# Year Reel > Turn a plain dated log of your own year - runs, books, gigs, films, classes, anything - into a year-in-review reel of vertical story cards. The page counts the facts for free; a metered build writes the cards from those facts only, and the page re-checks every number. URL: https://year-reel.skillsafe.ai/ API: https://year-reel.skillsafe.ai/api.html Built on: the SkillSafe skill @iart-ai/wrapped-video (https://skillsafe.ai/skill/@iart-ai/wrapped-video) - its year-in-review scene grammar: an opening card, big-number reveals, top-five lists, superlatives, a year-type card and a closing card made for sharing, framed as vertical 9:16 stories. Model: gpt-terra, metered per run. Counting, the draft reel, card images and exports are free. ## What the page does for free - Reads one entry per line. Dates: 2026-03-14, Mar 14, 14 Mar, 03/14, 03-14-2026, optionally after a weekday (month first unless the first number is over 12). A line holding only a month name is a header for the undated lines under it. - Pulls #tags and amounts with a unit (km, mi, pages, min, h, kg, lb, laps, reps, episodes, songs) or a currency sign ($, pounds, euros). - Counts numbered facts (F1, F2 ...): entries, months with an entry, the busiest and quietest months (ties named), different things logged, totals per unit, the biggest single entry, the most repeated items, top tags, the longest run of consecutive days, the longest gap, the most common weekday, the first and last entry. - Gives the reel length by fixed rules, in order: fewer than 6 dated entries = needs more log (3 cards plus what to add); fewer than 20 entries or fewer than 4 months = short reel (5 cards); otherwise full reel (8 cards). - Builds a draft reel from the facts alone, plays it as tap-through story cards, and saves any card as a 1080x1920 PNG drawn in the browser. ## What the paid run returns One JSON object: task "reel", reel (full | short | too-thin), title, persona {name, why} or null, cards [{type, kicker, headline, big, unit, list, caption, facts}], asks (only for too-thin), share_caption. Card types: intro, big-number, top-list, superlative, month, persona, outro. ## Grounding Every digit on a card must come from the counted facts or the log; the model never counts or computes new numbers, never guesses pronouns or details about the person, and cites the fact ids each card uses. The page flags any number that is in neither, any big number its cited fact does not hold, and any length that differs from the count. ## Exports Card images (.png), the card script (.txt), the reel plus facts (.md), everything (.json), the share caption.