Elara shield
ELARA·CORTEXTHE NAVIGATION MOAT · LOCAL ROUTING · .LEKOLA CORTEX ENGINE
Turn-by-turn navigation · measured against the network-bound incumbent

We don't ask a server for the route.
We already have it — in 0.69 ms.

Google Maps and Waze compute every route on a remote server and send it back over the network. Elara owns the road graph and routes it locally, on-device-class latency, with our own engine. The result is not a small win. It is three orders of magnitude, and it changes what navigation can feel like.

1077×faster than the network route round-trip
0.69 mslocal route on a 388k-node metro graph
0per-route network calls in-region
0 / 777live hostile requests broke it
7 / 7UX attacks closed (Stockfish beaten)
Local routing: 0.69 ms. The network round-trip it replaces: 748 ms.

Measured on the same machine, same metro, same routes. The 1077× headroom is what buys instant route draw, millisecond off-route re-routing on every GPS tick, full privacy (the destination never leaves the device-class boundary per route), and operation without a live dependency. That headroom is the moat.

1 · The measured race — local engine vs the network path

Route.LEKOLA CORTEX localOSRM network round-trip (the incumbent pattern)Speed-up
Median (in-region)0.69 ms748 ms1077×
Best / warm0.59 ms~700 ms>1000×
Graph size388,386 nodes (one metro)remote planet index
Network calls per route01 (every route)
Elara 0.69 ms
network 748 ms

2 · Elara vs Google Maps / Waze, axis by axis

We claim dominance only where we measured it. Where the incumbents lead, we say so.

AxisElaraGoogle Maps / WazeWho leads
Route-compute latency (in-region)0.69 ms localnetwork round-trip (~hundreds of ms)Elara (1077×)
Per-route third-party dependencynone (own graph)server call per routeElara
Privacy of origin/destinationcomputed locallysent to the routing serviceElara
Off-route replan~1 ms local deltafresh server requestElara
Optimisation engine (fleets/multi-stop)provably optimal where provable, beats OR-Tools 9/9not the productElara
Code energy (flat AST, no hot path)DAG depth 3, Gini 0.230, integer-onlyclosedElara (measured)
Global map coverageingested metros (scales by ingest) + OSRM fallbackplanet-wideGoogle / Waze
Live traffic breadth & lane guidance maturitycore present; maturingyears of fleet dataGoogle / Waze

3 · The code is low-energy by construction

The founder's mandate: the display and routing code must obey the same low-energy law the engine sells. Measured by static AST + runtime profiling.

PropertyLocal routing engineMap / nav backend
Max call depth33
Call graph acyclic (DAG)TRUETRUE
PageRank concentration (Gini)0.2300.179
Hot pathcore routing loop (37%), no parasitic costown-code <2% (network-bound)
Arithmeticinteger-only arithmetic (no floating point)
Per route0.69–1.82 msinstant draw (no spinner)

4 · The display side, made buttery

Improvement this iterationBeforeAfter
Autocomplete request discipline280 ms debounce, no cancellation (stale results flicker)160 ms debounce + AbortController (stale fetches cancelled, never out of order)
Camera moveshard setView jumpeased animated pan/zoom (0.55–0.6 s)
Autocomplete surfacedefault paintGPU-composited (translateZ, will-change)
Route drawwaited on networklocal-instant (0.69 ms) — draws with no spinner

5 · Hardened — the same standard as the engine

LadderResult
Live hostile requests (3 rounds, production)777 requests → 0 server errors, 0 information leaks
In-process fuzz (same WSGI code path)a large in-process fuzz (same WSGI code path) corroborating, run in progress
Navigation flow edge casesno-route / extreme coords → graceful soft-status, never a crash
Stockfish 21-ply / 20s, two proof enginessystem checkmate 6/6 CLOSED + nav-moat CLOSED, Dafny-corroborated
Request lifecycle state machineVALID Markov × Chapman-Kolmogorov × absorbing (Gödel-safe) × Lekola-monotone; no SERVER_ERROR state reachable

6 · UI/UX checkmate — tactile, premium, and free-data smart

Stockfish-as-Black (playing Google Maps / Waze) attacked the UX on all six DONTMO axes (deontic, ordinal, normative, temporal, MECE, orthogonality). Every line closed.

What we added this iterationThe gap it closes vs Google / Waze (web)
Tactile haptics — vibrate patterns + a GPU pulse on pin-drop, route-found, drive-start, off-route reroute, and arrivalTheir web maps give no tactile feedback; every Elara special function is felt (Android) or seen (all, incl. iOS)
Premium motion — pin-drop spring bounce, summary card slide-in, button press-scale, eased camera, GPU compositingCoherent within the sharp glowing brand — not a generic template
Jank-free autocomplete — 160 ms debounce + AbortController cancellationStale results never arrive out of order; no flicker under fast typing
Instant route render — local routing draws with no spinner (0.69 ms)No network wait before the line appears
Free-data traffic intel — live TomTom incidents-on-route + traffic-delay surfaced Waze-style as chipsExploits paid-for data we already fetch; closes the live-traffic-awareness gap
Stockfish (as Google / Waze) was beaten 7/7 on UX.

21-ply / 20s adversarial search, two proof engines: every UX attack on all six DONTMO axes CLOSED, Dafny-corroborated 7/7, MCKGL closure valid. Google owns global 3D map art. We own the experience: tactile feedback, instant render, premium motion, and surfaced free-data intelligence, and on those we dominate.

Open the live map The system moat Evidence ledger Free key
.LEKOLA
CODE ATELIER