Turning scattered outdoor information into decisions people can inspect, question, and act on.
A permit, a map, and a trail description can each be correct while the trip assembled from them still fails. Wayproof is an open-source attempt to close that gap: connect routes, rules, access, and dated evidence without pretending the missing pieces are known.
Outdoor planning asks people to reconcile agency pages, reservation systems, maps, and trip reports. A road may reach a trailhead on a map without being suitable for a particular car. A permit may cover an entry without covering every proposed stage. A pet allowance can stop at a jurisdiction boundary halfway through the trip.
The product question is practical: What applies to this trip, how do we know, and what must still be checked? Wayproof complements navigation tools. It does not book a campsite, guarantee safety, or turn an old condition report into current availability.
The product personas are provisional, behavior-based hypotheses drawn from bounded public discussions. They guide acceptance scenarios; they are not a representative user study.
These are design targets. Supporting one bounded journey does not mean every persona already has a complete planning experience.
A restriction on the Kings Canyon continuation is a claim about that place. It does not automatically apply to the entire east-side approach. Likewise, the date Wayproof retrieved a page is not the date someone observed a road condition.
Planning adds a second chain: rule → requirement → fulfillment. A reservation or permit can satisfy only part of a party, a date range, or a route. Trip Readiness preserves those gaps; Pre-trip Recheck identifies volatile facts and the sources to revisit.
Unknown is a useful answer. Missing evidence does not mean permission, no fee, an open road, or available water. Conflicting sources remain visible. A nearby campground does not become an alternate trailhead without evidence of access.
Underneath, versioned JSON and Git hold the reviewed knowledge. Shared Python services supply the website, canonical command-line planner, and read-only agent interface (MCP). Keeping the rules in one service boundary reduces disagreement between interfaces. Git-backed storage keeps changes inspectable; an additional database remains a decision to justify with measured need. Explore the evidence model.
The human maintainer owns product direction, acceptance criteria, scope, and merge authorization. Agents assist with bounded source research, implementation, test execution, and reviewable proposals. The architecture makes that division enforceable for knowledge changes.
Code and presentation fixes can use ordinary bounded PRs when knowledge is unchanged. The Taboose repair below is one example: it changed discovery and presentation without inventing new facts. Passing validation means a proposal represents uncertainty correctly, not that the outside world has become certain.
Before committing to a demanding climb, the traveler needs to discover uncertain vehicle access and a pet prohibition on the western continuation into Kings Canyon. The test begins with ordinary search words, not a database identifier or a known detail-page URL.
The internal audit started at the homepage. “Taboose Pass dog” returned zero results. Searching just “Taboose” found the destination, but the pass page omitted the west-side pet warning. The road route showed zero planning facts even though scoped evidence existed. The map required visual hunting.
The deployed repair supports searches combining a place and a concern, including dog/pet synonyms. Pass and route pages surface scoped restrictions and road uncertainty before ordinary facts. Evidence links expose sources and dates. Map lookup and warning previews preserve that context.
The repair follows explicit relationships and spatial scopes. Connected-place warnings are labeled separately from facts about the selected place. An adjacent jurisdiction can deserve a warning without becoming an inferred traversable route.
Review also caught a map-loading defect: controls could be used before handlers or layers were ready. They now wait for readiness, while text links remain available. The audit’s fixture measurement reduced initial map HTML from 644,076 to 4,631 bytes by loading a compact search index on demand and limiting displayed results. That measures the initial document, not total map transfer or mobile latency; the large geometry payload was unchanged.
The October 7 internal production acceptance found the restriction, jurisdiction scope, provenance, and unknowns discoverable. It did not establish present road suitability, legal fallback parking, added approach distance or elevation, a precise turnaround, or a supported dog-compliant alternative. It was not an external usability study.
The useful answer is to identify the western restriction early and leave the unresolved access and itinerary questions visible. A clearer warning is a product improvement, not proof that the trip can proceed. See the acceptance audit and follow-up.
Canonical search; generated park, trail, camping, peak and pass directories; entity and evidence pages; published change history; focused Del Valle and Ohlone guides; maps from reviewed geometry; and the Taboose discovery and warning repair. Pages expose gaps and contextual rechecks alongside the facts.
Search the published knowledge · Inspect its history · Explore the map
The Schema v0 foundation, controlled knowledge publication, bounded intent resolution, requirements and fulfillment coverage, Trip Readiness, Pre-trip Recheck, a canonical CLI mode, and a read-only MCP adapter. Costs and deadlines receive bounded evaluation; published inventory is distinct from live availability.
An additive proposal service and a reviewed reference-only research path also exist. Media analysis is a synthetic, process-local rehearsal. These are engineering capabilities with defined limits, not a deployed end-to-end browser trip planner or a public media-upload service.
Richer intent and constraint resolution; live inventory and closure integrations; quantity and option selection; completion of the legacy CLI migration; a constrained MCP proposal adapter; broader ingestion and operational freshness scheduling. Human-readable browse, search, and detail experiences still need refinement and broader persona testing.
An installable offline web app (PWA), offline planning and maps, and foreground GPS are planned. A native iOS app and local agent are a later possibility, not a commitment. Real media processing needs a separate activation review.
M0, the trusted foundation, is complete. M1–M3 remain in progress; coverage is expanding, operational scale is unfinished, and M5 offline work is planned. The roadmap snapshot records the boundaries. This case study does not turn a milestone’s target outcome into a shipped feature.
The next product challenge is to make evidence easier to use without hiding its limits: shorter decision paths, clearer language, and broader testing against the people the product is meant to serve.
The repository links below pin the documents used for this dated account. Live product pages and the current roadmap may change after this snapshot.