Log a meal exactly as the plan says

It's eight in the morning. You ate the breakfast your plan asked for — same items, same quantities. The interaction shouldn't be a conversation; it should be one sentence. The plan already knows what you were supposed to eat, so the agent just needs you to confirm you ate it.

What the agent needs to hear

Behind your sentence, the agent works backwards to two pieces: which meal slot (breakfast, lunch, dinner, snack) and whether anything deviated from what the plan asked for. Time of day infers the slot most mornings; the default for the second piece is « nothing deviated, log it as planned ». Active plan and today’s date are implicit.

That’s why a clean morning takes one sentence. « Logged breakfast — same as the plan » carries the slot, the date, the items, and the quantities — all pulled from the plan’s day. When anything is ambiguous, the agent renders a preview card. On a clean call, it skips straight to saved.

When part of the meal deviated

A pure on-plan log takes the plan’s defaults wholesale: confirm, saved. A partial deviation names what shifted (« had 50 g oats instead of 40 ») — the agent moves that one item, keeps the rest at planned quantities, and tags the meal as partial-deviation rather than as-planned.

Bigger deviations work the same way, with sharper phrasing. « Had eggs instead of oats » swaps an item; « no fruit this morning » drops one; « also had a coffee, 200 ml black » adds one. When the swap crosses a threshold — half the slot rewritten, or macros shift materially — the card reappears so you confirm before saving.

How the plan-log shapes adherence

A clean plan-log counts as planned execution: slot checked off, macros line up, streak advances. A partial-deviation log still checks the slot off but adherence scoring weights it lower — the system knows part of the meal drifted. A heavily-deviated log fills the slot but reads near-zero adherence for that meal, because what you ate stopped resembling what was queued.

This three-tier reading is what makes plan-following legible in the weekly review. A grid of slot ticks looks the same across clean and partial weeks; the adherence chart tells them apart. Honest tagging at log time is what keeps the two views in sync.

When the agent gets it wrong

Wrong slot is the most common miss — « logged breakfast » at 11 am can land on the breakfast slot when you meant the late-morning snack. Correct it in plain language: « that’s the snack, not breakfast ». The slot moves and the items re-resolve.

Missed deviations come next. You said « 50 g oats instead of 40 » and the card shows 40 — the agent treated your aside as commentary. « No, the oats were 50 g » nudges the quantity onto the right row. The third miss is shape: your sentence got logged as a freeform meal instead of a plan-log. « Log it from today’s plan, not as a new meal » switches the call.

What makes the fast-path worth using

Three things decide whether logging from plan beats logging in plain language: the plan’s slot is the meal you actually ate (deviations small enough to declare, not a full off-plan meal pretending to be on-plan), deviations are named, not hidden (a partial log with the right items reads honestly; an « as planned » log with secretly different food corrupts adherence), and the slot is the right one (breakfast at 11 am is still breakfast if that’s the slot you ate; lunch at 11 am is a slot mismatch). The system uses plan-logs as the cleanest signal of adherence; honest tagging of deviations is what keeps the signal real. Speak the meal the way you ate it — the plan does the rest.