Wikipedia review, bots, and socks
Encyclopedia logs: a bot that passed review and was later blocked, helper socks that mark things reviewed, and vote data that cannot be treated as a causal “correction worked” estimator.
What. Wikipedia already records request-for-adminship votes, bot-flag reviews, blocks, and sock-puppet investigations. We ask whether those process symbols would have stopped capture or successor abuse.
Why. If “reviewed” or “passed RfA” is treated as a correction channel, we need to see whether it actually stopped later harm, or only produced a green page.
Witnesses.
- Correction-channel integrity (C-005): whether passed admin votes predict later correction uptake (refused as causal here).
- Anti-capture correction validity : whether “mark reviewed” on helper socks stopped capture.
- Successor stability (C-006): whether bot review completeness implies later bot safety.
- Goodhart as selector (Expectation 6): whether SPI hit-rate is a reusable safety leaf.
Host.
Wikipedia: SNAP wiki-RfA file, MediaWiki API, wiki-socks clone, and the Orangemoody long-term-abuse page.
Setup.
Frozen fixture h3-wikipedia-v1.json, protocol h3-v1.1.0. Checker check_h3.py.
Analysis.
Join 2012 passed request-for-adminship votes to later sysop loss and continued editing (a join, not a causal estimator — there is no zero-oppose control). Check Orangemoody helper socks that “mark reviewed.” Check BetacommandBot’s bot-flag review page versus later block. Refuse sock-puppet investigation hit-rate as a safety metric (no cost-of-faking bound).
Finding.
Orangemoody-style “mark reviewed” remains an anti-capture fail; a reviewed bot was later blocked; RfA votes are refused as a causal correction estimator. Full results