Underlink

CDU IT Code Fair 2026 · Data Innovation Challenge · Top Enders

Covered is not the same as connected.

Underlink is a resilience monitor for remote mobile service in the Northern Territory. Coverage maps show where a phone should get signal on a normal day. They do not show what that signal depends on. Underlink reads the public radio licence register as a network and finds, for each remote Northern Territory community, the relays its service cannot route around.

In plain words: many remote communities get phone service through a line of radio towers on hills. If one tower in that line breaks and there is no way around it, the whole community can lose service, even when the map says it is covered.

Tap a relay to switch it off

Every relay on this chain is working. The community has service.

The shape of one community's licensed radio path to fibre, from a reported outage, drawn without locations. All five relays are single-path relays: there is no other licensed path around any of them.

Words we use

Relay
A radio tower that passes the phone signal on to the next tower.
Weak link (single-path relay)
A relay with no way around it. If it stops, everyone after it loses service.
Fibre town
A town on the fibre cable, where the signal joins the national network.
Radio chain
The line of relays that carries a community's signal to a fibre town.

What Underlink is

A coverage map answers one question: is there signal? Underlink answers the next three. What does that signal depend on? Which single failure would cut it? How long could a repair take? It rebuilds every number from public data with one command, so the NT Government can rerun it each month and watch the answers change.

A coverage map showsUnderlink adds
Signal on a normal dayThe radio chain behind that signal, traced back to fibre
Covered or not coveredThe relays with no way around them, and how many places hang on each
Nothing about failureBackup power, repair time in the wet season, and what still works when the tower is down

One engine, four outputs

For DCDD

Government dashboard

Six measures with a starting point, counts by land council region, and a star schema for the NTG data warehouse (tested in DuckDB).

See the measures

For carriers and emergency planners

Relay register (restricted)

Each single-path relay with a hashed ID, its repair window by month, and a correction file the carrier fills in. Never published.

Why it is restricted

For communities

“When the phone goes down” card

What still works nearby when the tower is down. A draft that a community would redesign with paid interpreters.

See the draft card

For everyone

This web app

Break a link and watch places go dark, see today's fires, floods and closed roads, and the three fallback layers.

Try it

Works with no internet

The public numbers and the six measures in one HTML file. Save it to a laptop, phone or USB stick and open it in any browser, with the Wi-Fi off.

Download the offline version (3 MB)

How it would run each month

  1. Registers update. ACMA licences (daily), carrier outage registers and ACCC site lists.
  2. Underlink rebuilds. One command, a few seconds. A test fails if any number is stale.
  3. The measures move. Six measures by region, loaded into the NTG warehouse.
  4. DCDD acts. Co-investment conditions, a letter to carriers before each wet season, and follow-up on outages.

What is new: Underlink reads the radio licence register as a network and finds, with a dominator tree, every relay a place cannot route around. A coverage map cannot show this. The Python prototype, this web app and an offline single-file version all run from the same code.

Try it: break a link

This small network is made up. It works the same way as the analysis of the real one: 358 Telstra radio sites and 313 links from the ACMA register. Switch relays off and see which communities lose their path to a fibre town. Real relay locations are not shown anywhere on this site.

Swipe the network sideways to see all of it.

Fibre town Relay Weak link (single-path relay) Switched off Community

What we found

In plain words: 23 larger communities reach the phone network through a radio chain. 18 of them have at least one weak link, and about 7,100 people live there. For 16 of those 18 the coverage map says "covered".

All numbers below come from outputs/public/numbers.json, which one command rebuilds from public data. Results are for the 116 larger places (NT Government types Town, Major, Minor and Village).

How the 116 larger places reach fibre

One square for each of the 116 places

Radio chain with a weak link Radio chain, no weak link At a fibre town Radio island No licensed radio site nearby

A radio island has licensed radio links but none that reach fibre, so it probably uses satellite or backhaul the register cannot show.

The 23 places on a radio chain

Does the result depend on our distance choices?

