How the "What is Right?" track holds up against real and worked organizational cases.
Six worked examples, run different ways and described here for what each actually shows, rather than folded into one undifferentiated “it works.”
Methodology, read alongside the cases below
The verdict below any given rule — coexists, asymmetric, disproportionate, structurally excluded — is a measurement, not a ruling. Seven conditions determine whether a given verdict can actually be trusted for the case it was run on.
Worked demonstration
An invented case built to show the mechanism end to end: two proposed rules tested against a fictional hospital's own stated values, including how the verdict changes between an everyday policy and a temporary crisis exception.
Extends this case
Real historical case, checked against outcome
The one genuine retrospective check among these four — a real historical case, graphed blind before comparing the system's verdict to what actually happened.
Real case, verdict recorded before the outcome was known
The one genuinely held-out prediction among these: a rule about labor's board-level veto power was tested and scored while Volkswagen's own supervisory board decision was still pending. The board's real resolution landed weeks later, through a negotiated compromise neither side simply won — checked here against the verdict recorded before any of it was known.
Real, still-unfolding case
Shows how much an honest accounting of everyone a rule would bind actually matters: the same rule, tested against a thin graph and a real one, produces two very different-looking answers.
Regulatory mapping, not a validity check
Two real regulations — the EU AI Act's Article 27 and the GDPR's Article 35 — each require naming, before a high-risk system or process goes live, who's likely affected by it and the specific risk to each. That's exactly what this hindrance check already computes. Not a filing generator, and not legal advice — a worked mapping for the compliance officer or lawyer who has to produce that document and remains the one accountable for it.
Real, pending legislation — two tracks, one finding
Canada's Safe Social Media Act, narrowed to one clause: the duty to mitigate “content that foments hatred.” A universalization test and a hindrance check, run independently on the same statutory text, converge on the same structural defect — and a built-in argument audit flags it a third way, unprompted.
Send a real governance rule or policy dispute — yours, your organization's, or one you've read about — and I'll run it through the hindrance check myself and write back with the result and the reasoning behind it.
The result is only as trustworthy as what I have to build from, so please include, as best you can:
A clean two-sided dispute, well described, is something I can stand behind. A case with several stakeholders needs multiple runs and a consistency check before I'd trust it, so I may ask follow-up questions rather than run something too thin — that's not gatekeeping, it's what keeps a verdict from being blamed on the system when it was really an underspecified case. No cost, no signup, and no promises on turnaround; I'm one person doing this by hand. If it turns out to be a good case, I'll ask before publishing it here — attributed or anonymized, your call.