Inspiration
Investment tools often optimize for more data, more alerts and faster execution. The real everyday problem is often behavioral: a person needs to understand the next decision, test it against their own rules and record why it was made. PEA Pilot makes disciplined investing visible and repeatable without encouraging more trading.
What it does
PEA Pilot reads a local portfolio and turns it into one prioritized, explainable decision brief. It surfaces a protected cash floor, concentration risk and critical alerts; lets the user simulate an allocation without placing an order; and reveals the maximum rule-compatible amount when a scenario fails.
Its Evidence Gate asks where a signal came from, validates the freshness of a dated HTTPS source, requires a counter-case and adds a deliberate pause for emotionally loaded or social signals. The user can export a timestamped JSON decision receipt whose noExecution flag is always true. Decision Memory tracks verified evidence, slowed decisions and respected guardrails without rewarding trading frequency.
An English Judge Mode and a five-step guided tour make the complete product testable in 90 seconds. A privacy switch masks amounts instantly. The interface supports dark mode, keyboard navigation, reduced motion, responsive layouts, offline PWA behavior and local-first storage. The public build contains only fictional data.
How we built it
The starting point was a pre-existing local PEA portfolio tracker. During Build Week, Codex powered by GPT-5.6 was used to inspect the baseline, research the official requirements, isolate a privacy-safe competition edition, design and implement the decision cockpit, add the simulation and evidence flows, improve accessibility, create automated tests, inspect desktop and mobile renders, audit the public package and prepare the submission evidence.
The human owner retained the key product decisions: the protected-cash rule, the decision not to execute trades, the scope of the fictional demo and the privacy boundary. The dated baseline and Build Week change log clearly distinguish prior work from the competition extension.
Meaningful Build Week extensions
- Explainable next-decision cockpit and discipline score.
- Rule-aware counterfactual simulator with no trade endpoint.
- Research-grounded Evidence Gate and exportable JSON receipt.
- Source-freshness validation and deliberate social-signal pause.
- Local Decision Memory with no engagement or trade-frequency incentives.
- English Judge Mode and a five-step, 90-second tour.
- Privacy mask, dark mode, keyboard tabs, focus states and reduced-motion support.
- Standalone fictional dataset and static-host fallback.
- Automated accounting, safety and experience tests.
- Reproducible public package, presentation and narrated demo.
Challenges
The largest constraint was transforming an existing personal tool without exposing private portfolio data. We preserved the original application, created an isolated competition edition and made fictional local data the default. Another challenge was reducing a dense multi-tab tracker to a coherent product story. The new cockpit leads with one decision and keeps every rule and piece of evidence inspectable.
Accomplishments
The project works without a bank login, cloud account, paid API or private dataset. The same build runs through the local Python rules engine or a privacy-safe static snapshot. It makes its limits explicit: it is a discipline aid, not financial advice, and it never places an order. Ten automated tests, a complete browser tour, public-package privacy scans and a real GitHub Pages deployment all pass.
What we learned
Explainability can be a product interaction rather than a disclaimer. A responsible financial assistant should show which rule fired, what changes in a scenario, which boundary would be crossed and what evidence could invalidate the decision. It should help the user slow down instead of optimizing for engagement or transaction frequency.
What's next
Next steps include user-defined strategy rules, encrypted multi-device synchronization, J+7 and J+30 decision reviews, and an optional GPT-5.6 Socratic reviewer that challenges evidence without producing a buy or sell recommendation.
Try it
Open the public Judge Mode and select Start 90s tour. The source repository includes the dated baseline evidence, Build Week change log, tests and complete judging instructions.
Log in or sign up for Devpost to join the conversation.