Reproducing the drafts

The scripts are mostly auxiliary checks. P11 uses a rigorous finite modular reduction, so its full coefficient certificates are a claimed part of its proof. Reproducing a computation does not certify the general proof or novelty.

Working directory and outputs

For each table row, first change into the named directory beneath papers/, then run the commands in the stated order. For example, from the workspace root:

cd papers/P03-local-permutation-polynomials
python3 check.py

Equivalently, from the workspace root, prefix each script with its directory, for example python3 papers/P03-local-permutation-polynomials/check.py. Local module imports resolve beside their scripts; generated files use the script's location, rather than the shell's working directory. Use a normal Python invocation: python3 -O disables the assertions that perform these checks.

Scripts listed as writing JSON overwrite their named result files. Save a copy first if comparing against the recorded results. Compare parsed JSON, rather than whitespace; ignore diagnostic elapsed-time fields in P11. Scripts listed as stdout-only do not update an adjacent JSON file. For those, capture stdout to a new file if desired, for example python3 check.py > reproduced-results.json in P03, and compare its parsed JSON with check-results.json.

Python and package requirements

A Python 3.10-or-newer environment can run the complete collection. P11's pinned python-flint==0.9.0 explicitly requires Python >=3.10. The other supplied scripts are compatible with Python 3.9, including P12's use of math.lcm.
P01, P07, and P12 require SymPy. The inspected installation has SymPy 1.14.0.
P05's normalization-correction check requires NumPy. The inspected installation has NumPy 2.0.2. That check uses floating-point LP-vertex calculations with explicit half-integrality tolerance checks, together with exact matching calculations; it is auxiliary evidence, not the hand proof of normalization.
P11's two full coefficient passes require python-flint 0.9.0, pinned in P11-partition-congruences/requirements.txt. Its local certificates.py module and input JSON files are included. Install the package in the selected environment; a copied vendor/ directory is not required. Both independent expressions use the same FLINT library, a limitation recorded in the mathematical review.
All other published draft check scripts use only Python's standard library. P03 includes both field_arithmetic.py and transfer_experiment.py. P10's exploratory SciPy/NumPy optimization scripts are excluded from the published draft folder; they are not proof dependencies.

One possible clean environment setup, using an already selected Python >=3.10 interpreter, is:

python3 -m venv .venv-reproduce
. .venv-reproduce/bin/activate
python -m pip install sympy==1.14.0 numpy==2.0.2 python-flint==0.9.0

Run this from the workspace root before entering individual paper directories. In the activated environment, python3 in the table refers to that environment's interpreter. The environment setup is provided for reproduction; no new environment or dependency installation was performed during the packaging repair.

Per-paper commands

Commands are relative to the paper directory in the first column. A successful run exits with status zero; failed assertions or missing imports exit nonzero.

Paper directory	Commands, in order	Dependencies	Output and reference results
`P01-lefschetz-classification`	`python3 verify.py`	SymPy	Writes `verification.json` and prints the same JSON. Deterministic random seed 20260908 is recorded. These are finite graph/witness and exact-rank checks.
`P02-recurrence-repair`	`python3 check.py`	Standard library	Stdout-only `PASS ... local Lucas congruences and minimality witnesses`; no output JSON is written. This script checks the Lucas application, not every theorem in the paper. Broader independent review evidence is linked in `README.md`.
`P03-local-permutation-polynomials`	`python3 check.py`	Standard library; included local modules `field_arithmetic.py`, `transfer_experiment.py`	Prints JSON; compare with `check-results.json`. Does not overwrite that file. Independent function-value and polynomial-transfer computations are compared.
`P04-carlitz-factorization`	`python3 check_small_cases.py`	Standard library	Prints a four-case JSON list; no result file is written. Compare the numerical counts with `validation.md`. The computations use prime fields q=2,3; the proof covers all finite fields.
`P05-symbolic-defect`	`python3 verify.py`<br>`python3 verify_correction.py`	First script: standard library. Second: NumPy.	Writes `verification.json` and `correction-verification.json`, respectively, and prints summaries. The second script documents its floating-point tolerance and exact matching comparison.
`P06-stirling-repair`	`python3 check_digits.py`	Standard library	Stdout-only `PASS: ... direct Stirling-triangle digit congruences`; no file is written. This is the digit-formula check; other independently reviewed local congruence tests are identified in `README.md`.
`P07-irreducible-compositions`	`python3 check_prefix.py`	SymPy	Writes `prefix_checks.json` and prints the same JSON. Checks selected compositions, factorizations, and integer gaps. `proof.md` is also included; the arbitrary-continuation proof is not inferred from these samples.
`P08-homological-shift-counterexamples/checks`	`python3 check.py`	Standard library	Writes `results.json` and prints a compact summary. The script is auxiliary; artifact hashes are in the collection manifest.
`P09-unique-addition`	No computational script is required or supplied.	None for the hand proof.	Read `manuscript.tex`, `dossier.md`, and the review linked in `README.md`. The absence of a check script is intentional.
`P10-restricted-sumsets`	`python3 check.py`	Standard library	Writes `check-results.json` and prints the same report for 1,848 parameter triples. `counterexample.json` preserves the original translated finite witness. Optimization discovery scripts are not included or needed.
`P11-partition-congruences`	`python3 certificates.py`<br>`python3 check_congruences.py`<br>`python3 check_initial.py`<br>`python3 check_independent.py --full`	Python >=3.10; python-flint 0.9.0 for `check_congruences.py` and the independent `--full` pass. Other steps use the standard library.	Writes `certificates.json`, `check-results.json`, `initial-checks.json`, and `independent-results.json`. The independent check reads `check-results.json` and `initial-checks.json`, so retain this order. The full passes cover the theorem-required 182,706 modular tests through coefficient index 1,901,366. Omitting `--full` runs only certificate reconstruction and small exact-integer checks; it does not reproduce the full coefficient verification. Runtime `seconds` fields may differ.
`P12-arithmetical-critical-groups`	`python3 check_examples.py`	SymPy; Python >=3.9	Writes `exact_checks.json`; prints `PASS: 8 extremal pairs, primitive kernels, and nonunit star scaling`. Uses exact Smith normal forms, rational arithmetic, and integer kernels.
`P13-nilperiod-rings`	`python3 check_corbas.py`<br>`python3 C20_verify.py`<br>`python3 C20_corbas_verify.py`	Standard library	The author check prints per-case and final PASS summaries only. The independent scripts write `C20-verification.json` and `C20-corbas-verification.json`, respectively, and print JSON summaries. All are auxiliary finite multiplication-table/power-map checks of the hand proofs.

Build the PDFs

Install a LaTeX distribution or Tectonic (the preparation used Tectonic 0.17.0). From the repository root, for example:

tectonic --keep-logs papers/P01-lefschetz-classification/manuscript.tex

The manuscripts use standard packages declared in their preambles. Rebuilding may change PDF metadata and therefore its hash; the manifest identifies the exact initially uploaded PDF and source bytes. The detailed original AI review logs are not included; those logs were not human peer review. Source papers are cited in the manuscripts rather than redistributed here.