FloodTruth AR — Phase I Capability Map

Every shipped feature mapped to its NSF SBIR Phase I objective. Reviewer self-serve.
Project Pitch: /field/ar-prototype.html · Tap-to-mark: /field/ar-tap-to-mark-prototype.html · Cross-section: /field/ar-cross-section.html

This page shows what's already built and deployed today vs. what Phase I funds. Phase I scope per the NSF Project Pitch is engineering-grade calibration + Bayesian integration + closed-loop refinement + EAP exhibit generation + Helene-area pilot deployment. The shipped prototype is the verifiable starting point: it is publicly deployed with no login, so every feature listed below can be exercised right now, without any funding committed.

SHIPPED deployed and public today PARTIAL demonstrated; Phase I completes PHASE I in scope, not yet built PHASE II deferred

Obj 1 AR Coordinate-Stack Engineering Validation

Success metric: <2 m horizontal positional error at 100 m range under typical field conditions, validated on ≥3 Helene-affected dams.
SHIPPED
WGS84 → ENU → camera-space → screen pinhole projection
Real coordinate stack, not a placeholder. Per-vertex projection every frame at 60 fps.
ar-prototype.html · geoToEnu, buildCamRotFromHPR, pinholeProject
SHIPPED
WMM-2025 magnetic declination correction (Android)
Compass heading is magnetic on most Android sensors; declination fetched from BGS WMM-2025 service, applied with WMM-2020 fallback. iOS uses true heading directly.
ar-prototype.html · fetchMagDeclination
SHIPPED
DSM-based per-vertex occlusion (USGS 3DEP + GeoTIFF ingest)
Terrain occluder built from USGS 3DEP point + grid queries (free public API) or user-uploaded GeoTIFF. Reprojection from any WGS84 UTM zone (EPSG 326xx/327xx) and NAD83 UTM zones (EPSG 269xx). Per-vertex ray-march occlusion, not bounding-box.
ar-prototype.html · loadDsmGeoTiff, buildTerrainDem, isVertexOccluded
SHIPPED
Edge-silhouette clipping via 10-iteration bisection
Polygons partially occluded by terrain are clipped against the ridge silhouette using midpoint bisection in absolute NAVD88 elevation. Rebuilds the visible ring against the crest instead of dashing the full outline.
ar-prototype.html · silhouetteT, clipRingToSilhouette
PARTIAL
Survey-grade RTK calibration against Helene-affected dams
The AR pipeline is deployed and demoable; field validation against RTK GNSS + LiDAR ground truth is the Phase I deliverable. Validation budget covers $40K for RTK rental, USGS 3DEP-primary LiDAR strategy with UAV LiDAR contingency, site visits to 3 sites, and a 4th opportunistic site if a strong partner emerges.
Budget detail: BUDGET_JUSTIFICATION_v0.md §3

Obj 2 HEC-RAS Integration

Success: AR-rendered shoreline cross-validated against HEC-RAS-modeled shoreline and against documented post-Helene high-water marks on Helene-affected dams.
SHIPPED
GeoJSON inundation polygon ingest (browser-side)
Accepts FeatureCollection, Feature, bare Polygon, or MultiPolygon. Recognized properties: depth_ft, depth_m, wsel_ft, wsel_m, WSEL, return_period_yr, depth_offset_ft. Storm × WSEL offset scenario filter works on any data with these properties.
ar-prototype.html · parseGeoJSON, setFeatures
SHIPPED
FEMA NFHL fetch + zone-aware overlay + BFE display
Auto-fetch from FEMA's public ArcGIS REST endpoint within configurable radius of observer. Zone-aware color coding (SFHA inland, SFHA coastal, Zone X shaded/unshaded, Zone D). Reticle HUD reports zone + BFE in NAVD88 ft at look-at point or observer.
ar-prototype.html · fetchNfhlAround, classifyNfhlZone, nfhlZoneAt
PHASE I
HEC-RAS adapter (RAS Mapper GeoJSON MVP + shapefile + HDF5 roadmap)
Phase I MVP path: browser-side RAS Mapper GeoJSON parse with auto-detect of common CRS (UTM 17N/16N, NC/TN State Plane). Tier 2: shapefile via shpjs. Tier 3: HDF5 server-side, deferred to Phase II. Validation harness with Jaccard / Hausdorff / depth-RMSE / datum-offset reports.
Architecture: HEC-RAS adapter design documented in the Phase I proposal (available to reviewers on request)

