# Reproducibility record

Status: provisional authoring artifact

Evidence cut: 23 July 2026

This record reproduces the synthesis manuscript as an evidence-bounded paper,
not as a CatDB runtime. The committed repository base is
`e2f0468cea7b86a85397c5bcc269d66333a435f0`. The synthesis files and live
Slice 5 work were uncommitted at the evidence cut, so the table below pins
those inputs by path and SHA-256 rather than treating the commit as a complete
source snapshot.

No repository PDF, cover, independent paper review, coordination edit, commit,
deployment, publication, or Slack notification was created by this authoring
pass.

## Evidence classes and boundary

The current verification record marks P1 through P18 as internally accepted
paper artifacts. This synthesis is unreviewed and provisional. Acceptance of a
paper artifact establishes its reviewed model and argument only. It does not
establish a runtime, backend, connector, optimizer, distributed system, user
interface, or benchmark.

The only executable result admitted by this synthesis is the accepted Slice 4
record. That result is a closed fixture-corpus comparison, not a performance,
scale, federation, materialization, distributed, provenance, identity,
workspace, optimizer, connector, or MDReader experiment.

Live Slice 5 work is excluded. There is no Slice 5 release record and no
terminal accepted Slice 5 code review. An untracked author-side candidate
record appeared during this pass and expressly states that follow-up review and
release acceptance have not run. The accepted P16 and P18 manuscripts record
author-audit snapshots over a 42-case corpus. In those snapshots, a Slice 4
selection produced 29 semantic and 13 structural comparisons, while P16
separately classified all 42 cases as semantic under the unreviewed Slice 5
candidate. Those counts describe pinned author-audit snapshots. They are not
accepted Slice 5 evidence. The live Slice 5 tree continued changing during
this pass, so no current live Slice 5 count is reported here or in the paper.

## Manuscript inputs

All hashes are SHA-256.

| Input | Status at evidence cut | SHA-256 |
|---|---|---|
| `papers/01-semantic-model/paper.tex` | internally accepted | `95ccf45afad07d5c461b5e1063c7d6d528f8c7e6ef290b8124086b3b858728e1` |
| `papers/02-physical-compilation/paper.tex` | internally accepted | `81646e98d79db87c34399e821a75b508cf99072f062d2cdae53dbd84bfd8901a` |
| `papers/03-schema-migration/paper.tex` | internally accepted | `9b8f6d37e6d152dabd3b624d07ff840bae23fa192c2e227b4a9e06d9f6ac57b5` |
| `papers/04-composable-integration/paper.tex` | internally accepted | `83164ab3dda34629a77e83d3242300f3e0c2dcbf4451fb03696fe437774f8e44` |
| `papers/05-query-federation/paper.tex` | internally accepted | `10fe1f32757155cbdc01153657ea39d38e369032260562887a191af1a5286e67` |
| `papers/06-incremental-materialization/paper.tex` | internally accepted | `c688a5ab4ba5e57a7ed90d66dd3218a6edf612973be8543f9406eb128f59b008` |
| `papers/07-identity-reconciliation/paper.tex` | internally accepted | `9c1c53c9a080813ae9579c5af084258a660559af2278684f71c2aca638175174` |
| `papers/08-bidirectional-mappings/paper.tex` | internally accepted | `6e511e130b32fcbccdc1c89f691f73a003b5d7cdf08e6165a61c191c345cd696` |
| `papers/09-versioned-workspaces/paper.tex` | internally accepted | `78afb7850fabc5f1373c0f62affb6cc939f545c756b6b0622616b0ac7111da03` |
| `papers/10-distributed-catdb/paper.tex` | internally accepted | `08112d7f3fc3820f269292ac8ebbcab13b52c0caf1fd34a12ea9f6a5b3806172` |
| `papers/11-consistency-models/paper.tex` | internally accepted | `8b077b8e377c63447469cd51668a7802fa44e64bffcbea2ca62832f55b8773fa` |
| `papers/12-semantic-sharding/paper.tex` | internally accepted | `9228abaa59e75297c05934651f7ec9a454e2f66a976970eeb0e8e9be6bb861f4` |
| `papers/13-provenance/paper.tex` | internally accepted | `7bda8be17691af6c13218f885434b884f028ba37784672d96b1280656db7e136` |
| `papers/14-cost-based-planning/paper.tex` | internally accepted | `b8b84498a21352d8b6a386aea07b36e375aa048c246dcde2334db815b166612e` |
| `papers/15-rust-runtime/paper.tex` | internally accepted | `5cfd1902dbf26a9e152e2e7cec448f026634a81983343ec09b7db01ef2f440d4` |
| `papers/16-haskell-semantics/paper.tex` | internally accepted | `2d92431b7be7444ca2f3ab23257aa7a7d25f91c4c548e6b477430fcd82c6db23` |
| `papers/17-lean-formalization/paper.tex` | internally accepted | `5f7f5684fc153bda6b8a7d876a9552341fca86516f03fe2c68566759032a122c` |
| `papers/18-mdreader/paper.tex` | internally accepted | `6031eaad35a3583f1ddf5613323bf2046ebc83a7a25c95167ee6a5d973d90f86` |

