# Paper 5 Reproducibility Record

**Manuscript:** *Semantic Query Federation over Heterogeneous Databases*  
**Recorded:** 2026-07-23  
**Repository revision inspected:** `79b654fd456257865a1fe0e64a776807a012a277`  
**Result:** internally accepted paper artifact; NOT RUN for a CatDB federation runtime experiment.

## Evidence boundary

This paper is a provisional design and semantics paper. The repository inspection performed for the manuscript found no implemented Paper 5 query executor, federation planner, connector layer, source profile, or performance harness. The workspace manifests contain the existing core, IR, schema, mapping, migration, and CLI crates, while `fixtures/README.md` explicitly excludes query-rewrite fixtures and `lean/COVERAGE.md` excludes SQL, connectors, and planning.

Consequently:

- no backend execution result is reported;
- no latency, throughput, network, memory, pushdown, or planning-performance number is reported;
- the running PostgreSQL, SQLite, and JSON example is an executable target specification, not a completed run;
- exactness propositions are paper-level mathematical arguments, not implementation verification;
- status labels in the manuscript distinguish `PROVED`, `SUPPORTED`, `OPEN`, and `OBSTRUCTED`;
- partial answers and heterogeneous-source behavior remain specified policies whose implementations and empirical evaluation are future work.

## Authoring inputs

The manuscript was grounded in the paper-local dossier and brief plus the shared research ledgers, terminology, dependency graph, roadmap, and benchmark plan. The two paper-local source hashes are:

| Input | SHA-256 |
|---|---|
| `research/evidence-dossier.md` | `1c41bd96b6522db97db45515fcde86ec98123fec97631779b0ae8be6fa16e72d` |
| `research/manuscript-brief.md` | `9cb1a3633273ff6d1e4618b710a45b18fbed83865fda5ca1a36c90d1c2c3ba14` |

Primary literature and official specifications are recorded in the paper-local `references.bib`. Citation resolution is part of the clean build: the final BibTeX log contains zero warnings, and the final LaTeX log contains zero undefined citations.

## Clean isolated build

The final verification started in a newly created directory containing only `paper.tex` and `references.bib`.

```text
/tmp/catdb-p05-r1fix.VmD056
```

Equivalent reproduction commands from the repository root are:

```sh
build_dir=$(mktemp -d /tmp/catdb-p5-final.XXXXXX)
cp papers/05-query-federation/paper.tex \
   papers/05-query-federation/references.bib \
   "$build_dir"/
latexmk -cd -pdf -interaction=nonstopmode -halt-on-error \
   "$build_dir/paper.tex"
```

`latexmk` began without an auxiliary file, invoked pdfLaTeX, invoked BibTeX, and repeated pdfLaTeX until references, the table of contents, and the bibliography stabilized.

### Toolchain

| Tool | Version |
|---|---|
| latexmk | 4.88, 9 March 2026 |
| pdfTeX | 3.141592653-2.6-1.40.29, TeX Live 2026 |
| BibTeX | 0.99e, TeX Live 2026 |

### Final build outputs

| Artifact | SHA-256 |
|---|---|
| `paper.tex` | `10fe1f32757155cbdc01153657ea39d38e369032260562887a191af1a5286e67` |
| `references.bib` | `82ab8edffa899841a94b52b1cc4e098ebdefa58f5ab96959dc91e8877d49affc` |
| temporary `paper.pdf` | `f8bac148312836f870389482a3cc37b4dd6651403e3815080defbafbfe4137b5` |
| temporary `paper.log` | `a9ecfda99442b8a3af98695b242647bb544e37a6d951dbfaa220125479821e13` |
| temporary `paper.bbl` | `903eb6205f4c1bcd554e88dd395cb42c3d397f82f7f3b52856fc1850c9653cf5` |
| temporary `paper.blg` | `6c2d529ae179019ef9e52812e9eeab32797f61abf37724511f370139da3ef9e9` |
| combined `paper.tex` plus `references.bib` source digest | `707f2ad65b006c25be4095d2988fcdc509bf4e75b0c0d5ba7b3458b44fe4ef1f` |
| repository `paper.pdf` | `f8bac148312836f870389482a3cc37b4dd6651403e3815080defbafbfe4137b5` |
| repository 300-DPI `cover.png` | `547edd5d5ef4eb756e965417ae180cd5b0c12ef386125b0953a941e9d5a2f4c6` |
| canonical AGY review | `f30501f7711cf987e734c8ebb59f85d1b2f67c3683858846e10ff023e21a31bc` |

