--- name: stoc-supplementary description: Use when architecting everything beyond the first twelve pages of a STOC (ACM Symposium on Theory of Computing) submission — the discretionary-read appendix, the reviewed table of contents, theorem-to-proof pointer discipline, and the division of labor between the submission appendix and the arXiv/ECCC full version. --- # STOC Supplementary STOC has no supplementary-upload channel, no ZIP, and no page cap. One PDF carries everything, and the 2026 CFP draws the line inside that PDF: the abstract, the **table of contents**, and the first twelve pages are read; material beyond that is read at the committee's discretion (checked 2026-07-08). So "supplementary material" at STOC means *the discretionary region of your own submission* — and designing it is a first-class writing task, because reviewers enter it selectively, through the pointers you leave. ## The table of contents is part of the supplement's interface Uniquely among major venues, STOC names the ToC as guaranteed-read material. Treat it as the map reviewers use to decide *which* discretionary pages to open: ```latex \tableofcontents % after the abstract, before Section 1 % Section titles should narrate the argument, not label boilerplate: % 4 Reducing multi-commodity routing to tree embeddings % 5 The potential-function analysis (proof of Theorem 1.2) % A Deferred proofs from Section 4 % B The lower-bound construction (proof of Theorem 1.4) % C Extension to weighted instances ``` A ToC whose appendix entries name the theorem each section proves lets a reviewer verify exactly the claim they doubt. "Appendix A: Omitted proofs" is a map with no street names. ## Placement decision table | Content | First 12 pages | Discretionary appendix | Full version only | |---|---|---|---| | Every theorem/corollary statement | Required | Restated with proof | Mirror | | Technical overview of the main proof | Required | — | Mirror | | Complete proof of the main theorem | Idea + hardest step | Yes, in full | Yes | | Routine lemma proofs | No | Yes | Yes | | Secondary results and extensions | One-sentence pointers | Statements + proofs | Yes | | Long definitions/preliminaries variants | Compressed | Full generality | Full generality | | Illustration figures, search procedures | No | Brief procedure note | Complete detail | | Anything the paper's acceptance depends on | Must be at least *stated* here | Cannot live only here | Cannot live only here | The invariant behind every row: a committee member who stops at page twelve must still be able to state your contribution correctly and believe it is probably true; a reviewer who continues must find every belief redeemable. ## Pointer discipline Deferred proofs are only as good as their forward and backward links: - At each in-body statement: `\emph{Proof in Appendix~\ref{app:lb}.}` — one pointer, exact target, never "proofs appear in the appendix" globally. - At each appendix proof: restate the theorem before proving it (the `thmtools`/`\restatable` mechanism keeps restatements literally identical), so the discretionary reader does not page-flip to reconstruct the claim. - Order appendix sections by the body's order of claims, not by proof dependency; a reviewer chasing Theorem 1.4 should not wade through the machinery of Theorem 1.2 to find it. - Give the appendix its own two-or-three-sentence opening paragraph stating dependency structure ("Appendix B is self-contained; Appendix C uses Lemma A.3"). ## What the appendix is not for - Not an overflow valve for a bloated body. Twelve pages that read like a compressed twenty signal that the writing pass never happened; cutting secondary material *entirely* (into the full version) usually beats squeezing everything in. - Not a second paper. Extensions that would each need their own referee ("we also resolve the directed case") dilute the committee's confidence budget; state one, defer the rest to the full version, or split the work. - Not a hiding place. Burying a known limitation on page 47 while the body implies generality is the kind of asymmetry that turns a positive review negative when discovered — and assigned theory reviewers do discover it. ## Appendix smells and their repairs - **The monolith:** one 30-page "Appendix A: Proofs" with no internal structure. Repair: one appendix section per body section or per major theorem, each named for what it proves. - **The dependency maze:** Appendix C's lemma silently requires Appendix B's notation established twenty pages earlier. Repair: per-section dependency sentences and local notation reminders at first reuse. - **The duplicate drift:** the appendix restatement of Theorem 1.2 differs by one quantifier from the body statement. Repair: `\restatable` or shared statement files — never retype (`stoc-reproducibility`). - **The dangling pointer:** the body promises "Appendix D" and the appendix ends at C. Repair: a pre-upload pass compiling with `\refcheck` or grepping every `Appendix~\ref` target. - **The kitchen sink:** half-finished generalizations kept "for completeness." Repair: cut to the full version; the submission appendix is a verification surface, not a notebook. ## Relationship to the full version The submission appendix and the eventual arXiv/ECCC full version usually share sources but serve different masters: the appendix is optimized for *selective verification during review*; the full version for *linear reading forever*. Before camera-ready, the full version typically re-expands what the appendix compressed, restores the general statements, and becomes the canonical document (`stoc-camera-ready` and `stoc-reproducibility` handle that transition and its drift risks). During review, though, the single submitted PDF must stand alone: STOC 2026 was double-blind, and a reviewer following an external link to your named preprint is an anonymity event you caused. ## Cycle-volatility warnings - The guaranteed-read wording (abstract + ToC + 12 pages) is 2026 CFP text; earlier cycles used a 10-page window, so re-read the live CFP each fall before budgeting pages (待核实 per cycle). - Font/margin specifics (11-point, 1-inch, letter) are also restated annually. ## Output format ```text [Appendix architecture] navigable / needs restructuring [ToC audit] narrates argument / boilerplate labels <- rewrite [Placement violations] [Pointer check] statement->proof links complete? restatements identical? [Deferral plan] ```