We treat a radio site within 10 km of a fibre town as fibre-connected, and the nearest radio site within 10 km of a place as the one serving it. Change either distance:

Checked against a second dataset

Cyclone replays: exposure, not recorded outages

Own site or place in the footprint Cut off by an upstream relay only

Each replay removes every relay within the chosen distance of a real Bureau of Meteorology track, then re-traces every place. The bars count larger radio-chain places that lose their path to fibre.

The 68 single-path relays

By land council group

Public outputs group the NT into two regions: Central, and the three Top End land councils together. With four separate regions, the two island councils had counts of 1 or 2 that anyone could work out from the totals.

How long a repair could take

In plain words: in the dry season most of these communities would be reconnected within a day. In the wet season, 17 of 23 could wait more than two weeks, because a tower on their chain is far from a sealed road.

Days to restore = 0.5 days on site + travel from the nearest of five crew bases + a wet-season delay when a radio site on the chain is more than a set distance from a sealed road. The delay and the distance are our assumptions, so you can change them.

23 radio-chain places, July and January

Change the assumptions

Wet-season delay (November to April)

Distance from a sealed road that triggers the delay

The count moves between 16 and 18 across all nine combinations. The day counts rest on the assumed delay; which places wait comes from road distance alone.

What still works when the tower is down

Triple Zero

The 000 wording follows ACMA's emergency call rules and the cyclone advice follows SecureNT. The code tests the card for required and forbidden phrases.

When the chain breaks: three layers

In plain words: a broken weak link cuts the phone network. Three layers can keep people in touch: a satellite dish that keeps the tower on air, a satellite truck at the evacuation centre, and the community's own small radios. Each layer has a different owner. Step through it below.

Swipe the picture sideways to see all of it.

Who owns each layer, who pays, what limits it

LayerOwnerExisting way to payLimits
1. A second path for the tower
Satellite dish at the tower; L-band for rain
CarrierMobile Network Hardening Program (backhaul redundancy); NT co-investment conditions (Rec 1)Monthly fees, not one-off cost, so 7 to 10-year grant rules fit badly. Rain can fade Ku/Ka for minutes. Needs power at the tower.
2. A network at the evacuation centre
Satellite truck, fly-in kit, cell on wheels
NBN Co, carriers, NTESSTAND satellite services, funded to 31 December 2027Few vehicles for the whole country; they must reach the place, and roads may be cut.
3. The community's own radios
Bluetooth phones, solar LoRa radios
The communityNone yet; a pilot (Appendix H)Messages, photos and voice notes only, slowly. Cannot reach 000. Someone local must look after the radios. Software is not security-audited. Start from the local radio station if there is one. The community sets its own channel keys; default keys are public.

None of the three layers reaches 000 for a phone with no signal. For 000, use a payphone, a satellite phone, or another company's mobile network if one is in range. A radio can list itself on public maps such as rmap.world; that shows where it is, so listing is the community's call.

Where a pilot could start

What government already funds

  • The Hardening Program's second round lets carriers claim satellite backhaul and scores the extra redundancy it adds.
  • STAND put NBN satellite services at fire depots and evacuation centres; they are extended to the end of 2027, with up to 500 more sites.
  • The government's LEO working group recommended trialling LEO in First Nations communities (2024), and LEO voice trials ran at 50 sites, four of them in the NT, through the 2024–25 wet season.

Already working elsewhere

Hazards now: fire, flood and closed roads

In plain words: fires and floods take relays down, and closed roads keep repair crews away. This map puts the current public feeds together. Pick the layers you need, then save it as a picture or a PDF.

Northern Territory hazards now

Community card: a draft for co-design

Each community would get a one-page card, "When the phone goes down". This sample is for a made-up place. The layout, symbols and wording are all open to change with the community; only the 000 facts are fixed. No community has reviewed this card.

When the phone goes down

Sample Community · information as at 29 September 2026

Illustrative sample

WORKS: known to work. NOT AVAILABLE or CASH ONLY: stopped in a past outage. ASK LOCALLY: nobody has published the answer, so ask the clinic, store, council or school before the wet season. A mobile can still call 000 if another company's network is in range, but most remote communities have only one.