The PDF has 42 letter-sized pages, is PDF 1.7, and is 424,617 bytes in this run.

### Warning inventory

| Check | Count | Disposition |
|---|---:|---|
| TeX errors | 0 | pass |
| Undefined citations or references | 0 | pass |
| Multiply defined labels | 0 | pass |
| BibTeX warnings | 0 | pass |
| Overfull boxes | 0 | pass |
| Underfull boxes | 5 | accepted |

The five underfull boxes are narrow table cells or compact status-table entries. They occur at manuscript source lines 636 twice, 1426 once, 1858 once, and 1859 once. Representative rendered pages show no clipping or overlap. The log also reports the expected legacy-status notice from the `everypage` package used for the arXiv-style side label; it is non-fatal.

## Visual inspection

Representative pages from a final-source build were rasterized at 110 DPI and inspected:

- page 1: title block, vertical GrokRxiv identifier, and abstract;
- page 11: heterogeneous-source running example;
- page 13: candidate placement figure and observation-contract table;
- page 23: snapshot proposition and failure-policy table;
- page 34: limitations and open questions;
- page 42: final bibliography page.

No clipped text, figure collision, table overflow, malformed heading, or doubled status punctuation remained after the final correction.

## Current reproducibility claim

What is reproducible now is the manuscript artifact and its internal evidence classification. A reader can rebuild the same 42-page content from the two paper-local source files using the toolchain above. Binary PDF hashes can vary across builds because PDF metadata and trailer identifiers may encode build-time state; the source, bibliography, and generated `.bbl` hashes are the stable content checks in this record.

The repository now retains the accepted PDF, its inspected LaTeX log, and the
300-DPI first-page cover. No benchmark result or runtime trace was created.
AGY round 1 reviewed the TeX source digest `2c9c4b4a...` and requested two
minor semantic clarifications. Round 2 verified both corrections and found no
remaining blocker, major, or minor issue. Round 3 bound the unchanged paper
and bibliography to the repository verifier's combined source digest
`707f2ad65b006c25be4095d2988fcdc509bf4e75b0c0d5ba7b3458b44fe4ef1f`
and ended `VERDICT: ACCEPT`. The final repository verifier passed every
source, freshness, page-count, LaTeX-log, cover-DPI, review-digest, and
terminal-verdict check.

## Required future execution bundle

An empirical Paper 5 evaluation should not be considered reproducible until each run archives at least:

1. exact CatDB revision, dependency lockfile, compiler and OS versions;
2. connector revisions and source capability profiles;
3. canonical query IR, mapping revisions, selected sources, and physical plan;
4. PostgreSQL fixture snapshot, SQLite fixture snapshot, and immutable file digest;
5. request and response bytes, calls, rows scanned and returned, parse rejects, CPU, memory, and wall time by stage;
6. source snapshot identifiers, retry history, failure injection, and availability policy;
7. result rows with answer status, missing contributions, warnings, and provenance;
8. baseline definitions for central execution, pushdown-heavy execution, syntactic catalog federation, and single-source controls;
9. repeated runs, seeds, cold and warm-cache labeling, and uncertainty intervals;
10. an independently rerunnable checker for semantic equivalence and for every claimed optimization.

Until that bundle exists, the paper's execution plans, connector contracts, and evaluation tables are falsifiable specifications, not measured results.