Log how you feel before starting

Tuesday morning. The plan says heavy squats but you slept poorly and the right knee aches. Before you start, you tell the agent how you actually feel — a number, a word, a few specifics. It rides with the session and shapes what the weekly review can say later.

What the agent needs to hear

Behind your sentence, the agent works backwards to two pieces: a general state (a 1–5, a word, a short phrase) and any specific signals worth flagging (sleep, soreness, motivation, pain — all optional). « 3/5, slept badly » carries the general state and one signal; that’s already enough.

The assessment attaches to the session you’re about to start. If you’re already mid-warm-up and the session is open, it attaches to the active one instead — same log, just hooked to the right row. You don’t have to name the session; the agent picks the one in scope.

How to phrase the state

Three registers work. Numeric — « 3/5 », « 4/10 » — is fastest and reads cleanly on a trend chart. Verbal — « feeling rough », « solid », « about average » — gets mapped to a numeric internally, so the trend still works; the agent confirms the mapping if it’s ambiguous.

Multi-signal — « energy 4, sleep 2, motivation 5 » — gets separate dimensions tracked independently, when you offer them. « 3/5, knee stiff » is one rating plus one note; the agent doesn’t invent dimensions you didn’t mention.

Skipping is the default. No log means no signal — not a flagged absence, not a nudge in the review.

What this signal is for downstream

Pre-workout assessments are the user’s input into the readiness model. The recovery-readiness read pulls from these alongside workout load and streak data — one of three streams when you log them, silent when you don’t.

A single log is noise. Consistent logging across weeks builds a baseline the agent can detect deviations from: « your average pre-workout energy this month is 4; today’s 2 is the lowest in three weeks ». The weekly review uses the same baseline to connect a bad-readiness day to a poor-performance session, when both are logged on the same day.

Skipping is fine — but it limits what the system can correlate later. The pattern of « knee stiff » three times in a month is only visible if each one made it in.

When the agent gets it wrong

Three patterns recur. The agent treats your input as a post-workout assessment — same fields, wrong half of the session. Push back: « that’s pre-workout, before the session — not the closing read ». The agent reattaches it.

The rating gets bolted onto yesterday’s leftover open session because you forgot to close it last night. « Attach to the one I’m about to start » settles it; the agent opens a new session and the assessment rides with it.

A specific signal gets dropped. You said « energy 3, knee stiff » and only the energy rating came back. « Add the knee note » fixes it — the agent appends, doesn’t overwrite.

What makes the log worth keeping

Three things decide whether this pre-workout log informs your week usefully: the rating is honest (a 3 you logged as a 5 to feel motivated corrupts both the baseline and the post-hoc correlation with performance), specific signals are surfaced when they matter (« knee stiff » today plus « knee tweaked » three weeks later becomes a pattern; alone it’s just one note), and the log attaches to the right session (pre-workout, before the warm-up — not bolted onto a session that’s already half-done). Logged consistently, these pre-workout reads are what make the readiness model and the weekly review legible instead of guesswork.