Source

• The footage app reports every action to PostHog (search, add to bin, download, rating, abandon). search-quality-job copies that stream every 6 hours and labels each page-1 search; the search is also matched to our own trace for latency and mode.
• Everything a person does after a search, until their next search in the same brand (or 30 minutes), counts for that search.
• Response is the app's own search timer: from the moment it asked until the answer was parsed, before anything was drawn. First result is when the first card on the page had something on screen (a video frame at the scene start, a still, a finished audio or project tile), on the same clock; the app reports it as footage_search_first_result, and footage_search_rendered once every card settled. Searches made before the app reported these show '–'.

Searches and attempts

• A search is one thing an editor looked for. It is made of one or more attempts: related queries, each within 10 minutes of the previous.
• An attempt is good when the editor added a clip to a bin, downloaded one, or rated it helpful / great; otherwise it is bad (they rephrased, the app recorded an abandon, there were no results, an error, a negative rating — which beats an add — or nothing happened).
• A search is a success (found) on the attempt that was good, or the editor gave up; the two add up to 100% of searches. The chart splits successes by the attempt they came on (1st, 2nd, 3rd, 4th+).
• On each attempt one of three things happens: success, continued (bad, and the editor went on to a related query) or gave up (bad, and they did not). The three add up to 100% of that attempt.
• Gave up is where agentic search would land: the editor kept rephrasing and left without footage.

Which searches

• Only semantic attempts count by default. An attempt is set aside when the engine routed it as a job code, an exact file / asset name or a SKU, when the query reads like one (FCTOR260701-T10082, IMG_4151.mp4, A001_C016), when it is a known talent's name, or when it is empty (a browse).

Who

• Editors (the default) = everyone not seen as a lead, admin, QC, strategist or vfx: the app's editor role, internal (its tag for staff with no privileged role — an editor who never rated a search lands here, and so does a strategist who never rated; our data cannot tell them apart) and unknown (no role reported). Leads and admins search to review, and a strategist's success is a copied link we never see.

Period

• Periods are whole weeks of full UTC days ending at today's midnight (4 weeks by default), so every week in a period is complete. All numbers are totals for the period. The two charts show a bar per day for 1- and 2-week periods and a bar per week for longer ones (successes by the attempt they came on, over gave up, with the success %); the weeks slide with the period (the first starts on the period's first day) and are not calendar weeks.

Colors in the tables

• Percentages (success, success 1st try in the tables) are colored on fixed thresholds: green over 50%, amber from 25% to 50%, red below 25%. They are absolute, not relative to the average. Counts are not colored.

Caveats

• A brand the alias table does not know shows as brand:xxxxxxxx. Adds made later from a collection are not attributed to a search. Rows are re-labelled for 3 days as late signals arrive.