-
-
Mission control: the queue, backward-planned start-by times, and live KPIs
-
The computed week — painted free hours (left) and the backward-planned schedule (right)
-
Energy matching — foggy 9 PM re-ranks the queue to ten-minute wins
-
Evidence-verified capture — every task quotes your exact words: no citation, no task
-
The honesty loop — pace multipliers tuned by the time work actually took
-
Focus sprints sized to your energy, hosted by a zen capybara
🚀 Inspiration
Every planner we've ever used has the same fatal flaw: they all tell you when it's due — none of them tell you when to start. Meanwhile every AI todo-app floods your list with tasks it hallucinated from your syllabus, and you spend more time babysitting the AI than doing the work. We're busy students. We needed mission control, not another spreadsheet with deadlines screaming at us from a red column.
⚡ What it does
Momentum is a single-file productivity command center built around a real scheduling engine:
- Backward planning — add tasks with deadlines, estimates and work type; paint the hours you're actually free on a weekly grid. The engine plans backward from every deadline into your free blocks, as late as safely possible, with durations adjusted by your real pace. Earlier hours stay open for later-deadline work.
- Energy-matched re-ranking — the queue re-ranks against your current energy. A foggy 9 PM surfaces ten-minute wins; a sharp Saturday morning unlocks deep work. Same tasks, different brain, different plan.
- Evidence-verified capture — paste a syllabus or a chaotic brain-dump; every extracted task must quote your exact words back at you (the gold receipt): no citation, no task. Unverified items are flagged and dropped.
- The honesty loop — finishing tasks with the real time taken tunes per-category pace multipliers. A 30-minute essay takes 45? Momentum learns, and the plan bends toward reality instead of your optimism.
- Focus sprints + a capybara — a timer sized to your energy (10/25/50 min), a slice-it-smaller button, and a zen capybara in the corner keeping the mission-control vibe warm. 🦫
- .ics export — the whole plan exports as a calendar file with start-by times.
🔧 How we built it
One self-contained HTML file — zero dependencies, no build step, no backend, no account, works fully offline in any modern browser. The scheduling engine is a deterministic backward-planner (LST-safe, earliest-open-block reservation) written from scratch in plain JavaScript. The capture pipeline runs a rule-based extractor (dates, subjects, durations, evidence spans) fully offline; an optional OpenAI-compatible key adds a model path that never leaves the browser. State lives in localStorage on this device only — no server, no tracking. The UI is an industrial mission-control design: warm paper + deep ink, burnt-orange reserved for actions and live readouts, Fraunces / DM Sans / DM Mono (tabular numerals so digits never shift as they tick), WCAG-AA light and dark themes, responsive from 375 px phones to desktop, full keyboard navigation, reduced-motion respected.
🧗 Challenges we ran into
Three hard ones. (1) The scheduler: ordering forward from “now” produced terrible plans (everything crammed early); solving backward-from-deadline with slack-aware reservation took several iterations to get provably right. (2) Trust in AI extraction: we refused to ship “magic AI tasks” — the evidence rule (every task must quote your exact words, or be dropped) was our answer to hallucinated tasks, and building receipts you can visually verify was harder than a naive parser. (3) Making a single file feel like designed software: we iterated the type system (tabular mono digits), focus states, and both themes until it held up on a phone and on a projector.
🏆 Accomplishments we're proud of
- It actually computes: same input → same plan, a real algorithm not a list with dates.
- The evidence rule: no-citation-no-task is a genuinely new take on trustworthy AI capture — receipts, not vibes.
- Zero dependencies, fully offline, privacy-first: your week never leaves your device.
- It's usable at 9 PM when you're fried — the state most planners ignore.
- A capybara that meditates with you. (The organizers asked for one. We delivered.)
📖 What we learned
Backward planning beats forward cramming — structurally, not just aesthetically. That “small tasks first” is wrong when you have deep-work energy available — context decides priority. That trust in AI tools is built by showing evidence, not by being confidently fast. And that a single HTML file, treated with real engineering discipline (determinism, WCAG-AA, keyboard paths), can feel like shipped software.
🎯 What's next for Momentum
Momentum was built for the Triple H 2026 productivity theme, and it's ready to use today. Next: calendar round-trip (two-way .ics sync), pace multipliers with confidence bands as data accumulates, shared team week-grids, and an optional end-to-end-encrypted sync so your week still never leaves your control.