The synthesis outline used in this pass has SHA-256
`52a368e529f749bdd9a2a59b34d8ebc8598fa3cf65f93d1e1887a6de150adad0`.

## Program-level inputs

| Input | SHA-256 |
|---|---|
| `research/paper-manifest.md` | `bc6cdb5fa6ca411181db8d4a42ecdba47d4542b2615052aaa6c1802f4eb8417a` |
| `research/paper-dependency-graph.md` | `c9a26ba393f322ab7a30824dd4c650f34d9e7fbbde37ce9fdd0c4db5ff806c87` |
| `coordination/verification-status.md` | `f5c5f0b5ba287c3aaaed6364a100bbd12aab10926526f780febe53298c300651` |
| `coordination/releases/slice-4.md` | `afe1f08544ffb80b9dadd017a721371292599b863a5e06d92a231f2c90599254` |
| `reviews/slice-4-claude-code-review.md` | `69664f8f17cbac5a84b36e819c942bb2a26aa10b8ff00c8cd94914543c6289af` |

The manifest dependency graph contains 19 nodes and 70 edges and is acyclic.
The longest recorded blocking path is
`P1 -> P3 -> P4 -> P13 -> P7 -> P9 -> P10 -> P11 -> P12 -> P14 -> S`.
The runtime and application path is
`P1 -> P3 -> P4 -> P13 -> P15 -> P18 -> S`, and the formal path is
`P1 -> P3 -> P4 -> P16 -> P17 -> S`.

## Accepted executable evidence

The accepted Slice 4 release and terminal review record:

```text
fixture cases=33
slice-4 semantic-compared=29 structural-compared=4
slice-4 disagreements=0
```

The accepted release can be checked from the repository root with:

```sh
python3 scripts/validate-fixtures.py \
  --schema fixtures/schema-v1.schema.json \
  --cases fixtures/cases

python3 scripts/audit-fixture-manifest.py fixtures/manifest-v1.json

cargo run -q -p catdb-cli -- fixtures compare fixtures/cases \
  --left rust --right haskell --capability-set slice-4
```

These commands were not rerun during this authoring pass because the shared
working tree contained active, unreviewed Slice 5 changes. The synthesis uses
the accepted Slice 4 release and terminal review records, not a result from the
changing tree. The accepted result means only that the released Rust and
independently implemented Haskell programs agreed on 29 admitted semantic
cases in the closed 33-case corpus. Four cases remained structural-only. It is
not a theorem, runtime-conformance result, performance result, distributed
result, or scale result.

