Cases

What ARGUS would have seen.

Three real situations, looked at again — using only what was public before each one happened. We already know how they ended. So the most useful line in every case below is the one that says what ARGUS would have missed.

Read this first

In plain words: this is looking backwards, not predicting. It is easy to look clever about something that already happened.

The rules we followed

  • Only information that was already public before the event counts.
  • Every case says what ARGUS would have shown — and what it would not have told you.
  • No case claims ARGUS predicted anything. It didn't. Nobody ran it at the time.
  • Sources are listed so you can check us.

Same reason ARGUS scores its own forecasts with a Brier score: being wrong should cost something.

Why there are no screenshots on this page

You would expect screenshots of ARGUS analysing these cases. There are none, and the reason matters.

Until recently ARGUS's correlation signals only looked at events from the last few hours or days, measured from right now. Feed it events from November 2024 and nothing fired — they were outside every window. The only way to get a convincing screenshot was to re-date old events to today and photograph the result.

That is staging, so we did not do it. Instead we fixed the engine. It now takes an "as of" date and replays itself as it would have run at that instant — with a guard so that events after that moment cannot leak in and flatter the result.

So honest runs are now possible. What is still missing is the historical evidence set for each case below — every event sourced, placed and dated by hand. Until that exists, this page describes the mechanism, checked against the source code rather than remembered from feature names. Where the code did not do what we first claimed, we corrected the page rather than the story.

Backtest clock lives in runCorrelationEngine(..., asOf), covered by tests that assert historical events fire nothing on the live clock, fire correctly as of their own date, and never leak the future.

01 · Two cables cut in the Baltic Sea

In plain words: a ship went dark, sailed over two internet cables, and both broke. That is the exact shape of thing ARGUS was built to notice.

What happened

On 17–18 November 2024, two undersea cables broke in the Baltic a day apart: the BCS East-West Interlink (Lithuania–Sweden) and C-Lion1 (Finland–Germany).

The window we looked at: 15–18 November 2024.

What was already public

SignalWhere it came fromGrade
Cargo ship leaves a Russian port, 15 NovPublic ship-tracking (AIS)A2
Its AIS transponder goes quiet for about 7.5 hoursPublic ship-tracking (AIS)A2
Its track crosses both cable routes near the break timesAIS + published cable mapsB2
Operators confirm both cables are downCable operatorsA1

Reading a grade: A2 means the source is very reliable and the claim is probably true. B2 means usually reliable, probably true. Nothing here needed secret information.

What ARGUS would have shown

ARGUS carries 530+ real submarine cable routes from the TeleGeography dataset. When a critical event lands within 40 km of a cable's route, that cable is marked at risk on the map.

So once the breaks were reported, ARGUS would have tied each report to the exact cable system it sat on, drawn the affected corridor, and kept the sources attached. That is genuinely useful — it turns "a cable broke somewhere in the Baltic" into a named system, a route, and a map you can hand to someone.

What it would not have told you

It would not have flagged the ship beforehand. We checked our own code rather than assume. ARGUS has no detector for a transponder going quiet, and it does not compare a vessel's track against cable routes. Its Dark Fleet signal is a different thing — unnamed vessels loitering near one of 32 named chokepoints — and the Baltic break sites are not chokepoints.

And it could not have told you why. Accident or sabotage was never conclusively established; investigators themselves disagreed.

Honest summary: on this case ARGUS is good after the fact, not before it. Ship-track-versus-cable correlation is a real gap, and it is on the roadmap because of this case.

Sources: Wikipedia (2024 Baltic Sea cable disruptions) · CNN, 20 Nov 2024 · Lieber Institute, West Point.

02 · India and Pakistan, April–May 2025

In plain words: after the Pahalgam attack, the public signs of escalation piled up for two weeks. ARGUS would have shown the curve bending — not the date of the strike.

What happened

On 22 April 2025 an attack at Pahalgam in Indian-administered Kashmir killed 26 people. Clashes along the Line of Control followed. On 7 May India launched strikes ("Operation Sindoor"); Pakistan retaliated. A ceasefire was agreed on 10 May.

The window we looked at: 22 April – 6 May 2025 — everything before the strike.

What was already public in those two weeks

SignalGrade
The attack itself, and the claim of responsibilityA1
Repeated exchanges of fire along the Line of ControlB2
Diplomatic expulsions and mission downgradesA2
Suspension of the Indus Waters TreatyA1
Airspace closures and airline reroutesA1

What ARGUS would have shown

The Conflict Escalation signal does not care about headline volume. It compares the count of graded events to that region's own 14-day baseline, and asks how unlikely the new rate is. Through late April that line steepens sharply, and Regional Instability corroborates it from a second direction.

You would have opened the map on 5 May and seen, in one screen: a rising theatre, the graded evidence behind it, and a brief you could hand to someone — with every source attached.

What it would not have told you

The date. The targets. The name of the operation. ARGUS would have said "this is getting worse, quickly, and here is the evidence." It would not have said 7 May, and it would not have named a single site. That information was not public, so no open-source tool had it.

Sources: Wikipedia (2025 Pahalgam attack; 2025 India–Pakistan conflict) · public timelines of the crisis.

03 · Israel and Iran, June 2025

In plain words: this is the case where the honest answer is mostly "no". A surprise first strike is planned in secret. Public data does not contain it.

What happened

On 13 June 2025 Israel struck military and nuclear sites in Iran. Iran retaliated with ballistic missiles and drones. The United States struck fortified nuclear sites, and Iran fired at a US base in Qatar. A ceasefire was announced on 24 June, ending what became known as the Twelve-Day War.

The window we looked at: the weeks before 13 June 2025.

What was already public

SignalGrade
Nuclear-programme findings and IAEA board activityA1
Embassy advisories and staff drawdowns in the regionA1
Airlines rerouting away from certain airspaceA2
Reporting on force posture and readinessC3

Notice the C3. Posture reporting is exactly the kind of claim that sounds most exciting and holds up least. ARGUS grades it low on purpose, and a low grade limits how far it can move a brief.

What ARGUS would have shown — and what it would have missed

It would have shown pressure: advisories, reroutes and nuclear-file friction stacking up in one region, each graded, on one map.

It would not have given you 13 June. Not the date, not the targets, not the decision. A surprise strike is a decision taken privately by a small number of people. It does not appear in open sources until it happens.

We are stating that plainly because the opposite claim is the easiest lie to tell about a tool like this. Open sources show you pressure. They do not show you decisions.

Sources: Wikipedia (Iran–Israel war; Prelude to the Twelve-Day War) · Britannica (12-Day War) · France 24 retrospective.

So what is it actually for?

In plain words: ARGUS is a way to see and grade what is public, quickly, with the evidence kept attached. It is not a crystal ball.

What these three cases show

  • It is strong where the evidence is physical and public — ships, cables, aircraft, earthquakes, fires. Case 01 is its home ground.
  • It is useful where pressure builds in the open over days or weeks. Case 02: it turns a messy fortnight into something you can show someone, with sources.
  • It is limited where the deciding happens in private. Case 03: it shows the weather, not the order.

Every one of these was run knowing the ending. That is why each case carries its own "what it missed" line, and why ARGUS keeps a scored ledger of its forecasts instead of a wall of wins.

See the formulas behind these signals →

Next

Guide · Tour · Methods · GitHub