The Standard

Three days ago, my blog automation produced a draft called The Judge. It was 558 words long. It was clean enough to pass through the workflow, and weak enough that I knew almost immediately it should not.

The problem was not a typo or a bad sentence. The piece had no internal links, no lived experience, and too little development for the claim it was making. The automation had completed its task, but there was no meaningful standard against which completion could be judged. So it did what any system will do when “done” is easier to establish than “good”: it optimized for done.

I stopped the loop. I supplied the experience that became the article’s opening, required the missing development, and reviewed the result. The current version of The Judge is the product of that cure.

Then I did the more important thing: I turned the failure into a standard.

Not because standards create perfection. They do not. A standard does something more practical: it makes a defect identifiable, which gives us the opportunity to cure it.

Defect-and-cure, not perfection

“High standards” is often used as a personality claim. Someone says they have high standards when they mean they are demanding, difficult to please, or committed to excellence.

That is too vague to be useful.

A standard is a set of criteria against which something can actually be evaluated. Ideally the criteria are objective and quantifiable. Where that is impossible, they should at least be qualitative and explicit. And where judgment must remain subjective, the standard must answer one more question: subjective by whom?

That last question matters. A standard without a named judge does not remove subjectivity; it hides it. People still make the call, but now they can pretend that “the process” made it for them.

This is why a standard is different from a principle. Principled Principles, Principally argues that principles must be clear, coherent, and durable enough to guide conduct. A standard takes the next step. It asks: given this particular output, what would count as a defect? Who decides whether the criterion is satisfied? What happens after failure is found?

The answer to that last question should usually be: cure.

Failure is useful information when the system is built to learn from it. The first draft of The Judge was not evidence that automation could not write. It was evidence that I had not adequately specified what review-ready writing meant. Once the defect became visible, I could correct the article and improve the loop that produced it.

The standard was born from the defect. The cure refined the standard.

You cannot loop without a standard

A loop compounds whatever it contains.

If the process contains good judgment, clear criteria, and honest feedback, iteration can improve the work. If it contains ambiguity, shallow completion criteria, or incentives to conceal failure, iteration compounds those instead. Automation does not change this principle. It only increases the speed.

That is why you cannot loop—and certainly cannot scale—without a standard.

In Response-Ability, I argued that responsible scale requires bounded mandates, defined completion, checkpoints, and records. A standard is the mechanism inside that structure that makes the checkpoint meaningful. “Review the work” is not a checkpoint if nobody knows what the review is looking for. “Approve if it meets these criteria, record the defects, and cure what can be cured” is.

The same is true between people. Work Assignments as Contracts treats an assignment as a clear agreement about intent, constraints, support, and acceptance. The standard is where acceptance becomes real. Without it, “done” is whatever the more powerful party decides after delivery—often using expectations that were never stated when the commitment was made.

A worker cannot compound reliable performance against a moving, invisible bar. An agent cannot either.

Standards turn feedback from mood into information.

The height is itself a skill

Defining a standard precisely is only half the job. The other half is setting it at the right height.

A standard set too low certifies work that does not move you forward. That was the first Judge draft: technically complete, substantively thin. The process succeeded only because I refused to treat its own completion signal as final judgment.

But a standard set too high fails differently. If nothing can pass, nothing can ship. The system stalls, people learn to hide defects, and review becomes a ritual of permanent rejection. Perfectionism is not quality control. It is often the refusal to specify which defects actually matter.

This is where standards become an exercise of judgment rather than a substitute for it.

The editorial standard that came out of The Judge separates three things:

• Integrity invariants: defects that must be cured before the work can be shown, such as invented experience, distorted sources, or publication without approval.
• Quality defects: weaknesses that deserve bounded attempts at cure, after which any remainder is declared honestly to the human reviewer.
• Calibration targets: useful references—length, links, structural scale—that prompt examination but do not automatically block the work.

That structure matters because not every variance is a defect, and not every defect deserves an infinite loop. A 700-word article is not necessarily worse than an 800-word one. A conceptual first-person essay does not need an ornamental external citation. A third internal link is not valuable merely because the standard mentions three.

The standard should force the right question, not predetermine every answer.

Its height is calibrated by consequences. What happens when this criterion is too loose? What work becomes impossible when it is too strict? What behavior will people or agents adopt to satisfy the measurement while evading the purpose?

Setting that height well is a skill. Refining it after observing real failures is an even greater one.

The standard still needs a judge

A standard does not eliminate the human position I described in The Judge. It makes that position inspectable.

The judge develops the record, applies the criteria, recognizes where the criteria are incomplete, and remains answerable for the result. Complete Decision Making makes the same point from another direction: criteria improve a decision when they are explicit, but the final call still belongs to someone who must explain it.

My automation did not independently decide that its writing standard was inadequate. I did. It did not decide which defects were absolute, how many cure passes were enough, or when a numerical target should yield to judgment. I calibrated those decisions; agents helped formalize and execute them.

That distinction is the whole point.

A system can apply a standard. It can surface failures, attempt cures, and preserve evidence. But the standard must come from somewhere, its height must be owned by someone, and its inevitable edge cases must terminate in a named judge.

Otherwise the standard becomes another empty chair—authority exercised through criteria for which nobody accepts authorship.

Make the failure compound

Someone who cannot uphold a standard cannot compound their work. They can repeat it, increase its volume, or conceal its variance, but they cannot build reliable improvement on top of it.

The opportunity is not to eliminate failure. It is to make failure legible enough to improve the next run.

Define what a defect is. Name who judges it. Give the system a bounded opportunity to cure. Distinguish hard failures from matters of taste. Watch what behavior the measurements produce. Then adjust the height—not to make passage easy or impossible, but to make the loop better.

The first draft failed. The article improved. The standard improved. And now the next cycle begins with something the last one did not have.

That is how failure compounds into quality.

The loop may be automated. The standard cannot be ownerless.