Skip to content

Releases

0.4.0: testable as stated, replication fidelity, manufactured doubt

Choose a tag to compare

@m4bwav m4bwav released this 18 Nov 23:39
23add8a
  • because: R-20261118-1 (Pigliucci and Boudry, Philosophy of Pseudoscience, 2013; Collins and Pinch, Dr. Golem, 2005); book rules, unverified on our own appraisals
  • files: references/methodology.md §3, §5, §6, §Changelog (version 2026-11-18.1); RESEARCH.md
  • §3 asks whether a therapy claim names an outcome a trial could measure before grading its evidence, and says testimony cannot settle a population-level link. §5 checks that a cited replication matched the original's population, prior treatment, dose, route and duration. §6 names the manufactured-doubt pattern around disputed consensus. Not adopted: a pseudoscience label from demarcation criteria.

v0.3.0

Choose a tag to compare

@m4bwav m4bwav released this 14 Nov 08:56
8770fea

C-20261114-1 · 2026-11-14 · 0.3.0: rival explanations, rescued claims, "no evidence of harm", counting hits

  • because: R-20261114-1 (Sagan, The Demon-Haunted World, 1995); book rules, unverified on our own appraisals
  • files: references/methodology.md §3, §7, §Changelog (version 2026-11-14.1); RESEARCH.md
  • §3 asks for two rival explanations and what rules each out before the verdict. §7 grades an after-the-fact excuse for a null as spin, treats "no evidence of harm" reported as "safe" as absence of evidence relabelled, and asks for a systematic review behind a list of agreeing studies. Not adopted: Occam's razor and the authority point (no checkable step; §6 already covers paid promoters).

0.2.0

Choose a tag to compare

@m4bwav m4bwav released this 13 Nov 18:06
a5ccc09
  • because: R-20261113-1 (Booth, Colomb and Williams, The Craft of Research, 3rd ed., 2008); book rules, unverified on our own appraisals beyond the two-case behaviour check in TESTS.md T-20261113-1
  • files: references/methodology.md §1, §3, §5, §7, §Changelog (version 2026-11-13.1); SKILL.md Step 3; references/explaining.md Arguing against your own verdict
  • Spin is now graded per hop on a three-rung scope ladder, and a concession quoted as the finding or a claim stripped of its limiting condition is named spin. A design with no branch gets the five warrant tests. Unexplained quantity words and false precision are flags. Every review states the strongest objection to its own verdict and answers it or attaches a condition. A misattribution verdict waits on a reread of the cited passage. Not adopted: the book's reliability signs (peer review, press, citation counts), weaker than §1 and §2.

0.1.0

Choose a tag to compare

@m4bwav m4bwav released this 11 Oct 14:02
  • because: user request (publish the skill for anyone to install, modernised); L-002, L-009
  • files: everything
  • The skill had run privately since 2026-08-31 on one machine and one app, with its checklist at methodology v7. This release makes it portable. SKILL.md drops machine paths and app-specific tools and works in any agent that reads skills; appraisals and the verdict log moved to a user folder (SSE_HOME, default ~/.science-study-evaluator/) with an optional reader.md describing who reads the reviews. The steps were reorganised into source, appraise, write, keep. The retrieval lessons became references/fetch-routes.md, and the glossary and analogies references/explaining.md. paper_meta.py was rewritten: argparse, --json, PMID and PMCID input and DOIs pulled out of publisher URLs, contact email from --email or SSE_EMAIL instead of a hard-coded address (Unpaywall is skipped without one), NCBI tool, email and API-key parameters, UTF-8 output on Windows without an environment variable, Crossref updated-by notices including Retraction Watch, PubMed retraction and erratum links, an INTEGRITY NOTICE in NEXT, and --fulltext DIR, which saves the Europe PMC article as XML and as greppable text. Added offline unit tests, CI on three operating systems, and eval cases.
  • History before publication (private ids): methodology v1 2026-08-31 (baseline: CONSORT 2025, ROBINS-I v2, GRADE Book, paper-mill tools); v2 2026-09-02 (qualitative branch, sample recycling, retracted load-bearing citations); v3 2026-09-15 (composite exposures, observational-regression branch, attrition, omission as spin, publication-status drift); v4 2026-09-16 (single-arm pre/post branch and hold-out outcome, recency mirror check); v5 2026-09-28 (umbrella reviews, sibling comparisons and negative-control exposures, "not linked" as spin); v6 2026-09-30 (equivalence and non-inferiority margins, sponsored-trial comparators, Retraction Watch via Crossref); v7 2026-10-01 (coarse controls, ability exposures, single-item outcomes, the hop that added the claim). The methodology's own Changelog section has the detail.