Evidence Synthesis Traceability & Extraction QA

    1

    Generates source-traceable evidence matrices and extraction QA reports from research papers and technical documents.

    Free

    1 useSecurity scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Evidence Synthesis Traceability & Extraction QA

    Example session with this skill installed

    Run an extraction on the three PDFs in /papers using the intervention-schema.md. Identify any conflicting values between reports and prepare an adjudication queue for the primary investigator.

    • Read your context and instructions
    • Compiled the evidence synthesis traceability
    • Generated the data export

    I have completed the extraction for the three papers in the /papers directory. The extraction followed the intervention-schema.md protocol, identifying one potential study duplication and two conflicting outcome values.

    Evidence Synthesis Package

    Manifest Summary

    FileStatusPage CountExtraction Mode
    smith_2023_primary.pdfSuccess14pdf-text
    smith_2023_appendix.pdfSuccess5pdf-text
    jones_2022_pilot.pdfSuccess8pdf-text

    Evidence Matrix (Excerpt)

    Study IDFieldValueStatusAnchor
    SMITH-23Sample Size450reportedp. 4, Table 1
    SMITH-23InterventionCBT-ireportedp. 2, Methods
    JONES-22Sample Size42reportedp. 3, para. 2
    JONES-22Mean Outcome12.4conflictingp. 5, Table 2

    Adjudication Queue

    Ref IDFieldValue A (Anchor)Value B (Anchor)Conflict Type
    ADJ-001Mean Outcome12.4 (Table 2)12.8 (Results text)Within-report discordance
    ADJ-002Study IdentitySMITH-23 PrimarySMITH-23 AppendixPotential companion reports

    Next steps

    1. Review the adjudication-queue.csv to resolve the mean outcome discrepancy for JONES-22.
    2. Confirm the study family link between the Smith 2023 primary paper and its appendix.
    3. Inspect qa-report.md for a full breakdown of not_reported fields across the corpus.

    smith_2023_primary.pdf

    CSV · data export

    Generated

    Example file from a real run - the skill writes it into your workspace.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Extracting intervention and outcome data from academic papers or technical reports often results in "black box" summaries that hide uncertainty and lack traceability. This skill solves the trust problem in evidence synthesis by generating a source-traceable evidence matrix where every cell is anchored to a specific page, table, or quote.

    What it does

    • Anchor-backed extraction ensures every data point includes a source ID and precise location (page, section, or figure).
    • Status-aware logging preserves explicit states like not_reported, unclear, or conflicting instead of using blanks or zeros.
    • Conflict adjudication flags disagreements between independent extraction passes or companion reports for human review.
    • Duplicate detection links related reports via DOI, registry IDs, and recruitment periods to prevent double-counting.
    • Automated QA validation runs integrity checks on sample sizes, units, and effect estimates to catch extraction errors.

    How it works

    1. Initialize scope by defining the review question, target fields, and eligible source set in a formal extraction schema.
    2. Inventory sources using scripts/manifest.py to generate a file manifest with SHA-256 hashes and page counts.
    3. Execute extraction in pdf-text, ocr-review, or structured-records mode depending on the source material quality.
    4. Generate QA package including the evidence matrix, data dictionary, adjudication queue, and reproducibility manifest.

    Frameworks & tools

    The skill utilizes Python scripts for manifest generation and matrix validation (scripts/manifest.py, scripts/validate_matrix.py). It produces standard outputs in Markdown, CSV, and XLSX formats, following structured schemas defined in YAML.

    Why this beats prompting it yourself

    Standard prompts often hallucinate missing values or guess at sample sizes when data is ambiguous. This skill enforces a "fail-closed" logic that requires a specific evidence anchor for every populated cell, making the output audit-ready and reproducible.

    Use cases

    • Systematic reviews to build a first-pass extraction matrix for intervention and outcome data.
    • Technical audits to reconcile independent extractions and flag discrepancies for a senior reviewer.
    • Evidence mapping to identify gaps in reporting across a corpus of clinical or technical papers.
    • Data cleaning to link companion reports and companion registry entries into a single study family.

    Known limitations

    Does not perform clinical interpretation, meta-analysis, or risk-of-bias certification. Requires expert review for all interpretation and sign-off. Source files must be text-readable or OCR-ready.

    Requires expert review for interpretations. Not for clinical sign-off or meta-analysis. Page anchors depend on source document formatting.

    How to use

    1. 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. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    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 forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 27 days ago

    • Free with an Agensi account

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Listed27 days ago

    What's inside

    Frequently Asked Questions

    Unlike standard prompts that may guess missing values, this skill enforces 'anchor-backed extraction,' meaning every data point is tied to a specific page or quote. It eliminates 'black box' summaries by using explicit labels like 'not_reported' or 'conflicting' to ensure full auditability.

    The skill requires your source documents (PDFs or technical reports) to be text-readable or OCR-ready. It uses a manifest system to track files by SHA-256 hashes, ensuring that even if filenames change, the evidence remains linked to the correct source material.

    You receive a comprehensive toolkit including the extraction engine, specialized Python scripts for manifest generation and matrix validation, and structured YAML schemas for defining your data fields. It also includes templates for the evidence matrix and the QA validation reports.

    This skill acts as a first-pass extraction and QA tool; it does not perform clinical interpretation, meta-analysis, or risk-of-bias grading. All outputs are designed to be 'audit-ready' but require a human expert to perform the final review and sign-off.

    Yes, the skill includes a conflict adjudication feature that flags disagreements between different extraction passes or companion reports. This allows you to quickly identify and resolve discrepancies without manually re-reading every document.

    New in Research & Analysis