## Isolated manuscript build

The final synthesis source hashes are:

```text
c4ad37fbb17f49e69a61d7302b7cb83a3b3ce9f8745d4e087a029e0858f810fd  paper.tex
056fe8d8f8cd4ae9a9bd466e1cb2490a14017b1daa96b39d5d1bcb739676e95a  references.bib
```

A clean build can be made without writing generated files into the repository:

```sh
build_dir=$(mktemp -d /tmp/catdb-synthesis-build.XXXXXX)
cp papers/synthesis/paper.tex papers/synthesis/references.bib "$build_dir/"
(cd "$build_dir" && latexmk -pdf -interaction=nonstopmode \
  -halt-on-error paper.tex)
```

The inspected build was created in
`/tmp/catdb-synthesis-author-audit-v2.L4QuM9`. Exact output hashes:

```text
ea9ee10b9415bdec9008018a0e068a057a8dc3f95d42eabe60a7061593826c6b  paper.pdf
f6745221309cf90cf3884db49cf16ac5c6877015890d7359466cdb2d18df1076  paper.log
```

`pdfinfo` reported PDF 1.7, 38 letter-size pages, 483278 bytes, producer
`pdfTeX-1.40.29`. The workspace `paper.tex` and `references.bib` were
byte-identical to the isolated copies after the final build.

Build audits:

- TeX errors: 0
- TeX warnings: 0
- BibTeX warnings or errors: 0
- undefined references or citations: 0
- duplicate destinations: 0
- overfull boxes: 0
- underfull boxes: 0
- unique citation keys: 61
- unique BibTeX keys: 61
- generated bibliography items: 61
- citation/BibTeX symmetric difference: 0
- embedded fonts: all listed fonts embedded, subset, and Unicode-capable

## Visual inspection

The PDF was rendered with:

```sh
render_dir=$(mktemp -d /tmp/catdb-synthesis-render.XXXXXX)
pdftoppm -png -r 120 \
  /tmp/catdb-synthesis-author-audit-v2.L4QuM9/paper.pdf \
  "$render_dir/page"
```

The inspected render is
`/tmp/catdb-synthesis-author-render-v2.VCZMOw`. All 38 pages were inspected.
Contact sheets covered the entire document, pages 5 and 6 were also checked
individually after the final status edit, and landscape pages 21, 22, and 30
were checked in their native orientation. The title block, status notice,
contents, equations, dependency graph, code-result panel, portrait tables,
landscape tables, appendices, URLs, and bibliography were legible and
unclipped. No overlap, truncation, blank-page defect, or malformed glyph was
observed.

## Reopening and promotion conditions

Any source, evidence, status, or implementation change after this evidence cut
reopens the manuscript. Promotion requires:

1. continued exclusion of Slice 5, unless it gains a release record and a
   terminal accepted code review;
2. a current pinned evidence inventory and reproduction closure;
3. all five synthesis questions and claim IDs rechecked;
4. complete citation closure and a prose-only Humanizer audit;
5. a clean isolated build and full-page inspection; and
6. a separately authorized independent synthesis review on the exact source.

None of these requirements permits an accepted paper model to stand in for
runtime evidence.

## Terminal manuscript gate

The canonical CatDB verifier digest for `paper.tex` plus `references.bib` is:

```text
926a1d649ed8dc226421e9d61b8af8f830dd8c6d48d01933fed2a46971ebab98
```

AGY round 1 reviewed that exact digest and ended in `VERDICT: ACCEPT`; the
canonical review is byte-identical to round 1. The repository artifact hashes
are:

```text
paper.pdf  ea9ee10b9415bdec9008018a0e068a057a8dc3f95d42eabe60a7061593826c6b
paper.log  f6745221309cf90cf3884db49cf16ac5c6877015890d7359466cdb2d18df1076
cover.png  91d5339dda79469fb2abc7e6b12175b0cc87b2e72f0ab2810a43ee1cfb801777
review     1cabe2ab4951f3236a43f415ef9680300c5b5e98420a1df51ba1d77339ebb9c7
```

The final verifier passed source and review digest identity, PDF freshness,
38-page minimum, LaTeX-log inspection, 300-DPI cover resolution, and terminal
review verdict. This internal acceptance is tied to the frozen evidence cut at
`e2f0468`; it synthesizes all eighteen accepted paper artifacts while keeping
unreleased Slice 5 work outside the positive evidence set.