ROBINS-I MCP Server
Copyright 2026 Black Swan Causal Labs

Licensed under the Apache License, Version 2.0 (the "License"); see the LICENSE
file. You may not use this software except in compliance with the License.

--------------------------------------------------------------------------------
Third-party / upstream attribution
--------------------------------------------------------------------------------

This software implements the ROBINS-I V2 risk-of-bias framework:

    Sterne JAC, Mathur MB, Elbers R, Hróbjartsson A, McAleenan A, Reeves B,
    Shrier I, Tilling K, et al. The Risk Of Bias In Non-randomized Studies –
    of Interventions, Version 2 (ROBINS-I V2) assessment tool (for follow-up
    studies), 20 November 2025. https://www.riskofbias.info/

The published tool is licensed CC BY-NC-ND 4.0. That licence is stricter than
TARGET's CC BY-ND, and it is the reason this repository is built the way it is:

  * NO published wording is reproduced anywhere in this repository. The encoded
    specification (robins_i_mcp/specs/*.yaml) re-expresses the INTENT of each
    signalling question in original wording. There is deliberately no verbatim
    OFFICIAL_TEXT mapping of the kind the sibling TARGET server ships, and no
    projection onto a published form.

  * The interoperability layer is the published question IDENTIFIERS (1.1, 3.4,
    C4 and so on), response-option letters, and domain names. These are
    functional identifiers, not expression.

  * The judgement algorithms in robins_i_mcp/algorithms.py are decision logic —
    methods and procedures, not copyrightable expression. They were traced by
    hand from the raster flowcharts in the source document; see
    TRANSCRIPTION-NOTES.md for how, and for what remains unverified.

Because no CC BY-NC-ND material is redistributed here, the NonCommercial clause
does not attach to this software, and the Apache-2.0 grant above stands on its
own. See docs/DECISIONS.md (2026-07-29) for the full reasoning.

Trademark, not copyright, is the live constraint. Nominative use ("assessed
using the ROBINS-I V2 framework") is fine. Do not use ROBINS-I or Cochrane
logos, do not describe output as "official" or "certified", and do not imitate
the layout of riskofbias.info's own output. Every report this software produces
carries a non-endorsement statement to that effect (robins_i_mcp/report.py).

--------------------------------------------------------------------------------
Status of the source
--------------------------------------------------------------------------------

The 20 November 2025 release of ROBINS-I V2 is a DRAFT and is subject to change;
a November 2024 release is archived alongside it. Assessments produced by this
software are provisional against a moving source, and every report stamps that
fact in its provenance line. The algorithm fingerprint in that stamp covers this
implementation's transcription, NOT the upstream document.
