Travel Intelligence is one build old.

A deterministic, client-side scoring model: every candidate itinerary is scored 0–100 on Cost, Comfort, Upgrade Potential, Award Value, Flexibility, and Risk, weighted by your ranked priority list. Outputs an Executive Summary, Best Recommendation, Top Five Alternatives comparison table, Upgrade Strategy, Award Strategy (with a genuinely-strong redemption bar, not a blanket "use your miles"), Purchase Timing, and 5–8 input-specific Travel Advisor Insights.

Every recommendation explains its own weighting instead of hiding behind a single score.

• JUL 2026Full Input Card built: trip, flight preferences, ranked optimization priority, credit cards, 13 airline loyalty programs, and airport flexibility — matching the fields specified in the original agent brief.
• JUL 2026Hotel loyalty programs removed from the Input Card for now — the scoring engine wasn't using them, and a checkbox that doesn't affect the result is worse than no checkbox. Revisit once hotel signal actually feeds the model.
• JUL 2026Pricing model reworked after an early version clustered nearly every Business-class result at the same floor price regardless of carrier or route — replaced with per-cabin base prices, seeded variation, and cabin-appropriate floors so the spread looks like a real fare set rather than a repeated number.
• JUL 2026House style adopted from the sibling portfolio: cream/serif shell, persistent nav, pill-nav, phase/step page patterns — matching thefitnesslog.org and streamingscout.org rather than a one-off look.

Index (the optimizer itself), About, Guide, and this Roadmap — a static, no-build site, deployed to Netlify.

Something to open and run a real trip through, without opening a Cowork session first.

• JUL 2026Deployed to a new Netlify site. Custom domain connected shortly after — see phase 06.

A GitHub repository matching the other two sites' continuous-deployment governance model — push to main, Netlify auto-deploys. This build was handed off as files to push from your own machine rather than pushed directly from this cloud session, since no GitHub credential is available here.

Parity with the other two sites' deploy pipeline, and a real commit history for this project.

An installable version that adds itself to the Home Screen: its own icon, no browser chrome. Same site, no separate codebase, no App Store submission — a web app manifest and an icon set, nothing more.

One tap from the Home Screen before a trip, instead of finding a browser tab.

• JUL 2026Add to Home Screen shipped: a web app manifest plus icon set (192px, 512px, and an Apple touch icon) makes the site installable with its own icon and no address bar, on iOS and Android alike.

The single biggest gap today. Wiring in real data sources — Google Flights/ITA-style fare search, seats.aero-style award availability, live seat maps and upgrade fare rules — behind Netlify Functions, matching the backend pattern Vinyl Scout and The Fitness Log already run. This is what turns the reference-range planning model into an actual live search.

The recommendation stops being a documented estimate and becomes a real, bookable answer.

A domain matching vinylscout.org, thefitnesslog.org, and streamingscout.org. Registering one required entering payment, a decision left to Susan and made by Susan directly, not taken on the agent's behalf.

One address to remember, not a netlify.app subdomain.

• JUL 2026travelintelligence.org registered and connected via Netlify DNS. TLS certificate provisioning is automatic and typically completes within a few minutes of a new domain being connected.

Remembering preferred airlines, seats, hotels, upgrade successes, disliked airports, connection tolerance, and typical travel style across sessions, so later recommendations get more personalized instead of starting from the same defaults every time — matching the North Star in the original agent brief.

The agent gets better at optimizing for you specifically, trip over trip.