Restore the ability to build the thesis, and commit the corrected build

[salindne]

Neither Thesis/ nor ThesisPublished/ held the master document, so the thesis could not be rebuilt from either tree. The README recorded that fact; nobody drew the consequence — 104 .tex files across two trees had no automated check of any kind, which is how a malformed nested subscript and an unbalanced parenthesis have sat in Thesis/ unnoticed.

This adds Thesis/thesis.tex, reconstructed, and Thesis/thesis.pdf, what it builds.

Two results

The thesis builds. Zero errors, zero undefined references, zero undefined citations, 271 pages.

It works as a gate. Deleting a single \end{align*} fails loudly: Paragraph ended before \align* was complete.

The corrected thesis is now readable as a document

Thesis/thesis.pdf is not the PDF at the repository root, and the distinction is the point. The root PDF is as submitted in 2020. This one is the corrected thesis, so it is the only place the errata are visible rendered rather than as .tex source.

Verified rather than asserted: the post-C4 Degree-3 doubling row 74 3 89 19 is present and the superseded 73 3 101 19 is absent, so E-T10 really is in there.

How much is inferred: the preamble, all of it

The include order is exact, read off the committed thesis.aux of the original build — titlepage, frontmatter, chapters 1 to 7, no appendix.

The preamble was lost with the original file. About forty macro definitions were recovered from usage. The pattern that unlocked most of it: \be/\ee are example environments, not equation delimiters, matching \bd/\ed, \bl/\el, \bt/\et and \br/\er for definition, lemma, theorem and remark. That one correction took the error count from 247 to 47. subroutine then turned out to be a float rather than a theorem — it takes [htbp] and contains \caption — and \texorpdfstring revealed hyperref.

Evidence the inference is close:

	published	rebuilt
pages	267	271
extracted words	106,780	104,013
NUCOMP	268	265
Mumford	105	103
balancing	116	116

Two options settled by measurement, not taste

singlespaced: 268 pages at the time of testing against a published 267, where my initial doublespaced guess gave 368. The page-count comparison caught a wrong assumption, which is the whole reason for having it.

utopia: the only one of the class's five font options that builds, after carlisle supplied the missing scalefnt and dropping amssymb resolved a \circledS clash with mathdesign. It takes the page count from 276 to 271 and the word gap from 11% to 2.6% — which also explains the earlier discrepancy, since this thesis is mostly formulas and math extracts differently under different font encodings.

Three guesses, named rather than left to be discovered

All three are in Thesis/README.md:

1. \s is \quad, inferred from context alone, on about 1,200 sites. Nothing else in the reconstruction has that reach.
2. The title page fields the class demands — degree, graduating year, month, department — were in the lost preamble. The values are plausible, not sourced. Page one is where a wrong one shows fastest.
3. utopia is suggestive, not confirmed.

A hypothesis was also disproved on the way: the appendix is not the missing content. Including it adds one page and 114 words, because those three stubs really are empty.

What this unlocks

PR22 was planned around a gate that did not exist. Its stated acceptance test is "LaTeX builds", which was never runnable; the plan had grown a syntax checker as a substitute. With a working build that drops to a nice-to-have and PR22 gets the real thing.

PR31 becomes possible at all — you cannot fill three appendix stubs you cannot compile. The % \appendix line is left commented in thesis.tex with a note pointing there.

Requirements and one caveat

TinyTeX plus tlmgr install for memoir xpatch algorithms algorithmicx multirow grfext hyperref float url carlisle ly1 mathdesign and the Utopia fonts. No full TeX Live needed.

thesis.pdf is a committed build artifact and will go stale. Thesis/README.md says to rebuild it in the same commit as any .tex correction, and nothing enforces that. A candidate for a future gate rather than something this PR solves.

Gates

readme-paths 98/98 · check_paths --strict 259/259 · no .mag file touched, so the formula gates are not applicable · ThesisPublished/ untouched.