# P14 reproducibility record

Status: internally accepted manuscript, `OBSTRUCTED`

Date: 23 July 2026

## Evidence boundary

This manuscript is a source-level research artifact. A reader can reproduce
its LaTeX build and audit its citations, definitions, evidence labels,
falsifiers, proposed evaluation, implementation order, and exact nonclaims.

The current boundary is:

| Layer | Present evidence | Not established |
|---|---|---|
| P14 evidence dossier | verified primary-source ledger, falsifiable cost contract, unit and exclusion tables, counterexamples, claim anchors, benchmark design, and implementation roadmap | optimizer, connector contract, calibration, plan-choice experiment, or scale result |
| Existing CatDB repository | prerequisite schema and mapping work owned by earlier papers | P14 candidate plan language, statistics snapshot, cost model, planner, adaptive runtime, vector adapter, or benchmark |
| P14 manuscript | 36-page provisional contract, paper reasoning, prior-art boundary, exact nonclaims, and reproduction requirements | implemented optimizer, executed benchmark, global optimality, or billion-row evidence |

Every CatDB-specific P14 execution, calibration, plan-quality, adaptive,
materialization, vector, and scale claim remains `OPEN`, `OBSTRUCTED`, or
`UNRESOLVED`. Primary literature supports only the established external
results cited in the manuscript.

## Clean isolated build

The build must run outside the repository so no PDF or auxiliary file appears
beside the source:

```sh
p14_build_dir="$(mktemp -d /tmp/catdb-p14-author.XXXXXX)"
cp paper.tex references.bib "$p14_build_dir/"
cd "$p14_build_dir"
latexmk -pdf -interaction=nonstopmode -halt-on-error paper.tex
pdfinfo paper.pdf
```

Equivalent explicit passes are:

```sh
pdflatex -interaction=nonstopmode -halt-on-error paper.tex
bibtex paper
pdflatex -interaction=nonstopmode -halt-on-error paper.tex
pdflatex -interaction=nonstopmode -halt-on-error paper.tex
```

Acceptance checks:

```sh
rg -n \
  "Overfull|undefined references|undefined citations|LaTeX Error|Emergency stop|Fatal error" \
  "$p14_build_dir/paper.log"
rg -n "Warning|error|not found" "$p14_build_dir/paper.blg"
texcount -inc -sum "$p14_build_dir/paper.tex"
```

The render pass also remains outside the repository:

```sh
p14_render_dir="$(mktemp -d /tmp/catdb-p14-render.XXXXXX)"
pdftoppm -png -r 110 "$p14_build_dir/paper.pdf" "$p14_render_dir/page"
```

## Final author-pass observation

The corrected round-1 response build for the source hashes below used:

- build directory: `/tmp/catdb-p14-r2.RAfNdA`;
- render directory: `/tmp/catdb-p14-r2-visual.quj8Ip`;
- format: 12pt article, US letter, PDF 1.7;
- page count: 36;
- PDF size: 357,123 bytes;
- TeXcount text words: 7,581;
- TeXcount summed words: 7,899;
- bibliography entries emitted: 34;
- overfull boxes: 0;
- undefined references: 0;
- undefined citations: 0;
- BibTeX warnings: 0;
- LaTeX errors: 0; and
- underfull boxes: 9.

The underfull boxes occur in two long bibliography entries with long primary
source URLs. They do not clip, overlap, or hide text.

All 36 pages were rendered and inspected in four nine-page contact grids.
Pages 1, 10, 13, 15, 25, 26, 31, 33, and 36 were also inspected individually
at rendered resolution. These checks cover the title and GrokRxiv sidebar,
table of contents, dense equations, unit and roadmap tables, decision
certificate, appendix listings, and every bibliography page. No heading,
paragraph, equation, table, listing, citation, URL, page number, or sidebar is
clipped or overlapped. No page contains a broken glyph or unreadable element.

Temporary build hashes:

- `paper.pdf` SHA-256:
  `a94cb96bc61dfad0c7be1f636b2ceda8e2bec2362374b4e2f9d1bbaadb0956ee`;
- `paper.log` SHA-256:
  `47d7a648607cf737911833f052a9c3cfd5c9baf057a98bed5cefa6ad6df7479c`.

Repository source hashes:

- `paper.tex` SHA-256:
  `b8b84498a21352d8b6a386aea07b36e375aa048c246dcde2334db815b166612e`;
- `references.bib` SHA-256:
  `f8cf8fca55b883a2c1320aa1b6a791638be69095f8acddc511d8ed071d80afd8`;
- `research/evidence-dossier.md` SHA-256:
  `22886746a3bf97e2aa7f9b99273e2866c911140582276aa979e563752b8392ef`.

