When the agent missed a plan match

Monday morning. You open last week's review and the headline looks off — Thursday is sitting in the off-plan column, but you remember eating almost exactly what the plan said. The only thing you swapped was chicken for turkey. The agent didn't bridge the gap on its own, and now you want to confirm by hand that Thursday's lunch was the planned meal.

What the agent needs to hear

Behind your sentence, the agent is working backwards to three pieces: which day and meal slot you’re talking about, what you actually ate that you think was the plan, and how it deviated from what the plan listed — a substitution, a different amount, or a different food entirely. « Thursday lunch, I ate turkey instead of the planned chicken » is enough. If you have the date but not the slot, name what you remember of the meal and the agent will find it.

Naming the deviation type sharpens the lookup. A swap (one food for another) is different from a portion drift (same food, way more or less), which is different from « I ate the whole plan but the catalog name didn’t match what I logged ».

How auto-match works

You don’t have to log « from the plan » for a meal to count as planned. When you log freestyle and the plan has an active slot at that hour, the agent reads both and tries to bridge automatically. Same food (by catalog ID, by name, or by category), a quantity within roughly twenty percent of the plan’s target, same meal slot — that’s a planned-slot hit, no action needed. Alternatives you declared in the plan also auto-match the same way.

This is why most of your freestyle days post-hoc look the way you expect them to. You eat what you eat, the agent reconciles at read time, and the weekly review reflects reality without you babysitting the tags.

Why some matches slip through

The heuristic has limits, and they cluster around three shapes. A swap to a food in a different catalog category — chicken to turkey, beef to bison — won’t auto-bridge because the foods share a meal slot but not a category root. A custom item the catalog can’t normalize (your own protein cookie, a spelling that drifted from the master entry) won’t match either. A portion well outside the twenty-percent band — half a serving, a double — falls out of tolerance even when the food itself is right.

The signal that one slipped through is the weekly review: coverage looks low for a day you remember eating cleanly. That gap between memory and number is the cue to pull the log up.

When the agent gets it wrong

If the entry the agent surfaces is the wrong one, name the swap out loud — « the turkey was the substitute for the chicken, mark it as planned » or « the protein cookie was supposed to be the planned snack ». The agent persists the link directly: the log line gets tagged as a clean match or a declared substitute, and adherence recomputes on the next read. Thursday stops sitting in the off-plan column without you opening anything.

The same flow handles portion misses and custom-food misses — you’re telling the system « this log line was the plan, here’s the planned item it maps to ». For ambiguous matches the agent can’t decide on its own — two planned items that both could be the partner — it asks before persisting. The app’s reconciliation sheet is the fallback if you’d rather pick from a list visually.

What makes the reconciliation worth doing

Three things decide whether a manual reconciliation actually helps the numbers: the identification is right — you know which log entry and which planned item are the pair, not just « Thursday was fine » — the intent is honest — a clean match, a declared substitute and an accepted portion drift are different links and the score reads them differently — and the timing matters — reconciling inside the active week moves the headline adherence and streak immediately, while older weeks recompute the snapshot but the review already sent to you stays as it was. Reconcile while the meal is fresh; the score you keep is the one that matches what you ate.