# P12 Reproducibility Record Date: 2026-07-23 Status: **PROVISIONAL AUTHOR MANUSCRIPT; RUNTIME EVIDENCE OBSTRUCTED** This record covers the reproducibility of the manuscript package for Paper 12, not the reproducibility of a semantic-sharding implementation. The repository contains internally accepted P2, P5, P10, and P11 manuscript contracts. It does not contain normalized executable interfaces or independent oracles for their query semantics, source coverage, epoch/migration, and consistency boundaries. The manuscript therefore reports an evaluation contract and falsifiers, not benchmark results. ## Evidence boundary | Question | Evidence in this package | Status | |---|---|---| | Does the manuscript encode the requested typed contracts? | `paper.tex`, Sections 3-9 | Yes | | Are the 12 proposed invariants preserved? | `paper.tex`, Table 5, IDs `SHARD-01` through `SHARD-12` | Yes | | Are the 18 counterexamples preserved? | `paper.tex`, Section 11, IDs `CE-01` through `CE-18` | Yes | | Are all 15 claim anchors preserved? | `paper.tex`, Table 6, IDs `P12-C01` through `P12-C15` | Yes | | Is primary prior art cited and closed? | `references.bib` plus the generated `.aux` and `.bbl` in the isolated build | Yes, 36 of 36 keys | | Is a P12 runtime implemented? | No executable P12 implementation is identified | OBSTRUCTED | | Has native pruning advantage been measured? | No result or benchmark is present | OBSTRUCTED | | Has distributed planning or cross-shard performance been measured? | No result or benchmark is present | OBSTRUCTED | | Has vector or full-text locality been measured? | No result or benchmark is present | OBSTRUCTED | ## Clean isolated build Run from `papers/12-semantic-sharding`: ```sh p12_build_dir="$(mktemp -d /tmp/catdb-p12-final.XXXXXX)" cp paper.tex references.bib "$p12_build_dir/" cd "$p12_build_dir" latexmk -pdf -interaction=nonstopmode -halt-on-error paper.tex latexmk -pdf -interaction=nonstopmode -halt-on-error paper.tex pdfinfo paper.pdf pdffonts paper.pdf ``` The second `latexmk` invocation is an explicit fixed-point check. It should report that all targets are up to date. The build output is intentionally temporary. Do not copy `paper.pdf`, auxiliary files, or a cover letter into the repository while P12 remains provisional. ## Citation-closure check The final build used 36 unique citation keys and a 36-entry bibliography. The following check compares the keys emitted to the `.aux` file with the keys in the BibTeX database: ```sh sed -n 's/^\\citation{\(.*\)}/\1/p' paper.aux \ | tr ',' '\n' \ | LC_ALL=C sort -u > cited.keys sed -E -n 's/^@[[:alnum:]]+\{([^,]+),/\1/p' references.bib \ | LC_ALL=C sort -u > bib.keys comm -23 cited.keys bib.keys comm -13 cited.keys bib.keys ``` Both `comm` commands must produce no output. The generated bibliography must contain 36 `\bibitem` records. The paper-local source ledger contains one row for every BibTeX key. During this author-side pass, all 30 DOI strings returned HTTP 302 DOI-resolver redirects. The USENIX, Microsoft, and NeurIPS no-DOI records returned HTTP 200. The two official VLDB 1986 PDF endpoints reset command-line connections in the earlier source pass, so terminal full-text availability is not claimed; the binding dossier retains their official proceedings provenance. ## Log and layout checks Run these checks in the isolated build directory: ```sh rg -n 'Overfull \\[hv]box|Underfull \\[hv]box' paper.log rg -n 'undefined references|Citation .* undefined|There were undefined' paper.log rg -n 'LaTeX Error|Emergency stop|Fatal error|duplicate destination|Warning' paper.log rg -n 'Warning|error|not found' paper.blg texcount -sum paper.tex ``` All `rg` checks above must return no matches. `texcount` is descriptive, not a semantic acceptance test. For visual inspection: ```sh p12_render_dir="$(mktemp -d /tmp/catdb-p12-render-final.XXXXXX)" pdftoppm -png -r 96 paper.pdf "$p12_render_dir/page" ``` Inspect every rendered page. Pay particular attention to the landscape dimension table, the distributed-join table, the invariant table, the claim ledger, the exact-nonclaims list, and the bibliography. ## Observed isolated build The round-one correction verification build used: - Build directory: `/tmp/catdb-p12-agy1-author-verify.xrPMMe` - Render directory: `/tmp/catdb-p12-agy1-author-render.UeDCky` - Engine: `pdfTeX-1.40.29` through `latexmk` - PDF version: 1.7 - Page count: 40 - Page size: US Letter, 612 by 792 points - File size: 375,051 bytes - Unique