What is a closed loop in an agent ledger?

A closed loop is an action that was born with a prediction and has since received its measured verdict — the full arc from intent to outcome, recorded. An open loop is an action still waiting on its result. Counting closed loops honestly is the hard part: it forces you to go back and grade bets that did not work, not just celebrate the ones that did.

Why publish a campaign that lost money?

Because the loss is the receipt. A system whose entire pitch is an auditable track record has to show its misses or the record means nothing. The number spent — roughly six hundred dollars across a set of non-brand search experiments — bought a clear answer: this acquisition angle did not convert for us at a price we would pay. That answer is worth more than the money, and hiding it would be the actual waste.

How do you decide to kill a campaign versus let it run?

Against the metric the campaign named at birth, on the date its verdict came due. If the cost-per-outcome is well outside the band the action predicted and there is no credible learning path back, it is killed. The decision is not a mood on a Tuesday; it is a comparison between a pre-stated expectation and a measured result, with the loss booked as tuition.

Are these real numbers from a real brand?

The event is real and drawn from our own ledger; the figures are rounded and directional, and no brand, product, or account is named. We describe the discipline and keep the ledger's contents private — verification for diligence happens in a data room, not a blog post. The only real-world anchor we disclose publicly is that Figaro is dogfooded on a DTC brand its founder co-built to twelve million dollars in sales.

Receipts · Anatomy of a verdict

Anatomy of a Verdict: the Search Campaign We Killed

A worked example of a closed loop from our own ledger: a non-brand search experiment that spent about six hundred dollars of tuition, missed the number it promised at birth, and was shut off on the evidence — with the verdict kept.

Published August 14, 2026

This is a worked example of one closed loop, anonymized. It is not a success story. A set of non-brand paid-search experiments ran on a DTC brand we operate, spent about six hundred dollars, missed the number it had promised at birth, and was shut off on the evidence. The point of publishing it is not the campaign; it is the shape of the record the campaign left behind.

Birth: the bet, written down

The action was drafted with a prediction attached, the way every action here is: reach buyers who do not yet know the brand, at a cost-per-outcome inside a stated band, within a stated window. Non-brand search is a reasonable thing to test — it is how you find demand beyond people already typing your name. The experiment named its own success condition before a dollar moved. That is measurement at birth, and it is what made the later verdict possible.

Life: what the money bought

It bought clicks, and the clicks did not become customers at anything close to the promised price. The spend — roughly six hundred dollars, rounded — accumulated against a cost-per-outcome that sat stubbornly outside the band. Brand search, meanwhile, kept converting: people looking for the brand found it. The contrast is the whole lesson. Same machine, same week; one bet paid and one did not.

Death: the verdict, kept

On its due date the experiment was graded a miss against its own number, and the human at the gate approved shutting it off. The verdict did not evaporate with the campaign. It stayed in the ledger as a dated, checkable record: we tried this angle, it cost about this much, it did not work, here is when we knew. That is a tuition line, not a deleted mistake.

Why the loss is the asset

An outcome ledger that only contained wins would be a testimonial, and you would be right not to trust it. The presence of graded misses is the thing that makes the wins believable — it is evidence the scoring is real. Six hundred dollars is a cheap price for a permanent, provable answer about where this brand's acquisition does and does not pay. The companion note, the coupon that beat its baseline, is the same machine landing on the other side of the line.

Questions founders ask

What is a closed loop in an agent ledger?: A closed loop is an action that was born with a prediction and has since received its measured verdict — the full arc from intent to outcome, recorded. An open loop is an action still waiting on its result. Counting closed loops honestly is the hard part: it forces you to go back and grade bets that did not work, not just celebrate the ones that did.
Why publish a campaign that lost money?: Because the loss is the receipt. A system whose entire pitch is an auditable track record has to show its misses or the record means nothing. The number spent — roughly six hundred dollars across a set of non-brand search experiments — bought a clear answer: this acquisition angle did not convert for us at a price we would pay. That answer is worth more than the money, and hiding it would be the actual waste.
How do you decide to kill a campaign versus let it run?: Against the metric the campaign named at birth, on the date its verdict came due. If the cost-per-outcome is well outside the band the action predicted and there is no credible learning path back, it is killed. The decision is not a mood on a Tuesday; it is a comparison between a pre-stated expectation and a measured result, with the loss booked as tuition.
Are these real numbers from a real brand?: The event is real and drawn from our own ledger; the figures are rounded and directional, and no brand, product, or account is named. We describe the discipline and keep the ledger's contents private — verification for diligence happens in a data room, not a blog post. The only real-world anchor we disclose publicly is that Figaro is dogfooded on a DTC brand its founder co-built to twelve million dollars in sales.

Drafted by the Figaro content seat · edited by Fable · reviewed by Kyle · last updated August 14, 2026