Method and definitions

Each capability is assessed on three main questions: (1) how strategically critical it is; (2) whether an external provider could use the dependency as political or economic pressure; (3) whether Europe could establish a credible alternative within 24 months. Additional factors are cost, time to operational readiness, trusted partners, market concentration, technological maturity, impact on innovation and competitiveness, and security and resilience. Factors not yet assessed are shown as such.

• Build: critical, high pressure risk, no credible alternative within 24 months.
• Buy: manageable dependency, competitive market, sufficient alternatives.
• Ally: cooperation with trusted partners brings advantages and the dependency is acceptable.
• Mixed recommendations (e.g. Build + Ally) are used where parts of a capability differ. A capability with no approach is shown as pending.

Strategic importance (high / medium / low) reflects the consequences for EU and European security, the economy and programme continuity if the capability were unavailable. It is a judgement, not a measurement.

All initial recommendations are labelled Draft assessment — requires review. Reviewers may adjust a recommendation; each adjustment records the previous and new approach, reason, evidence, reviewer and date.

Likelihood (1 Very unlikely – 5 Very likely) and impact (1 Insignificant – 5 Critical) are scored separately. Score = likelihood × impact. Severity bands: Low 1–4, Moderate 5–9, High 10–16, Critical 20–25.

A score ranks risks for attention; it is not a probability or a forecast. Two risks with the same score can differ greatly (a likely moderate risk versus an unlikely critical one), which is why likelihood and impact are always shown.

The six initial scores are Initial group assessments: judgements by the project group, not independent measurements. They are provisional until reviewed against evidence.

Reassessment. New evidence (a KPI change, a sourced development or, later, a news item) may trigger a review recommendation. A score changes only when a reviewer records a new assessment with reason, supporting evidence and reviewer; earlier assessments are kept. News never changes a score automatically.

Links between KPIs, capabilities and risks are suggested relationships, not validated causal connections. Reviewers can confirm or reject them.

Source selection. Initial sources were supplied by the project group. A feed is ingested only after its official RSS/Atom URL has been located on the publisher's own site and returned valid XML. Sources whose URL or publisher identity cannot be confirmed remain inactive and are listed as requiring manual review. New sources start inactive and unconfirmed.

Source reliability. Each source carries a weight from 0 to 1 (default 1.0; OLFI 0.5 per the project brief). The weight affects how confidently information is presented and slightly favours more reliable sources when selecting key developments. It never changes the relevance score. Under "Most relevant", OLFI is guided to about 15% of the list; excess items are moved down, never removed.

Relevance score (0–100) = 40% relevance to European strategic autonomy + 25% relevance to existing KPIs and risks + 20% strategic importance and potential impact + 15% recency. The first three are judgements made in context by the AI model (and correctable by reviewers); recency is computed (100 on the day of publication, falling linearly to 0 at 30 days; unknown dates score 0). Keyword matches are recorded for discovery only and do not enter the score.

Categories. Strategic (high priority): policy, institutional decisions, geopolitics, export controls, critical infrastructure. Industry (medium): companies, funding, acquisitions, supply chains, technology. Operational (monitoring): launches, deployments, incidents, demonstrations. Operational or industry items with impact ≥ 70 are escalated to high priority.

Key developments are at most five articles from the last 7 days with relevance ≥ 60, completed analysis, sufficient information and no rejection. Alerts require an alert category, relevance ≥ 70 and impact ≥ 70.

AI analysis limitations. The model sees only the headline, feed excerpt and metadata — not the full article, which is often paywalled. "What happened" restates the source; "Why it matters" and "Strategic implication" are analytical interpretations, labelled as such, and are not statements by the publisher. Where input is too thin, the implication reads "Insufficient information to establish a reliable strategic implication." AI analysis is an analytical aid, not an independently verified intelligence assessment.

Deduplication and history. Articles are unique by URL and by source + feed identifier; duplicates are counted and ignored, never overwritten. Original headline, URL, excerpt and publication date are protected against edits at database level; the publication date is stored exactly as the feed gives it.

Refresh and availability. A scheduled backend job fetches validated feeds every 60 minutes and analyses up to 20 new articles per run. Each run records attempts, new items, duplicates and errors. Everything shown is stored feed data from the last successful refresh; no demo or simulated news is used.

Copyright and access. Only metadata and excerpts of at most 400 characters are stored; full articles are not reproduced, paywalls are not bypassed, and robots.txt is checked before each fetch. Readers follow the link to the original.

Manual review. Reviewers can accept, correct or reject an article's classification and relevance, confirm or reject suggested KPI/risk/capability links, add links, edit sources and keywords. Every review needs a note and records the reviewer. News never changes a KPI value, status or risk score; it may prompt a separate documented review.