citation keys: 36 - BibTeX entries: 36 - Generated `\bibitem` records: 36 - Overfull boxes: 0 - Underfull boxes: 0 - Undefined citations: 0 - Undefined references: 0 - LaTeX errors or fatal stops: 0 - Duplicate destinations: 0 - Log warnings: 0 - BibTeX warnings or errors: 0 - Embedded fonts: all listed fonts embedded and subset - `texcount` words in text: 6,324 - `texcount` words in headers: 199 - `texcount` words outside text: 75 - `texcount` sum: 6,651 All 40 pages were inspected through five contact sheets covering pages 1-8, 9-16, 17-24, 25-32, and 33-40. Pages 7, 8, 10, 11, 21, 27, 29, 35, 36, and 40 were also inspected at original render resolution to verify the mapping scope, row-migration obligations, partition table, vector/full-text baselines, claim ledger, disputed claims, review provenance, and bibliography. No clipping, overlap, broken table continuation, missing glyph, or unreadable bibliography entry was found. The repository verifier independently accepted the temporary PDF against the canonical combined TeX/BibTeX source digest: ```text 518ed9ca8fbc51b2da15737d12261a4b1e85d254c5a2a327033ecc6eb68a0630 ``` Temporary build artifact hashes: ```text paper.pdf 5e251cd4c07fcd1b8c58b43216eab3d254eeb56801f80c824bfd32c0c1e27682 paper.log c893e697f037b55c9c90e21b7e98e964e2a74f6e6f1482462b9e572c8e554af8 ``` Repository source hashes at the verified build: ```text paper.tex 9228abaa59e75297c05934651f7ec9a454e2f66a976970eeb0e8e9be6bb861f4 references.bib 696eba92923e86880935d70caa9de1c45922ead9cfee6ee814fe7357fdbbab1f research/evidence-dossier.md 8ba87bd7d4e31dc2b4ae4873a81d545b446c0ff4dbec2f2eafbe7a9ef8ed9729 related-work-ledger.md 686d892b97e35c12dea7be3b1e981200e4a96fb6c86ba21023059d03924c9166 novelty-ledger.md 02a6758acd5a573b77da58c92a8e02b204ff97117a2c659a0036dd69b2702500 disputed-claim-ledger.md 5b6e7539657409b8ce27a14d1ffed066ed14f51ceeca2f404b441ad93b75ffec primary-source-bibliography.md 485e1533bbb28587edc7ca14bfa7fe13ddfdd3a77de82a542f348b34bcb846d6 review-contract.md c7cb2d591deca4bf5b026825bf54e0393b5713515db28fd7aa92ea6ca0756571 ``` The self-referential reproducibility and humanizer records are intentionally excluded from this source-hash block. ## Future semantic reproduction contract A future implementation may promote P12 claims only after the following evidence exists: 1. An executable query/completeness profile is derived from the accepted P2 contract and checked by an independent oracle. 2. An executable source-coverage, capability, and policy profile is derived from the accepted P5 contract. 3. An executable topology-epoch and migration-read profile is derived from the accepted P10 contract. 4. Executable consistency profiles for mutable references, indexes, and materializations are derived from the accepted P11 contract. 5. Every route decision emits a typed obligation and trace. 6. Every excluded partition has a machine-checkable disjointness, policy, or compatibility certificate. 7. Every selected partition participates in a completeness comparison against an independent oracle. 8. Join, exchange, reference, index, and materialization tests cover the corresponding `SHARD-*` invariants and `CE-*` counterexamples. 9. Baseline and treatment measurements are preregistered and report failures, timeouts, partial answers, metadata overhead, and provenance quality. Until those gates are met, the correct artifact status remains **PAPER REASONING with explicit assumptions; implementation OBSTRUCTED**. ## Terminal manuscript gate The corrected manuscript was reviewed at the exact combined source digest ```text 518ed9ca8fbc51b2da15737d12261a4b1e85d254c5a2a327033ecc6eb68a0630 ``` AGY round 2 rechecked all three round-one findings and ended in `VERDICT: ACCEPT`. The canonical review is byte-identical to round 2. The repository artifact hashes are: ```text paper.pdf 5e251cd4c07fcd1b8c58b43216eab3d254eeb56801f80c824bfd32c0c1e27682 paper.log c893e697f037b55c9c90e21b7e98e964e2a74f6e6f1482462b9e572c8e554af8 cover.png 238640fd344428efe706773733aa912a77cab065f15e4536eb0e08834e9703ae review e732d3fd4a0152a6c15059ab5d1a22e707f746d9243933b36b465545570d8d77 ``` The final verifier passed source and review digest identity, PDF freshness, 40-page minimum, LaTeX-log inspection, 300-DPI cover resolution, and terminal review verdict. This internal manuscript acceptance still does not establish executable P12 code, benchmark data, a publication submission, or any of the future semantic reproduction obligations above.