Design a logging palette that earns its keep

Every failed logging scheme I have watched die went the same way: designed in a meeting, twenty fields wide, abandoned by clip thirty. The schema that survives is the one you wrote for yourself, sized to decisions you actually make while footage plays. A palette earns its keep when every field pays rent on every clip.

Start with three fields you decide every clip

The test for a day-one field is blunt: while the clip plays, you can make this call in a couple of seconds, every single time. For a game day that is usually take (keep or junk), action (what play this is), and who is in it. That is the whole starting palette. Lens, weather, mood, sponsor visibility: real fields, someday, for some project, but on day one they are the fields you fill twelve times, feel organized about, and quietly abandon, taking your momentum down with them. Three fields filled across 400 clips beat twenty fields filled across thirty, and it is not close. If a field is on the bubble, watch yourself log ten clips and count how many times you filled it. The footage votes; count the ballots.

Single-select unless values are simultaneous

Model the question, not the interface. A one-answer question gets a single-select: the play was a run or it was a pass, and the moment "both" feels true you have discovered two fields. Multi-select is for facts that are plural, like everyone visible in the frame. The discipline matters because a controlled vocabulary is what keeps search alive. Left as free text, "TD", "td", and "touchdown" become three different searches forever, and by November your season has dialects. A controlled list, with the synonym declared once, makes them one thing every filter can find. Keep the lists short while you are at it: ten options is a usable menu, forty is a second job. And when an answer will not fit the list, that is what a free-text note is for; do not widen a controlled field to absorb exceptions.

Write definitions a model can act on

A field definition is not documentation. It is an instruction, written for the two readers who will actually obey it: the next human who logs for you, and the AI. "Action = the football play type, never camera movement" is one line, and it stops both of them from tagging a whip pan as a play. In ClipLogger that line is wired straight into the model's prompt, and closed fields force the model to answer inside your options, so a sharper definition comes back as better suggestions on the next pass. If you cannot write the instruction in one line, the field is not a decision yet. It is a vibe, and vibes log badly. The definitions also outlive you: when a second logger joins mid-season, the palette is the training document, and it is already written.

Put the fields on keys

Logging at watch speed means the mouse never gets touched. Bind values to keys and let modifier layers multiply the keyboard: the base layer stamps takes, one layer holds actions, another resolves people. The best version is code entry against a lookup table: hold the modifier, type 72, and the roster resolves it to the athlete's name while the play is still running. A field you can stamp without looking down gets filled on every clip; a field that needs a dropdown gets filled until the first coffee. The mappings live in the palette too, so the scheme travels with the project instead of living in one logger's muscle memory. Watch someone log with their palette on keys and it looks like typing, not tagging; the clip never pauses.

Grow the vocabulary from real footage

Plan three fields, then let footage argue for the rest. Add an option the moment a clip demands one and the demand repeats; add a whole field only when you catch yourself wanting to record the same extra decision for the third time. The palette after one real game day beats the one from a week of whiteboarding, because footage is opinionated and meetings are not. This holds in any tool, spreadsheets included: schema design is not a kickoff deliverable, it is a byproduct of logging honestly and noticing what you reach for. The twenty-field template feels like preparation. The three-field palette that grew to seven by October is the one that describes your footage.

A field you do not decide on every clip is clutter with a label.