `https://doi.org/...` link. No citation was repointed to a different source; only style normalized and links added.


## Em-dash count

| File | Before | After |
|---|---|---|
| dissertation.md | 0 | 0 |
| deck.md | 0 | 0 |
| chapters/*.md (9 files) | 0 | 0 |

The em-dash character (U+2014) does not appear in any deliverable. Em-dash discipline was enforced during drafting (house style in `expansion_plan.md` A.8), so no conversions were required at the polish stage; the final grep confirms zero remain.


## Banned phrases removed / adjudicated

The MITRE-humanizer purge scan flagged two tokens. Both classes are **technical terms of art** in design-based econometrics and the cited engineering literature, are load-bearing, and were correctly **retained** (the MITRE-humanizer strips non-load-bearing boosters, not domain terminology):

- **"robust" (15 occurrences):** all are "robustness battery / robustness check / robustness pass / robustness column," "cluster-robust standard errors" (a named estimator), "robustness against slip" (an engineering property), or appear inside published paper titles in the reference list ("Robustness, Optimization, and Applications"; "Robust and Efficient Stereo Feature Tracking"). No generic-booster use ("a robust framework / robust solution") was present.
- **"leverage" (3 occurrences):** all are the correct economic/statistical noun sense, "leverage on the cross-generation question," "outsized leverage," "a clean lower-leverage estimate." No banned verb "leveraging" was present.

No instances of robust, cutting-edge, leveraging (verb), seamless, navigate the complexities, testament to (in body prose), or at the end of the day required removal from the body. The deck was already clean of every banned phrase.


## Length and structural integrity (no truncation)

- Lines: 1,577 (revised) = 1,577 (original). Headers: 170 (revised) = 170 (original); zero added, zero dropped on either side.
- Characters: 471,763 (revised) vs 461,392 (original), the revised file is **10,371 characters larger** because the clickable citation links add markup.
- Words: 65,846 (revised) vs 65,883 (original). The 37-word `wc -w` delta is a tokenizer artifact: a bare year ("2003") counts as one token while the converted link `[\[39\]](#ref-crisp2003)` tokenizes differently. Normalized-prose length is also larger in the revised file, and all 8 chapters, 4 appendices, and 128 reference entries are present in both. No section, data point, or citation was lost.
- Reference list: 128 numbered entries, 128 `<span id="ref-...">` anchors, 125 clickable `https://doi.org/...` DOIs, zero bare DOI tokens.


## Citation resolution check

- 509 in-text citation links of the form `[\[N\]](#ref-key)`.
- 0 unresolved citation markers (every `#ref-key` link target resolves to a defined anchor).
- 0 leftover author-year or bracket-key strings in the body.
- 13 reference-list entries are carried in the corpus but not cited in the body prose (`a2023`, `anon2016`, `anonndg`, `chaisemartin2020b`, `li2002`, `li2011`, `mesalam2024`, `mishkin2002`, `nikitas2019`, `shen2003`, `sopegno2024`, `toupetnd`, `washington1999`). This is a complete-corpus reference list, not a defect; no in-text marker is unresolved.


## Items flagged for the author

1. **Uncited corpus entries (cosmetic, optional).** 13 of the 128 reference-list entries are not cited anywhere in the body (listed above). If COLLEGIUM house style requires a strict cited-only reference list, prune these 13 or weave a citation for each into the relevant chapter. Left as-is, they are harmless complete-corpus carryover. No fabrication was introduced to "use them up."
2. **Highest-value evidence gap (author already named it, restated for the committee).** The single binding limit on the design's confidence is the reconstruction of a clean per-drive autonomous-drive-fraction series across all generations (`expansion_plan.md` Part C item 3; carried in 4.2.3, 5.6.5, 6.3). It is correctly disclosed as a measurement-error caveat that attenuates the within-rover coefficient toward the null, so a null is read conservatively. Close this before execution.
3. **Date stamp.** The title block reads `Date: 2026-06-15`. Refresh if the defense submission date moves.


## Result

```json
{ "id": "JPL_AUTONOMY_EDL_03", "status": "ok", "em_dashes_remaining": 0 }
```