Editorial team and review
Every evaluation on Argupedia is work in progress. This page explains who is responsible for it, how it is produced and how you can tell how thoroughly it has been reviewed.
Responsible
The editorial team consists of the founders Julian Ohm and Paul Ohm. Julian Ohm is responsible for the content under German media law (legal notice).
How an evaluation is produced
- Draft. A language model drafts the evaluation under the open standard that applies equally to every debate: arguments from both sides, and for each argument its value, impact and plausibility with reasons and sources. The rules are described under How Argupedia scores, the conversion anchors under Baseline, the values under Values and the evidence tiers under Plausibility.
- Automated check. Before anything is published, a checking program recomputes every figure: scores, totals, the balance from 20,000 draws, the caps of the evidence tiers and whether every source marker points to a source. An evaluation that fails this check does not go online.
- Review rounds. An AI model goes through the evaluation again: it looks for calculation and source errors, missing arguments and new legislation, data or studies, and proposes corrections. Notes from the editorial team and from readers feed into the same rounds.
- Human approval. A member of the editorial team looks through the errors found and the proposed changes and approves or rejects them. The round then carries their name in the log. If a figure changes, a new version is created; the old one stays visible in the log.
The rounds repeat. Over time, every evaluation is reviewed more often, corrected more often and becomes more accurate, and the log shows how far along that path it is.
How to read the review status
At the foot of every debate is the review box. Its counter shows how many review rounds the evaluation has been through and since when, how many of them corrected something and how often a human approved it. Expanded, it lists every round with its date and result: for an AI review the model, for an approval the name. The log is only ever added to, never rewritten.
If no human has approved an evaluation yet, the top of the page carries the notice “AI evaluation · not yet approved by a human”. An approval covers the figures the human has read. If an AI later changes figures, the notice reads “AI re-evaluation · not yet approved by a human” until a human has read the new version. Such pages remain readable, but are only listed for search engines after approval.
Sources and quotations
Every figure in a justification carries a source marker that points to the debate's list of sources. Quotations from the political debate are given verbatim and with their source. Where a figure is set rather than measured, the justification says so, and the plausibility is lowered accordingly.
Corrections
All changes to evaluations are listed chronologically under Changes, with the verdict before and after. Please send errors, missing arguments or newer sources to info@argupedia.com. Every note goes into the debate's next review round; if it leads to a change, the change appears in the log.
Open data
The entire stock, with all arguments, figures and sources, is available as an open dataset so that anyone can recompute it.