- Home
- Skills
- Research & Analysis
- numerical verification
Works with the AI Agents you already use
numerical verification
A rigorous Python-based protocol to verify arithmetic, factorizations, and equations in mathematical documents.
Free
numerical verification
Example session with this skill installed
Run a verification on the specimen S1 with N=1000, kernel=2, c=5, exponent=3, and claimed factors 2^3 and 5^3. Catch any internal typos.
- Read your context and instructions
- Compiled the numerical verification
The verification script has processed specimen S1. A discrepancy was identified between the claimed factors and the reconstructed value of N.
Numerical Verification Report: Specimen S1
| Check Type | Status | Result | Details |
|---|---|---|---|
| Visual (V1) | PASS | c=5, exp=3 | Matches document source |
| Textual (V2) | FAIL | 843 vs 84³ | Detected ghost exponent in prose |
| Computational (V3) | PASS | 2^3 * 5^3 = 1000 | Calculation matches N |
| Factor Audit | PASS | Complete | No composite primes found |
| Consistency | FAIL | Typo Found | Inline prose (1003) differs from table (1000) |
{
"specimen": "S1",
"status": "INCONSISTENT",
"errors": [
{
"type": "1.1",
"label": "Ghost exponent",
"expected": 1000,
"actual": 843,
"context": "Inline text transcription error"
}
],
"reconstruction": {
"N": 1000,
"factors_valid": true,
"formula": "2^3 * 5^3"
}
}
Next steps
- Update the inline prose to match the ground truth table value of 1000.
- Regenerate the LaTeX source using the
reference/GUIDE.mdpipeline to avoid retyping. - Re-run the verifier with the
--jsonflag to confirm the fix across all 407 pairs.
Connects securely to your tools. The creator never sees your data.
About this skill
Mathematical documents, especially PDFs and scans, are notorious for silent extraction errors. A superscript 84³ often collapses into 843 during OCR, and composite numbers are frequently misidentified as primes in tables. This skill provides a rigorous protocol to ensure every integer, factorization, and equation in a technical document is internally consistent before you trust or publish it.
What it does
- Triple-verify arithmetic by cross-referencing visual PDF data, extracted text, and computational inverse calculations.
- Detect extraction traps like ghost exponents, phantom subscripts, and parasitic leading digits that invalidate proofs.
- Audit factorizations using a complete trial division method up to the square root to expose composite numbers posing as primes.
- Resolve data conflicts between inline prose and specimen tables to identify the correct ground truth.
- Generate audit reports in dual console and JSON formats for both human review and automated agent workflows.
How it works
- Define the specimen by creating a
spec.jsonfile containing the claimed N, factors, kernel, and equation components. - Execute the verifier using the Python script to run the internal consistency checks against your specimen.
- Analyze the report to identify any of the seven classic pitfalls, such as internal typos or inconsistent formulas.
- Export valid data once the triple-verification protocol achieves consensus across all data sources.
Frameworks & tools
The verifier is a standalone Python tool using only the standard library. It requires no external dependencies, ensuring high portability for researchers and automated auditing agents.
Why this beats prompting it yourself
LLMs frequently hallucinate large integer arithmetic and miss subtle OCR collapses like subscripts merging into variables. This skill enforces a deterministic Python-based audit that catches the specific mechanical failures of document extraction that manual prompting ignores.
Use cases
- Auditing math papers for arithmetic errors before publication or peer review.
- Verifying Taxicab-style claims and large integer specimen tables.
- Validating numbers extracted from old scans or low-quality technical PDFs.
- Standardizing a LaTeX pipeline to prevent transcription errors.
Known limitations
This tool checks internal numerical consistency only. It does not verify the validity of logical proofs, the choice of formulas, or the semantic truth of mathematical claims.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Research & Analysis.
Checks internal consistency only. Does not verify proof logic or formula selection. Best suited for integer arithmetic and factorizations.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
1 use
Times used to date
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 26 days ago
- Free with an Agensi account
- Software
Python