Congress doesn’t hand-write statute language. Staffers, agency counsel, and outside consultants draft it, get paid for it, and move on. So a law is a document like any other: version 1, shipped with bugs. Someone finding an edge case the drafters didn’t think of isn’t “cheating” — it’s the same skill as a pen-tester finding a hole in code nobody meant to leave open.
This lab is not here to tell you that’s good or bad. The moral question is separate from the legal one, and the whole point is to let you sit in that gap instead of resolving it for you. The three cases below are archetypes — drawn from the shape of real events but not pinned to named companies. ◐ mine marks the lab’s own framing and placements: teaching synthesis, not measurement.
A power-hungry AI buildout needs electricity now, faster than the grid can deliver it — so it runs its own generators under a rule meant for temporary equipment: a category written years earlier, for a smaller world, that technically also covers gear left running indefinitely. A team with the money to hire lawyers who read regulations for a living finds that gap and builds inside it.
Legal, on the reading they bet on — and that’s the interesting part. The public reaction is anger, but notice it aims at the result (the emissions, the neighborhood downwind) more than the method (reading a document more carefully than the people who wrote it). Then regulators tighten the rule afterward. That’s the tell: you don’t rewrite a rule you think was followed the way you meant. You rewrite one that got read differently than you wrote it.
A different kind of gap: not a regulatory loophole but a legal grey zone — say, an AI lab facing a claim that some of its training data was acquired improperly (piracy, not scraping). Rather than hunt for daylight in the statute, this team settles — pays a large, known, bounded sum — and keeps building. No edge case exploited, no gap driven through; a bet that certainty is worth more than the fight.
Both panels above assume there was a real gap — legally sound (A) or a genuine grey zone (B). This is the null case. A frontier lab’s restricted-access model becomes the target of a large-scale extraction effort: tens of thousands of fraudulent accounts systematically pulling the model’s most advanced reasoning, tool-use, and coding behavior. There’s no ambiguity being exploited — the access was fraudulent from the first query. The lab responds by escalating past a legal complaint to government oversight, reframing the dispute from “IP theft” to national-security-adjacent capability extraction.
Here’s why it belongs on the same bench: placed on the Legal / Reputational axes, this case sits in a different quadrant entirely — high legal exposure, but the reputational exposure runs the other direction. It’s the victim’s reputation (as a company worth stealing from) that rises, not the actor’s that falls. That asymmetry is the teaching point: the two-axis model built for Panels A and B stops describing the case here — and noticing why the axes break is worth more than pretending it fits.
Same three buckets as every RVP lab — Resources, Variables, Parameters — pointed at both plays at once.
| Panel A · the gap gets used | Panel B · the gap gets paid for | |
|---|---|---|
| Parameter | The regulation as literally written — immovable at the time. | The legal uncertainty around the underlying claim. |
| Variable | The distance between what the drafters meant and what the text says. | Whether to litigate the gap or close it with cash. |
| Resource | Legal talent good enough to find the gap first. | Capital, spent once, to buy certainty instead of a verdict. |
Every move here trades one kind of exposure for another. Drag the two sliders and watch the dot move. Then hit each panel’s preset and see where it lands — the two plays sit in opposite corners. ◐ mine — the placement is the lab’s read, not a measurement.
Your move: the instructor hands you a third case — any company, any decision. Place it on these same two axes with the sliders, and defend the placement in one paragraph. There is no right coordinate. There is only a placement you can argue for.
Both moves are rational. Neither is automatically right or wrong — that’s the point, and the lab won’t tell you which to admire. What it will teach: a law is a first draft written by someone who got paid to imagine every edge case and didn’t quite manage it. If you can out-think the draft, in a capitalist system, you’re allowed to. Whether you should is a second, separate question — and collapsing the two together is exactly how people end up either excusing everything or condemning everything, instead of actually thinking about it.
These three cases are archetypes, not case studies. Each is drawn from the shape of real, reported events — a regulatory-exemption fight, a data-provenance settlement, a fraudulent-access extraction campaign — but the lab deliberately does not pin them to named companies or live dockets, because the skill it teaches (spot the gap; notice when your frame stops describing reality) doesn’t depend on the names. The RVP framing, the two-axis placement, and the “first draft” metaphor for lawmaking are ◐ mine — a teaching synthesis, not a measurement. Spot something off? Email User Zero — corrections get acknowledged right here.