There is a moment every Monday at 08:30 when SENA is, briefly, its own harshest reader. Before a single new proposal goes out, it pulls up everything it told Atelier Mira last week (every prediction, every drafted action, every confidence range) and marks it against what the world actually did. The result is a ledger. It is not flattering. It is not supposed to be.
Every proposal SENA made, re-read the morning after against what actually happened. Click a row.
Payment or a payment date within 3 working days of the reminder.
Both paid Wednesday. €18,400 in. Relationship intact. Tone mattered.
Sixty percent, and proud of it
Read the header line again: three hits, two misses, one declined, one still open. A sixty percent hit rate. No vendor puts sixty percent on a slide, and that’s precisely what makes the number useful. It wasn’t produced by a marketing department; it was produced by a procedure: state the expectation in advance, in checkable terms, then let reality mark the paper. A system that reports 95% has almost always done something subtler than lying: it has simply never written its expectations down anywhere reality could get at them.
The structure doing the work here is small and unglamorous: every proposal carries a falsifiable expectation with a deadline. Not “improve collections” but payment or a payment date within three working days. Not “optimize the schedule” but zero idle hours on Line 2 and the Bruma delivery held to within two days. Vague hopes can’t be graded, which is exactly why most software speaks in them.
What a miss buys you
Look at what the two misses actually taught. The markdown that was supposed to move four slow movers moved one: the discount was too shallow, and that calibration error is now written against the pricing model, which will cut deeper next time and say so in its reasoning. The outreach to three quiet accounts got three silences, which says the signal may have come too late or the channel was wrong, and both hypotheses are now on the record with a follow-up date. A miss, written down, is training data with a receipt. A miss, hidden, is just the same mistake waiting patiently for its next opportunity.
You don’t learn to trust a system by watching it be right. You learn to trust it by watching what it does when it was wrong.
Why software almost never does this
Not for any technical reason: the reconciliation loop is the easiest part of SENA to build. It’s an incentive problem. A recommendation engine that ends at the recommendation can never be caught; by the time the outcome lands, everyone has scrolled on, and the vendor’s track record resets to zero with every sales pitch. Closing the loop means volunteering to be caught, weekly, in writing, by your own product. It’s uncomfortable. It’s also the only mechanism that turns a tool’s output from claims into a track record, and a track record is the only form of trust that compounds.
One row in last week’s ledger is SENA declining its own idea: an early hardwood reorder killed for resting on a single price datapoint. Declines are graded like everything else: if the price spikes later, the decline was a miss, and it will say so. Restraint gets audited too.
There’s a compounding effect operators notice after a few weeks of this. The ledger makes trust specific. You stop asking “do I trust SENA?” (too big a question to answer) and start knowing precisely where it runs hot: strong on receivables and scheduling, still earning its stripes on pricing moves. That specificity is what lets the autonomy dial move right for one routine and stay left for another, on evidence instead of vibes. The ledger isn’t a report about the product. The ledger is the product.