Methodology
A daily-updated, evidence-cited database that reviews and scores open-source projects for ministry fit. It is a discovery and evaluation layer for operational decisions. It does not make theological judgments. Read the story behind the scores.
Plain-language key: what the interface calls the Trust Score is the Ministry Readiness Index (MRI) defined below; evidence coverage NN% is the 0–1 confidence value shown as a whole percent; a project is a repository. This page keeps the full technical vocabulary so every displayed number can be traced to its rule.
In plain language: every tool is assessed on six things — how useful it is for ministry work, how well it's maintained, how safe its security and licensing are, how hard it is to adopt, how much you can trust its community, and how well AI agents can work with it. Security or licensing problems cap the final score no matter how good the rest is.
The number you see on the boards — the Trust Score — sums these into one 0–100 readiness rating. A tool that holds people data (giving, children, pastoral records) is additionally flagged so you can apply extra care before adoption.
MRI = min(45, 0.25U+0.20M+0.20S+0.15A+0.10C+0.10I) when S < 30; else raw composite
Displayed MRI decays −2 points per calendar month without a full review, with a floor of 20 while unarchived. Archived repositories display 0.
Tie-break: mri_displayed → S → M → U → full_name
The interface displays confidence as an integer percent labeled evidence coverage (0.70 → 70%); the 0–1 value is used everywhere internally. Tier A reviews auto-publish only at 0.80 confidence or higher, with clean flags and passing validators. Reviews at 0.70–0.79 or carrying caution flags go to human review with a Discord alert. Sensitive capabilities or people-data always require founder review, regardless of confidence. Reviews below the confidence floor remain withheld.
Methodology changelog
2026-09-23
Baseline contract
Six-dimension weighting, the S-gate cap, score decay, tie-break order, confidence handling, and provenance labels established.
2026-09-27
Conservative automated adjudication
Shifted from all-human-review to automated publish for high-confidence, no-safety-flag reviews. Flagged and low-confidence reviews defer to a human queue with a Discord alert where the founder can veto or approve.
2026-09-27
Include-and-flag licensing
Repos with non-standard licenses are included and flagged with a visible caution badge rather than excluded. The honest score sorts them down naturally; Safety/Care tags (people data) still defer to the founder queue.
Refinry helps you evaluate software; it does not operate those tools for you.
