Izere/Case studies/Rail Baltica
Backtest · CEF · Rail Baltica TEN-T · Latvia · Estonia · Lithuania

The €18B overrun.
14 months of warning signs.

Rail Baltica is the largest cross-border infrastructure project in EU history. Cost trebled. Timeline doubled. The signals were in the data the whole time. This is the long-form Izere backtest — what the model would have seen, when, and what corrective action it would have surfaced.

€23.8B
Latest cost estimate
€5.8B
Original baseline (2017)
+14.2 mo
Izere lead vs. ECA
2035
Latvia revised completion

In January 2026, the European Court of Auditors flagged Rail Baltica as the largest single budget failure in CEF history. The cost had trebled since baseline. Three Member States revised their completion windows separately. By then, the cost gap had compounded for fourteen months.

The signals were visible from Q1 2024. Bidder concentration in Latvia's contract awards had crossed a structural threshold three quarters earlier. Cross-state coordination cadence had collapsed by 60% in Q3 2024. The €5.8B baseline had been quietly amended four times before the formal cost reset was announced.

None of these signals appeared in any single document. They lived in the structure of the procurement and governance graphs — exactly where Izere is designed to read.

What Izere would have seen.

This is a backtest. We ran the v2.4 production model against the Rail Baltica programme graph reconstructed from public sources between January 2023 and the present. The model was given no information unavailable to the European Court of Auditors at the time. It was given only the structure: contracts, awards, sub-recipients, governance changes, milestone amendments, disbursement events.

The model's risk score for Rail Baltica was 41/100 in January 2024 — moderate, in line with comparable cross-border infrastructure programmes. By May 2024 it had moved to 78/100. By Q3 2024 it crossed 84. The European Court of Auditors detection — January 2026 — was 14.2 months after Izere's threshold breach.

Risk trajectory · Rail Baltica
Izere v2.4 backtest vs. ECA detection.

Izere risk score Traditional supervision Programme events
100 75 50 25 Q1 '24 · CONCENTRATION CROSS MAY '24 · IZERE FLAG · 78/100 Q3 '24 · COORDINATION COLLAPSE JAN '26 · ECA DETECTION Jan 2023 Q1 2024 Q3 2024 Q2 2025 Q1 2026 +14.2 MONTHS LEAD TIME
Signals

Four structural signals.
Five quarters of warning.

Each signal is reproducible from public data using documented features (see Methodology § 03). Auditors can recompute any value below from raw inputs.
F.001 · Bidder concentrationQ1 2024

Latvia bidder pool collapses 47%.

Across 18 contract notices in 2024 Q1, the Herfindahl–Hirschman concentration index crossed 0.62 — three quarters earlier than ECA detection. Qualified bidder count fell from 11.4 (2018 baseline) to 6.0.

HHI Q1 240.62
Threshold0.55
F.003 · Coordination cadenceQ3 2024

3-state meeting cadence drops 60%.

Joint coordination meetings between Latvia, Estonia, and Lithuania programme authorities collapsed from 4.2/quarter (2022 baseline) to 1.7. Document throughput halved over the same window.

Meetings/Q1.7
Baseline4.2
F.004 · Milestone amendmentsQ2 2024

Quiet baseline reset.

The original €5.8B baseline was amended four times across 2024 — none of them publicly announced as a cost reset. Cumulative amendment delta: +€2.4B before any formal disclosure.

Amendments4
Δ disclosed+€2.4B
F.010 · Co-financing gapQ4 2024

2027–28 financing unconfirmed.

National co-financing for the 2027–28 implementation window remained unconfirmed across all three Member States as of Q4 2024. Implied funding shortfall of €3.2B against the (then-quietly-amended) baseline.

Confirmed28%
Required100%
Counterfactual

If Izere had been deployed in 2023.

What would the May 2024 risk flag have surfaced as actionable intervention? The system's playbook generates a recommended response based on which features are driving the score.

For Rail Baltica, the dominant attribution was F.001 (bidder concentration) and F.003 (coordination cadence). Standard playbook: deploy DG REGIO technical assistance to the Latvia managing authority; convene an emergency 3-state coordination session; commission an independent cost reset before the next CEF disbursement window.

None of these are operationally complex. All would have been substantially cheaper than the eventual €18B correction. The constraint was never capacity — it was timing.

Estimated correction cost · May 2024 vs. Jan 2026
MetricMay '24Jan '26
Cost gap to baseline€2.4B€18.0B
Schedule slip+18 mo+96 mo
Member States affected1 (LV)3 (LV/EE/LT)
CEF disbursements at risk€420M€1.84B
Avoidable by intervention~€15.6B

Why traditional supervision missed it.

It is tempting to read this case as a failure of the European Court of Auditors. It is not. ECA's mandate is retrospective sampling and special reports; their detection latency on a programme this large and this politically sensitive is consistent with their statutory role.

The failure is structural. EU supervision relies on programmes self-reporting against linear KPIs. By the time a milestone is formally missed, the underlying structural risk has been compounding for months. Linear KPIs cannot see the kinds of signals that drove Rail Baltica's failure — they live in the relationships between actors, not in any single actor's reporting line.

Linear monitoring asks "did the milestone slip?" Graph monitoring asks "is the structure that produces milestones still intact?"

Izere is designed to read structure. The Rail Baltica backtest is one of seven we have published; performance is consistent across cross-border infrastructure (CEF), national recovery plans (RRF), and operational programmes (Cohesion). Backtests are reproducible — we publish the feature pipeline and ship the inputs as part of every methodology release.

What deployment would look like.

Rail Baltica is now a closed case — the cost has compounded, the schedule has slipped, the institutional response is set. The question is what happens to the seventeen comparable programmes currently in CEF's portfolio.

An institutional Izere deployment for DG MOVE or INEA would surface structural risk across the entire CEF portfolio in real time, with monthly briefings to leadership and an audit pathway every score can be contested through. The deployment is reviewable in advance — methodology, model card, conformity statement, and audit access — before any procurement decision.

Request a CEF backtest Read methodology

Sources & data lineage

Reproducible from public data.

[01]European Court of Auditors · Special ReportRail Baltica programme — first observations on cost and timeline. January 2026.
[02]CEF · INEA programme reportsRail Baltica annual implementation reports, 2018–2025. Cost baselines, milestone status, financial execution.
[03]TED · PPDSAll Rail Baltica contract notices and awards across LV / EE / LT, 2018–2025. Bidder identity, concentration metrics derived per § 03.
[04]RB Rail AS · public reportingJoint venture quarterly publications. Coordination cadence derived from meeting minutes and document throughput.
[05]Member State managing authoritiesLatvian Ministry of Transport, Estonian Tehnilise Järelevalve Amet, Lithuanian Susisiekimo ministerija — programme-specific reporting.
[06]EU AI Act conformity recordIzere v2.4 model card and risk-management documentation, generated continuously during this backtest.