▶ Live demo — sueun-dev.github.io/proofplan · runs 100% in your browser, no install, no account.
ProofPlan is a glass-box decision engine for students. When five things all feel urgent, it tells you which one to start — and prints the proof. You compare your real options on evidence, it classifies the choice as a two-way door (decide fast) or a one-way door (slow down), and you leave with a Decision Receipt: the chosen path, the first move for today, and an if-then fallback for the one thing most likely to break it.
No account. No cloud. No black-box AI — every number is visible and editable.
Students rarely fail for lack of a to-do list. They stall because every option feels plausible, the deadline is close, energy is low, and the plan collapses the moment the first obstacle shows up. Generic productivity apps store tasks; they never make you say why one path beats another, or what you'll do when it breaks.
ProofPlan lives in that gap — the distance between a decision and the first real action.
- Frame the decision in one line, with deadline, energy, stakes, and the time you actually have today.
- Compare paths. Score each on Impact, Confidence, Effort, and Reversibility — then attach the evidence that makes it believable (not a vibe).
- Run a premortem. Assume it already failed; name the likely blockers and pre-decide an if-then fix for each.
- ProofPlan returns a ranked recommendation, a door classification, plan checks, and a Decision Receipt — live, as you type.
- Save to a local decision journal, copy a summary, or export the receipt as a PNG.
ProofPlan isn't another planner or AI brain-dump sorter. Its recommendation is built on four named, citable mechanics:
| Mechanic | What it does | Source |
|---|---|---|
| Glass-box weighted matrix | A transparent, editable score — Impact 30%, Confidence 20%, Feasibility 20%, Reversibility 15%, Evidence 15%, plus a deadline nudge. Unlike an AI suggestion, you see and can challenge every number. | Weighted decision matrix · decision hygiene, Kahneman, Sibony & Sunstein, Noise (2021) |
| The door test | Classifies the pick by reversibility. A reversible two-way door → decide fast at ~70% and iterate. An irreversible one-way door → the app changes posture: it stamps "VERIFY FIRST" and tells you to slow down and strengthen the evidence. | Bezos, Amazon shareholder letters |
| Premortem | Step 3 assumes the plan already failed — which surfaces materially more failure modes than asking "what might go wrong?" (prospective hindsight lifts reason-finding ~30%). | Premortem: Klein, HBR (2007); 30% figure: Mitchell, Russo & Pennington (1989) |
| If-then implementation intentions | Each blocker becomes a pre-decided "if X, then Y." A single if-then plan roughly doubles follow-through (one study: gym attendance 39% → 91%). | Gollwitzer & Sheeran (2006), 94-study meta-analysis; Milne, Orbell & Sheeran (2002) |
The output is a compact, screenshot-worthy receipt a judge — or future-you — can read in ten seconds. It changes with the decision: a reversible choice is stamped RECOMMENDED; an irreversible one is stamped VERIFY FIRST with a "slow down" instruction.
| Two-way door → decide fast | One-way door → verify first |
|---|---|
|
 |
A deliberately editorial system — built to look crafted, not templated:
- Type-forward: Fraunces (display serif) over Inter (UI) over IBM Plex Mono (the receipt's "instrument" voice).
- Warm paper + ink + a single electric-indigo accent, with a ruled-grid texture and a thermal-receipt artifact at its heart.
- Result-first: on mobile the Decision Receipt appears before the long form; on desktop it sits in a sticky rail and updates live.
- Tactile details (perforated receipt edge, dashed rules, an ink-style approval stamp), GPU-only micro-interactions, full
prefers-reduced-motionsupport, and no horizontal overflow from 320px up.
It's a fully static app — no build step, no backend.
# from this folder
python3 -m http.server 8027
# then open http://127.0.0.1:8027…or just open index.html directly in a browser.
To export a receipt: load a scenario chip (or Load demo), then click Export receipt PNG and download or screenshot it.
Vanilla HTML · CSS · JavaScript, the Canvas API (the Effort×Payoff map and the exported receipt), localStorage (the decision journal), and Google Fonts (Fraunces · Inter · IBM Plex Mono). Icons are inline SVG. No frameworks, no bundler, no server, no tracking — and it degrades gracefully to system fonts offline.
The brief asks for a practical tool that helps people stay organized, productive, or make better decisions in daily life, with a real way to take action. ProofPlan targets that prompt directly: it helps a student make one better decision under pressure, see exactly why, and walk away with a concrete first move and a fallback.
100% local. Decisions live in your browser's localStorage and never leave the device. There's no account, no analytics, and no network calls except loading web fonts.
Released under the MIT License.