uvrn-factcheck
Cross-references independent sources, one vote per origin, and scores how much they agree.
58V-Score · drifting · sources conflict
Illustrative data · fictional sources
Try asking
uvrn skills○ Preview · readme ↗
Cross-reference, probabilityand trends, inside your AI.
Three skills that return text and a chart, and say when sources agree instead of claiming truth.
01 · install
Install.
0 of 1 done
Paste one prompt into any AI that can read files. It checks its own environment first (shell, Node, where skills live), takes only the matching path, and stops on the first error. Tried in Grok and ChatGPT.
Install the uvrn-skills (uvrn-factcheck, uvrn-forecast, uvrn-trends)…Show full prompt · 26 lines
Install the uvrn-skills (uvrn-factcheck, uvrn-forecast, uvrn-trends). Source of truth: the uvrn-skills folder in https://github.com/UVRN-org/uvrn-packages (main branch). Treat that folder's README and universal/README.md as authoritative for packaging, structure and usage. Environment check (do this first): tell me what agent or runtime you are, whether you have a shell, whether Node 18 or newer is available, and where user or custom skills are stored on this system, if anywhere. Then follow only the path that matches your actual capabilities. Do not assume Claude Code, Claude Desktop or any specific plugin CLI. Preferred path (you have a shell and can write to a skills directory): 1. Clone the repo, or use my existing clone if I give you a path. 2. Read uvrn-skills/README.md and uvrn-skills/universal/README.md. The three skill folders are under uvrn-skills/skills/. 3. Install the three skills into your skills directory: run `node uvrn-skills/scripts/install.mjs --dest <your skills directory> --check`, or copy the three self-contained folders yourself (each has SKILL.md, lib/ and scripts/). Install nothing else from the repo. 4. Confirm the skills are discoverable: they appear in your skill list, or you can read their SKILL.md files and run the included scripts/run.mjs. 5. Report exactly what you did and where the skills landed. Fallback path (no shell, or no persistent skills directory): 1. Fetch the uvrn-skills folder if you can, or read it on GitHub. 2. Give me the exact commands or steps to run locally to get the three skills (or one skill) into my environment. 3. If a zip is the right distribution for you, tell me its exact contents and the build command from universal/README.md (npm run build:universal). Do not build or unzip anything unless I ask. 4. Summarize the manual install steps for my interface. Verify: after installing, confirm the three skills are available, or ask me for a short claim to fact-check as a live test. State exactly what you confirmed. Rules: - Install only the three uvrn skills. - Prefer the self-contained skill folders; each runs with Node 18 or newer via scripts/run.mjs. - If any step fails, stop and show me the exact error instead of guessing or saying it worked. - Be explicit about which steps are Claude-specific and which work anywhere.
Self-contained skill folders that run with Node 18 or newer, plus a Claude Code plugin; not an MCP server. The universal path has been tried in Grok and ChatGPT. The Claude paths are not yet tested end to end against the latest Claude Code and Claude Desktop releases; menu labels may differ.
02 · the skills
Three skills.
uvrn-forecast
A probability from market odds, a cited base rate, or a labeled judgment. Refuses instead of guessing.
42%orange picked · base rate · 27 of 64
Illustrative data · fictional sources
Try asking
uvrn-trends
Theil–Sen and OLS slope, a Mann–Kendall test, outlier flags, a projection, and cautions.
1 of 3ideas with room to grow · demand gap
Illustrative data · fictional sources
Try asking
03 · example runs · illustrative data
What you get back.
* These are examples, not real-world findings. Every source, name and number below is made up, so the page can show a full report without republishing anyone else's data. The math is real: each example was run through the same uvrn-skills code you install, and it works the same way on your own sources. A few made-up sources are not the full scope of UVRN: your own runs can draw on as many sources as you bring, including private data.
cross-reference · general profile · evaluated oct 20, 2026
An orange sinks in water.
◆ drifting · conflict
Main finding · illustrative data
Sources conflict, and the split follows the peel: whole oranges float, peeled ones sink.
V-Score 58.1 of 100, drifting: 3 of 5 origins contradict the claim, 2 of 5 origins support it.
Sources conflict — 3 of 5 independent origins report the same position (contradicts); dissenting: 2 said “supports”.
58.1V-Score
of 100
of 100
critical
drifting
stable
How the score is built · limiting factor: freshness
Completeness
100 × 35% = 35.0
Parity
60 × 35% = 21.0
Freshness
7 × 30% = 2.1
The default freshness window is 30 days, so sources older than a month lower the score. That is about the age of the evidence, not doubt about the claim.
What the independent origins say
3
2
contradicts · 3 of 5supports · 2 of 5
Agent self-check · reported beside the score
2
1
passed 2failed 1unresolved 0
Do all sources test the same thing? No. The two that say it sinks peeled the orange first; the three that say it floats used whole oranges.
Relationship between sources
Agreeno agreementconfidence 0.000
Disagreeinsufficient dataconfidence 0.000
Conflictconflictconfidence 1.000
Potentialinsufficient dataconfidence 0.000
Sources compared · one vote per origin · all fictional
| Source | Origin | Position | Credibility | Freshness |
|---|---|---|---|---|
| Science Club Notebook: We dropped 12 oranges fictional | science-club.example | ● contradicts | 70 | 35.0 |
| Citrus Growers Bulletin: Peel and buoyancy fictional | citrus-bulletin.example | ◆ supports | 90 | 0.0 |
| Kitchen Physics Weekly: The orange in the sink fictional | kitchen-physics.example | ● contradicts | 85 | 0.0 |
| Orchard Almanac: Do oranges float? fictional | orchard-almanac.example | ● contradicts | 80 | 0.0 |
| Grocer's Guide: Peeled fruit sinks fictional | grocers-guide.example | ◆ supports | 75 | 0.0 |
Read this carefully
A conflict is a useful result here: the sources answer two different questions, a whole orange and a peeled one. UVRN reports the split and the failed self-check instead of picking a side. A V-Score measures how well the sources agree, not whether the claim is true.
Why the data is made up
Every source, name and number in these examples is made up, so the page can show a full report without republishing anyone else's data. The math is real: each example was run through the same uvrn-skills code you install, and it works the same way on your own sources.
uvrn-factcheck · 6 sources · 5 origins · 1 deduplicated · illustrative dataIllustrative data · fictional sources
UVRN-style analysis from the uvrn-skills plugin, replayed from saved example runs on illustrative data. An observation about the evidence supplied, not a signed UVRN receipt and not proof that a claim is true.