Obj 3 Bayesian Breach-Parameter Posterior

Success: posterior inundation envelopes produced from hierarchical Dirichlet-weighted mixture of Froehlich, Xu-Zhang, Wahl breach-geometry likelihoods. Calibrated uncertainty interval delivered with every envelope.
SHIPPED
Python bayesian_breach package (10/10 tests pass)
Five modules: regression_likelihoods (Froehlich 1995/2008, Xu-Zhang 2009 simplified categorical, Wahl 2014 spreads), mixture_sampler (Dirichlet-weighted Monte Carlo), breach_hydrograph (broad-crested weir + reservoir routing via scipy.solve_ivp), aggregator (variance decomposition + tail gate + LCB), posterior_pipeline (couples breach hydrograph → floodline_port solver → posterior).
Source: /SBIR-EAP/bayesian_breach/ (7 unit + 3 integration tests passing)
PHASE I
Browser ↔ server integration (cache-first posterior delivery)
Bayesian inference takes 30-150 s per dam — not interactive. Phase I architecture: precompute-at-enrollment for enrolled dams, cache in Supabase, serve via Fly.io endpoint with auth gate + per-user monthly cost cap (per feedback_paid_api_gating rule). Browser fetches cached GeoJSON envelope and renders through existing AR pipeline. 202+poll fallback for cache miss.
Architecture: Bayesian breach-posterior service design documented in the Phase I proposal (available to reviewers on request)

Obj 4 Closed-Loop Field-Validation Workflow + EAP Exhibit Generation

Success: inspector workflow that tap-marks ground-truth discrepancies, refines breach posterior in real time, and produces every AR exhibit required by an EAP in a single session.
SHIPPED
Tap-to-mark prototype (Discrepancy mode)
Four-class discrepancy classifier (AR-wet-observed-dry, AR-dry-observed-wet, uncertain, agree). Inverse pinhole projection back to world coordinates. localStorage persistence. GeoJSON export with full observer-pose provenance.
ar-tap-to-mark-prototype.html · screenToWorldGround
SHIPPED
High-Water Mark mode + photo capture + FEMA HWM survey package
Six evidence types (debris/mud/vegetation/water-stain/silt/other) × three confidence levels. Per-mark evidence photo to IndexedDB. Export packages a ZIP with hwm_points.geojson + hwm_summary.csv (20 columns) + per-mark photos + README cross-walking fields to USGS Short-Term Network (STN) equivalents.
ar-tap-to-mark-prototype.html · setMode('hwm'), FEMA HWM package builder
SHIPPED
EAP exhibit batch export (Field + Simulated modes)
Ingest EAP point list (CSV / GeoJSON / manual). Field mode: overlay capture prompt at each point during a live AR session. Simulated mode: render each exhibit at point-correct projection over synthetic gradient for desk authoring. ZIP includes per-point PNG with metadata caption strip, manifest.csv (15 cols), manifest.geojson, README, and optional jsPDF composite.
ar-prototype.html · Panel 5 (EAP exhibit batch)
SHIPPED
Cross-section profile tool (observer → tap, ground + WSEL + crossings)
Leaflet map for tap point selection, Chart.js v4 ground/WSEL profile, per-segment crossings table (ENTER/EXIT of inundation along transect — the canonical evac-route output). CSV + PNG export. Right-click in AR view hands off observer + look-at + GeoJSON via sessionStorage.
ar-cross-section.html · also ar-prototype.html handoffToCrossSection
PHASE I
Bayesian back-feedback from tap-marks into breach posterior
The tap-to-mark GeoJSON export feeds the Bayesian Obj 3 backend. Phase I closes the loop: marks update posterior weights, posterior regenerates envelope, envelope re-renders in AR. The observation set is collected by the shipped prototype today; the importance reweighting is Phase I scope.
Spec: /SBIR-EAP/closed_loop_workflow_spec.md

Obj 5 Helene-Area Pilot Deployment

Success: pilot deployment on ≥1 Helene-affected dam with a state dam-safety program or regional AEC partner. Inspector usability + credibility feedback captured against engineering metrics.
SHIPPED
Helene-anchored default demo scene (Lake Lure Dam)
Cold-start demo loads a procedural inundation polygon over Lake Lure Dam (35.4260°N, −82.1840°W) — Hickory Nut Gorge bearing, real USGS 3DEP DSM in background. Three independent guardrails against demo-being-mistaken-for-real-engineering. Lake Lure is also the target dam for the public records request supporting Phase I validation.
ar-prototype.html · startDemoMode, makeSyntheticDownstreamPolygon
PHASE I
State dam-safety / AEC partner pilot
Letters of Support outreach drafted to NC DEQ Dam Safety, USACE Wilmington District, Kleinschmidt (HEC-RAS expertise), and NC State CCEE (academic computational hydraulics). Pilot site recommendations include Lake Lure, Bee Tree, Lake Susan, Cane Creek/Mills River (Helene-affected WNC).
LOS package: /SBIR-EAP/LOS_OUTREACH_v1.md
PHASE I
3-PE evaluation panel (Phase I final report deliverable)
Independent licensed NC / SC PE reviewers evaluate the prototype on a Helene-affected dam site. Subcontract line in Phase I budget ($35K total: 1 QA reviewer + 3-PE panel). Selection criteria: NC PE licensure for primary, geographic diversity for panel, freedom from COI with current employer.
Budget: BUDGET_JUSTIFICATION_v0.md §2

Phase II / commercialization deferred

The Phase II commercialization plan covers SKU definition, pricing, letters of intent, and rollout to the six primary buyer segments (state dam-safety programs, regional AEC dam-safety practices, FEMA floodplain managers, insurance adjusters, emergency responders, dam owners). The TAM estimate is in the high tens of millions ARR across the segment set.