Where to go

  • Satellite Wi-Fi phone: [location to be added by community]
  • Sky Muster at the school or evacuation centre
  • Payphone: ask locally whether it works when the tower is down

Before the wet

  • Charge power banks
  • Know your nearest working phone
  • Save clinic and council numbers
  • Know who to tell when the phone is down

000

In community language

To be written and voiced by community members with paid interpreters (Aboriginal Interpreter Service). No machine translation.

Draft for co-design. Built from public information. No community has reviewed this card. Is this right? Tell your community's chosen contact.

For government: measures with a starting point

In plain words: ask phone companies how each tower connects, how long its batteries last and when it has gone down, then check the weak links before every wet season.

Each recommendation in the report comes with a measure, an owner and a target. Public data gives the starting value today.

Three fields to ask carriers for

When the NT Government co-invests in a site, it asks the carrier for three restricted fields per site:

  1. Backhaul type: fibre, radio or satellite
  2. Battery hours at the site and at each relay feeding it
  3. A three-year outage log

This extends the restricted government access to asset locations that the 2024 Regional Telecommunications Review proposed (Recommendation 6).

Ready for a data warehouse

The pipeline writes a star schema (docs/schema.sql) that loads into DuckDB, and a test builds and queries it on every run.

dim_placedim_sitedim_event
fact_place_snapshotfact_chain_memberfact_replay_resultfact_fallbackfact_outage_event

fact_outage_event is where monthly carrier outage registers would land, so observed repair times can replace our assumed ones.

Method and data

  1. Build the network. 808 NT point-to-point link records licensed to Telstra become 358 sites and 313 links in 51 separate pieces.
  2. Trace each place to fibre. Every fibre-connected site joins one imaginary source; a dominator tree finds the relays each place's site cannot route around.
  3. Add four measures beside it. Cyclone replays, published backup power, a repair window from road access, and what still works nearby. They are shown side by side and never merged into one score.
  4. Check it. Nine sensitivity runs, the ACCC site list, and every outage reported in the news that we could find.
  5. Release by tier. Regional counts in public, a relay register for government and carriers only, and a card each community controls.

Against reported outages

The report tests the model against six outages that news reports described (Table 2). One fits in part. One was planned work on the community's own tower, so it does not test the chain. It misses four, and each miss traces to data carriers hold and do not publish: battery hours, fibre routes and which sites serve which places.

This site names no community. The report names them only because news reports did, and only to test the model.

18 sources

★ marks the eight datasets the organisers suggested. Licences and retrieval dates are in DATA_LICENCES.md and data/manifest.csv.

    Limits

    • The ACMA register shows only licensed radio. Fibre and satellite backhaul are missing; links it cannot show can only lower the count for the 23 chains.
    • Fibre towns are assumed to be sound. The nearest radio site stands in for the serving site; the ACCC check above tests that.
    • Replays show exposure without probabilities. Repair parameters are assumptions and are swept.
    • Population figures are 2020 estimates, and homeland populations change with the seasons.

    Data rules

    Underlink studies a carrier's network, so every finding describes equipment, links and missing data. We follow the CARE principles and the Maiam nayri Wingara principles of Indigenous data sovereignty, and take note of Walter's BADDR critique, which warns against data that is blaming, aggregate, decontextualised, deficit-based and restricted.

    Public (this site)

    Aggregate counts, two land council groups, the made-up network above. No relay locations, no site identifiers, no community rankings.

    Government and carriers

    The relay register with hashed identifiers, for DCDD and carrier field teams only.

    Each community

    Its own card and any outage records it chooses to keep. A custodian the community chooses approves or withholds anything about it. Automated steps cannot publish.

    • A test fails the build if any public table holds place keys or names.
    • Land tenure is used only to flag permit lead time (a section 19 lease on Aboriginal Land Trust land). No sacred-site data is used or inferred.
    • Community-language text is written and voiced by paid interpreters. No machine translation.