The five principles

A ledger may audit the circumstances under which work gets made. It may never grade the work itself.

By conditions I mean things a stranger could verify without having an opinion about the output:

• Hours: how much paid time a person actually had for the task, as opposed to how many tasks they closed.
• Training: paid hours spent getting better at the craft, not hours spent learning this quarter’s tool.
• Junior headcount: whether anyone is still being brought into the profession, because AI eats the entry rungs first and the damage only shows up a decade later.
• Human authorship: who made the thing, disclosed but not assessed.
• Rework: what was scrapped and redone, which is the real price of a fast process.
• Exposure: AI-mediated adverse events landing on people, by position and harm class. The working definition.

Every one of those is externally observable. None of them requires anybody to decide what good looks like.

And each can be set as a floor rather than a target. A floor harms nobody’s freedom: you can always exceed one, nobody is ranked by how far above it they sit, and there is no reward for faking it.

Merit is the other thing entirely. The moment a ledger scores whether work is good, somebody has to define good, that definition becomes the target, and every organisation reporting against it optimises for it. You have accidentally built an authority on quality: a body that decides what counts as a good diagnosis, a sound judgement or a well-made piece of work, standing over professions whose whole job is making exactly those calls.

The intentions of whoever holds that pen are irrelevant. This is the Act I bug applied to my own proposal: latch onto a measurable definition of good, bend the culture to maximise it, destroy the thing you were protecting. It happens by mechanism, not by malice.

So: this is the line between a labour standard, which is dull, enforceable, of the same family as working-time rules and health and safety, and a ministry of taste. Every other principle on this list is a design choice I’d happily lose an argument about. This one is the thing that stops the Human Ledger from becoming the exact failure it was written to fix.

The opposite of a human ledger is not an unmeasured world, because there is no unmeasured world. It is a world measured privately.

Insurers already price AI liability. Rating agencies are learning to price knowledge loss. Those instruments exist, they are proprietary, and they measure downside protection rather than human outcome.

Compulsory publication is not a softer version of the same idea; it is the whole idea.

Act I was an argument that every maximised metric eventually gets bent. A Human Ledger built as a set of targets will be gamed within two reporting cycles.

Built as thresholds you may not breach, it can’t be, because there’s no upside to exceeding a floor and so no incentive to fake one.

Floors govern process; trends govern outcomes

Exposure is the condition that forced this rule, and it applies across all six.

• Procedural cells take real floors. No person may receive an adverse AI-mediated decision without a named human with the authority to reverse it. A floor like that can be breached, caught and enforced, and a minimum review time is the second one.
• Outcome counts cannot. For error counts, adverse decision counts and override rate, a floor is either incoherent (you cannot promise to reject fewer people) or perverse (a minimum override rate manufactures spurious overrides). These are disclosed with their trend and never targeted. The discipline comes from having to publish a worsening line beside last year’s and explain it, not from a threshold.

This keeps Principle Three intact rather than bending it: setting a direction as a target is exactly the maximised metric that gets gamed, whereas publishing the direction is not. Why this was added.

A clean domestic ledger and a filthy supply chain are entirely compatible. We know this because we’ve already run the experiment with waste and with carbon, and had to invent border adjustment to close it.

Offshored volume has to land on the ledger of whoever commissioned it, or the ledger becomes a laundering mechanism with a nice logo.

A ledger that can measure everything will eventually be asked to, and AI makes measuring everything cheap. Every sector has work whose value can’t be seen while it is happening: the doctor chasing a symptom that doesn’t fit the protocol, the associate reading beyond the brief, the engineer building a prototype nobody asked for, the junior who takes three times as long because they are working it out. Measured as it happens, it looks like waste, and waste is the first thing an efficiency programme removes.

So the line works like a nature reserve: an organisation discloses how much protected time its people have, and nobody measures what happens inside it. The size of the space is a condition, and the contents are off limits.

This is also where Principle Two stops. Publish everything you measure, and choose deliberately not to measure everything. Moving that line should take more than an ordinary policy change, because it protects the place where the next thing gets learned. Why this was clarified.