AnTIE: one traffic brain, built by many hands
An open, neutral, Malaysia-hosted traffic-intelligence exchange. Nav apps, fleets and agencies push anonymised segment observations and pull back a fused live road state none of them can build alone. Separately, every local crowd is grey. Together, the same roads are already live.
Malaysia Traffic Intel ExchangeLive traffic is a density game. No Malaysian app can win it alone.
Every local nav app has a better Malaysian product — toll economics, AES alerts, Malay addressing, flood awareness. But each crowd on its own is one to two orders of magnitude too thin to colour the map, so drivers see grey where the foreign incumbent shows red and green. This is arithmetic, not marketing — and it has a measured answer.
- Push: speed samples, incidents, ✓/✕ verdicts — pseudonymous by schema
- Fuse: production trust & fusion maths, one new axis — member_id
- Pull: fused GeoJSON segment state, live deltas, cross-app verifies
The exchange, read off a dispatch board
One road segment, five states of the pipeline — pushed, fused, pulled, cross-verified, and the leading-hazard layer we have designed but not built — rendered as honest mockups, because AnTIE is in formation, not yet running as a national exchange.
The crossover, measured on a live system
We ran the exchange gate against the production gateway before asking anyone to join: same road, same code, one crowd grey — pooled crowds coloured. The honesty gate stayed on the whole time.
Coverage model at typical thin-app density: 1 member → 0.8% of corridors live · 4 → 52% · 5 → 71.5% · 8 → 96.2%. The gate counts samples on the road, not samples per app — pooling opens it with zero code changes.
Three calls in. A national map out.
The barter is deliberately cheap: a few REST endpoints on, an enriched fused state back. Your users, brand, routing engine and pricing stay entirely yours.
Members push
Three small integrations: anonymised speed samples, incident reports, and one-tap “still true?” verdicts. No raw GPS tracks, no user accounts, no trajectories — fields that look like them are rejected at the schema layer, by design.
The hub fuses
Map-matching onto a canonical road-segment registry, per-member trust weighting, k-anonymity floors and Bayesian confidence — the exact maths already running in a production Malaysian navigation stack, generalised so each company is one weighted source among many.
Everyone pulls
A fused, provenance-labelled traffic state richer than any single app could produce: GeoJSON per segment, live WebSocket deltas, and cross-app verify prompts — another company’s driver confirms your report, so a lone tip becomes consensus faster.
The interface, shown honestly
Draft v0 of the protocol — message shapes derived from endpoints that already run in production, generalised with one added dimension. Versioned in the path; additive changes only; failures return an actionable empty state, never a spinner or a fabricated colour.
POST /exchange/v1/observations
Authorization: Member <token>
{
"device_hash": "<member-salted rotating pseudonym>",
"samples": [
{ "lat": 3.14159, "lon": 101.6907,
"speed_kph": 48.2,
"observed_at": "2026-11-03T03:10:00Z" }
]
}
// plate? phone? track? → 400, rejected at the schema layerGET /exchange/v1/traffic/nearby?lat=&lon=&radius_m=
{
"segment_id": 12345,
"congestion": "medium", // fast|medium|slow|blocked|unknown
"speed_kph": 22.5,
"confidence": 0.71,
"sample_count": 41,
"source_mix": { "crowd": 38, "authority": 3 }
}
// below the k-anonymity floor → "unknown" —
// never a guess, never a fabricated colourSix pillars of the exchange
The segment registry
A national canonical road-segment identity — OpenStreetMap way + direction, open-licensed, versioned and diffable. Every member computes the same road graph, so fusion is deterministic and the registry itself becomes a Malaysian-owned mapping asset.
Fusion & trust, in production
Speed-window fusion, reporter-trust decay, spam quarantine, corroboration folds — not speculative. These run today in AnOmbak’s gateway; AnTIE adds a single new axis: member_id, threading ingest through trust to tiered pull.
The crowd-verify loop
Reports mature through verdicts. A fold needs distinct member × pseudonym pairs — two different companies’ users agreeing outweighs two users of one app, so consensus cannot be farmed from inside a single crowd.
Leading-hazard sensing
Kinetic Crowd Sensing: a Bayesian state estimator reading traffic physics to flag a hazard forming upstream, before the visible queue. Thin evidence collapses to silence. Designed; honestly, not yet built.
Reciprocity, not punishment
Pull quality is a deterministic function of contribution: contributors stream real-time; a member that stops pushing degrades live → delayed → coarsened, never a hard cliff. The tier is published on handshake, auditable by every member.
Owned by all, held by none
A member council — one member, one voice — owns the spec and admissions. Reference implementation and registry live in a neutral open-source home with no founder veto. Any member can run its own instance and diff the output: neutrality is testable, not promised.
The objections, answered up front
An exchange between competitors only works if the hard questions are answered in the architecture — not in a meeting room later.
“My competitor sees my users.”
No member ever sees another’s raw data or user base. Input is pseudonymous aggregates with per-member rotating salts — unlinkable across members and windows. The hub itself cannot re-identify a driver even if ordered to.
“The state surveils drivers through this.”
Impossible by construction: no identities accepted, no trajectories stored, coordinates coarsened before storage, aggregate consensus only, Malaysia-hosted, PDPA-by-design, open auditable code.
“Competitors coordinating is a cartel.”
The shared channel is traffic state only — never pricing, users, fleet or strategy. Open, fair, proportional access with per-member firewalls and recorded decisions; structured for Competition Act 2010 sign-off.
“One company runs the referee.”
The founding contributor hands over: code, spec and registry transfer to the co-owned entity on charter ratification. Deterministic reproducibility means any member can verify the referee is neutral.
Questions, answered
As a national exchange — not yet, and we say so plainly. The fusion, trust, k-anonymity and verification maths it needs are already running in a production Malaysian navigation stack (AnOmbak’s gateway), and the density thesis has been measured on that live system, not modeled on a slide. What is proposed is the multi-tenant exchange, the co-owned entity, and the other members. Nothing here describes a service running today.
Push your samples. Get back a national map.
AnTIE is in formation — the fusion, trust and verification maths are already measured on a live Malaysian navigation stack. What is proposed is the exchange itself. Apps, fleets and agencies that want the pooled road state can start the charter conversation today.