The paper directory contains no generated PDF, cover image, log, table of
contents, BibTeX output, or other auxiliary build artifact.

## Citation closure

The final BibTeX pass emits all 34 bibliography entries. The manuscript has no
missing or undefined citation.

The source set covers:

- System R, Volcano, Cascades, Calcite, and end-to-end optimizer evaluation;
- R*, Garlic, SemaGrow, Trino, and federated SPARQL evaluation;
- Eddies, ANAPSID, progressive optimization, LEO, Bao, and learned
  cardinality estimation;
- multiple-query optimization, semantic caching, view-based rewrites,
  dynamic materialization, and incremental maintenance;
- continuous consistency and probabilistically bounded staleness;
- relational and coarse-grained provenance; and
- product quantization, HNSW, GPU search, DiskANN, Faiss, and Milvus.

The evidence dossier independently checked DOI metadata and live primary
records on 23 July 2026. Bibliographic identity does not promote a CatDB
implementation claim.

## Future semantic reproduction contract

A future P14 result bundle must retain:

1. exact repository revision and dirty-state record;
2. frozen request, plan, capability, objective, policy, and result-envelope
   profiles;
3. immutable logical and physical inputs with content digests;
4. source and connector revisions, source frontiers, and environment records;
5. candidate generation, feasibility, pruning, and decision traces;
6. typed statistics with units, scope, method, uncertainty, expiry, and
   semantic revision;
7. predicted outcome vectors and frozen model revisions;
8. raw operator, transfer, queue, source-load, materialization, provenance,
   and vector-quality observations;
9. independent normalized result-envelope expectations;
10. all failures, timeouts, retries, cancellations, and censored observations;
11. cache and materialization compatibility evidence;
12. exact vector ground truth or a declared reference;
13. randomized-block assignments, seeds, warmup, repetitions, exclusions,
    stopping rules, and analysis code;
14. physical-only, capability-aware, semantic-feature, uncertainty, and
    adaptive ablations;
15. decision certificates and a replayable explanation for every chosen,
    probed, rejected, infeasible, fallback, or abstained case;
16. raw artifact checksums; and
17. retained null and negative results.

Reading expected fixture output as the planner result is not an independent
oracle. Equal visible rows without equal required provenance, coverage,
freshness, consistency, and approximation are not an exact result-envelope
match.

## Paper-review response

AGY round 1 accepted the original candidate and identified two nonblocking
hardening points. The corrected source now:

- treats declared rate limits, concurrency-pool limits, quotas, and saturation
  thresholds as hard feasibility constraints rather than soft source-pressure
  preferences; and
- requires filtered-vector ground truth and exact fallback to apply the
  request's identical metadata predicate before exact distance calculation.

Because those corrections changed `paper.tex`, the original acceptance digest
is retained only in the round-1 review artifact. Round 2 reviewed the corrected
combined source digest, verified both changes, found no remaining actionable
major or minor issue, and ended in `VERDICT: ACCEPT`. The canonical review is
byte-identical to round 2.

## Accepted repository artifacts

| Artifact | SHA-256 |
|---|---|
| combined `paper.tex` plus `references.bib` source | `477c284657b5836526cbd307629771f1de44602b379d028c52f91f94fda5f193` |
| repository `paper.pdf` | `a94cb96bc61dfad0c7be1f636b2ceda8e2bec2362374b4e2f9d1bbaadb0956ee` |
| repository `paper.log` | `47d7a648607cf737911833f052a9c3cfd5c9baf057a98bed5cefa6ad6df7479c` |
| repository `cover.png` | `0290121463aebdb23b2218a2531c269eccb4724f1557ccd2a7367ed39a36a798` |
| canonical AGY review | `73b722f7e3d7c7f1f7c3c2e78cde4eb7a2bc2f6433d5d1586c407f56b06359fa` |

The repository verifier confirms source-digest identity, PDF freshness, 36
pages, no rejected log finding, cover freshness, 300-by-300 DPI cover metadata,
the exact review source digest, and terminal `VERDICT: ACCEPT`.

## Deliberate absences

This author pass did not:

- implement a P14 optimizer, cost model, statistics layer, connector, cache,
  materialization policy, adaptive runtime, vector adapter, or benchmark;
- run Rust, Haskell, Lean, SQLite, PostgreSQL, HTTP, vector, distributed, or
  user experiments;
- claim global optimality or billion-row scale;
- create a repository PDF, cover image, log, or auxiliary file;
- edit README, CITATION, shared coordination files, code, Slack, or another
  paper;
- commit or publish changes; or
- turn literature results into CatDB execution evidence.

No optimizer, runtime, connector, benchmark, performance result, global
optimality result, or billion-row evidence is created by manuscript
acceptance.