The Meter You Quietly Tune

In defense of the tuner

Before we go further: this isn’t a story about dishonesty.

Meters drift. Clocks disagree. Events show up twice, or late, or never. Anything that counts at scale miscounts sometimes, and a team that corrects its meter is doing its job. Show me a metering pipeline that’s never needed a correction and I’ll show you one nobody has checked.

So the tuning isn’t the problem. The silence is.

One month in the summer of 2025

Here’s what silence costs, and how fast the bill shows up.

In the summer of 2025, inside a single month, two of the most admired companies in AI found out in public. Cursor moved its popular Pro plan toward usage-based pricing and communicated the change so quietly that users met it as surprise charges. Within weeks the company was apologizing and issuing refunds. Days later, developers noticed Claude Code’s usage limits had tightened without a word, and Anthropic, a company whose engineering nobody doubts, let a full news cycle pass before publishing the formal limits it could have led with.

Both changes were defensible on the merits. Prices change, limits exist for good reasons, capacity and abuse are real. But a quiet change isn’t experienced as a decision; it’s discovered. And nobody reads a surprise on their bill as a pricing update. It feels like something done to you.

The pattern is older than AI tooling. In 2017, Uber discovered it had spent two and a half years computing its commission on New York drivers’ gross fares instead of net. It called the mistake an accounting error and repaid about $45 million. The error was ordinary. What made it a national story was the two and a half years nobody could see it, including Uber. That’s the other thing a silent meter costs you: without a checkable record, even the vendor doesn’t know what its own meter has been doing.

What silence erases

The deeper cost shows up later, and Facebook paid it in full view. For roughly two years, its metric for average video view time overstated reality, and advertisers rebuilt strategies around the inflated number. The $40 million settlement wasn’t really about the bug. The sting was the allegation that Facebook had known for months before saying anything. True or not, Facebook couldn’t prove otherwise, and there’s the trap.

Once corrections happen in silence, an honest meter and a dishonest one look exactly the same from the outside. No one has to do anything wrong for that to be true; silence does it on its own. Every quiet fix made in good faith looks identical, from where your customer stands, to a fix made for a worse reason. You know which one yours was. Your record doesn’t.

The meter you quietly tune is the meter you can’t defend.

And if you think being right will save you, California’s largest utility would like a word. In 2010, PG&E’s new smart meters were accused of overbilling. Regulators ordered an independent evaluation, and it found the meters accurate. PG&E was right, and it barely mattered. The fight ate years, critics attacked the audit itself as conflicted, and the lesson stuck: proof assembled after the accusation, commissioned by the accused, is always suspect. Being right isn’t the same as being believed, and the proof has to exist before the question does.

What visibility buys

Fair question at this point: if most quiet fixes are never noticed, what does showing them actually get you?

Smaller arguments, first. A visible correction turns “this bill is wrong” into “walk me through March 12,” and a dated entry with a stated reason mostly answers itself. You’re trading the rare, expensive fight about a total for the occasional cheap question about a line item, and the line item conversation has a property the other one lacks: it ends.

It also buys you the benefit of the doubt at scale. Customers won’t check most of your entries, the same way nobody re-adds most spreadsheets. They extend trust to the thousands of numbers they’ll never look at because the handful they did look at held up, corrections included. That’s the strange economics of the margin note. The visible mistake is what makes everything else believable.

And it decides which story you’re in when a correction finally surfaces, because the same fact reads opposite ways depending on who brings it up. A fix your customer finds disclosed in the record reads as a well-run shop. The identical fix discovered sideways reads as something hidden. Nobody was ever really angry about Cursor’s prices or Anthropic’s limits. They were angry about finding out sideways.

No erasers

Bookkeepers ran into all of this five hundred years ago and settled it with a rule that still holds: no erasers. When an accountant finds an error, she doesn’t scratch out the entry. She posts a correction next to it, dated and explained. The error in plain view is what makes the rest of the book believable. The margin note beats the eraser every time, for one simple reason: the margin note can testify.

That’s the whole idea behind recordation, the thing we build at PacSpace. Usage gets committed to a shared record before the invoice goes out, and when a number needs fixing, the fix gets committed too: the error, the correction, and the reason, all legible to both sides. Nothing about your job changes. You still tune the meter, on the last day of the month if that’s when you find the problem. The tuning just happens where it can defend you later, and where your customer watches the month build instead of auditing it after the fact.

The month still has to reconcile, and the meter will always need tuning. Meters are like that.

So tune it. Just put down the eraser.

Related reading: Who Notarizes the Meter?, on whether the meter itself can be checked. This note is about what happens to its corrections.