Fix a weight or measurement entry

You logged yesterday's weight at 78.3, then re-stepped on the scale because the number felt wrong, and it read 77.8. The first reading was a bad pad placement. The chart now has a phantom bump until you fix it.

What the agent needs to hear

Behind your sentence, the agent works backwards to three pieces: which entry (weight or perimeter, and from which date), what’s wrong with it, and the right value. Today is the implicit date. The type — weight vs perimeter — comes from your phrasing: « weight » or a number alone reads as weight; a site name (« waist », « biceps », « chest ») reads as the perimeter at that site.

You can point at an entry by date (« yesterday’s weight »), by content (« the 78.3 weight reading »), or by combination (« last Sunday’s waist measurement »). The agent finds the matching log, shows you a card with the old and new values side by side, and waits for a nod before applying.

What you can change

The value is the most common edit. « 77.8, not 78.3 » updates the number and recomputes the trend curve on the spot. Unit corrections work the same: « 38 cm, not inches » flips the unit and re-scales. The date can change too — if you logged « today » but the reading was actually from yesterday morning, « that was yesterday, not today » moves the entry to the right day.

For perimeters, the site label is editable: « that was biceps, not forearm » re-attaches the value to the correct site and updates both site trends (the wrong site loses its phantom reading; the right site gains the real one).

If the entry never should have existed — you stepped on the scale fully clothed by accident, the tape slipped during measurement and you re-measured — deletion is the right move: « delete yesterday’s weight, bad reading ». The agent confirms once and removes it.

When the entry is missing entirely

If you forgot to log a measurement and want to add it after the fact, that’s not a correction — it’s a new log with an explicit date: « log Sunday’s weight, 77.5 fasted morning ». The new entry slots into the timeline and the trend chart re-draws.

If you logged at the wrong site entirely (you logged « waist 86 » but meant « hip 86 »), the cleanest move is to delete the wrong entry and log the right one, rather than editing the site label — the latter works, but the audit trail reads cleaner with a delete + re-log.

When the agent gets it wrong

If the card pulled the wrong entry (yesterday’s reading instead of the day before’s), say so: « no, two days ago — Saturday’s weight, not yesterday’s ». If the agent isolated the right entry but wants to change the wrong field (you said the value should be 77.8 and the agent is offering to change the unit instead), call the field: « 77.8 is the new weight, not the unit ».

If multiple matching entries exist (you logged the same weight value twice on the same day by accident), point with the most distinguishing piece: time, sequence, or « delete the duplicate ».

What makes the correction worth doing

Three things decide whether this correction lands cleanly: the pointer is unambiguous (so the agent edits exactly one entry, not all matching ones), the new value is the truth (a corrected reading, not a number you wish you’d hit), and the change applies only to the field you meant (value, unit, date, or site — not all four by accident). The system reads corrected logs the same way as originals — trend charts, longitudinal narrative, plan adherence all re-read from the new number — so a clean fix restores the curve, while a sloppy fix corrupts it twice over. Say what’s off, point at the one entry, confirm the card.