Validation and Verification Report

Integrity and credibility of the NFSI

NationFiles and the NFSI are designed to serve as a trusted reference in scientific studies, audits, and regulatory assessments. The following positions underscore traceability and defensibility.

• No invented values: All NFSI values and forecasts are produced solely from documented formulas and the sources listed in this report and the data inventory. There are no manual overrides or black-box corrections.
• Fully code-based and documented: Every calculation step (Layers 1–4) is implemented in code and described in this report with constants, formulas, and pseudo-code.
• Provenance and audit trail: Data origin, licences, and update frequencies are documented in the data inventory and legal sources. Changes to model and data are versioned and traceable via an audit trail.
• Governance includes version control, audit trail, and defined processes for external auditors.
• Algorithmic interpretations do not represent official assessments. Technical security, audit cycles (e.g. SCA/SAST), and infrastructure are described in the Global Security & Resilience Statement (/legal/security/). Error correction for raw data: report to data@nationfiles.com → source review → correction in the database → automatic recalculation; manual overwriting of scores is excluded (see Independence Statement /legal/independence/).

Absolute transparency

NationFiles is committed to full transparency in the methodology, data, and calculation of the NFSI. There are no hidden parameters, no opaque corrections, and no deliberately withheld information relevant to assessing the index.

What is disclosed:

• All formulas and calculation steps (Layers 1–4), including pseudo-code and the complete constants table, in this report.
• Complete data inventory of all sources with NFSI impact: origin, licence, update frequency, thematic weight. Access to legal sources (/legal/sources/).
• No secret weightings: The weight matrix (connector weights, groups) is documented; upon request, authorised auditors can access technical artefacts under audit agreements.
• Changes to model and data are versioned; an audit trail (e.g. timestamps, hashes) ensures traceability. Calculation logs per country/date can be provided during audits to verify intermediate steps.
• No black box: Every published NFSI value can be reconstructed from the documented formulas, constants, and input data. There are no downstream "secret corrections" or manual overrides.

Transparency is not a marketing claim but documented practice: this report and the linked references are the basis.

How to cite this report

This report is citable for media, academia, and public authorities. Use the following:

Short:

Weighting, scales, and normalization — how everything is weighted

So that the reader can see unambiguously how the system works: all value ranges, normalization rules, and weightings are described here in one place.

All stability scores in the NFSI system lie in the interval [0, 100] after processing. Higher = more stable (better), lower = less stable (worse). Exception: raw values before Layer 1 are source-specific and are mapped to 0–100 in Layer 1.

Per data source, one raw-value column (or a computed severity) is used. Over all rows of that source, minimum (min_raw) and maximum (max_raw) are computed; span = max_raw − min_raw. If span ≤ 0 or min/max not defined, all rows receive the neutral score_row = 50. Otherwise: per row normalized = (raw − min_raw) / span × 100, i.e. values in [0, 100]. Direction is fixed per indicator: “higher raw = worse stability” (e.g. conflict count) → score_row = 100 − normalized; else score_row = normalized. The result is clamped to [0, 100] and rounded to two decimals. Thus every row has a uniformly interpretable score: 0 = worst, 100 = best stability within that source.

Per connector, country, and date, all associated row scores (score_row) are collected. Security group (group 100): daily score is the minimum of collected scores; missing values are padded with 100 (conservative: one critical event suffices). All other groups: fixed dummy values 0 and 100 plus pad value 50 for missing entries are used so that all countries have the same effective array length; daily score is the arithmetic mean of this array. Then smoothing with previous day: score_final = 0.6 × dayScore + 0.4 × yesterday (yesterday: last available value; if missing, 70 for security, 85 for others). Value is always in [0, 100]. Recovery: if a day is missing, score is increased by 0.2 (security) or 1.0 (others) per day, up to 95, for at most 90 days.

Each connector has a configured weight (e.g. 40–100, see data inventory) and a thematic group (e.g. 40, 50, 60, 85, 100; group 100 = security). The effective weight in the weighted average is effW_n = group_n × (scoreValue_n/100) × updateMult_n, where scoreValue_n is the connector weight/100 and updateMult_n reflects update frequency (e.g. 1.0 annual to 7.3 multiple times daily). Missing connector data (no_data) receive the neutral score 50. The weighted average is formed with fixed dummies 0 and 100 (weight 1 each) → baseScore. Then conflict malus (if min(security) < 70), fragility malus (governance gap × population sensitivity), small-country malus (if pop < 5M), population bonus (log₁₀(pop)×0.5, max 4), WGI pull (governance 0–100 raises raw score). All intermediate results are clamped to [0, 100]; final raw score has floor 1 and cap 100.

The raw score from Layer 3 (nfsi_raw) is smoothed with the previous day’s NFSI (prevScore) unless there is a security crisis (min(security) ≥ 25). Inertia weights: standard 80% previous day, 20% today (inertiaWeight = 0.80). When many connectors are missing (no_data share ≥ 50%), 45% previous / 55% today (0.45) is used; when “score without L1/L2 data” 50% / 50% (0.50). The maximum change from the previous day is ±3 points. In security crisis (min(security) < 25) smoothing is skipped: NFSI_Today = nfsi_raw (immediate pass-through).

Worked numerical example (anonymised)

For illustration without disclosing the full weight matrix: a country has three indicators with Layer-2 scores 72, 85, and 90 on a given date. Thematic groups are 100, 60, and 40 with full value 100.

1. Step 1: Effective weights: 72×1 = 72, 85×0.6 = 51, 90×0.4 = 36. Sum of weights = 159.
2. Step 2: Weighted average (without dummies): (72×72 + 85×51 + 90×36) / 159 ≈ 80.2. With dummies 0 and 100 the order of magnitude remains comparable.
3. Step 3: Min(security) = 72 ≥ 70 → no conflict malus. NFSI_Base ≈ 80.2.
4. Step 4: Previous day = 79. NFSI_Today = 0.5 × 80.2 + 0.5 × 79 = 79.6 (simplified inertia).

The actual implementation uses additional factors (population bonus, governance pull, exact inertia parameters). This example is for traceability, not full replication.

Detailed mathematical calculations and procedure

The following formulas and steps match the implemented NFSI logic. All constants are fixed. Notation: raw = raw value, score = value in [0, 100], sums and weights as stated.

These values are used in the formulas:

• Layer 1: If no spread (min = max) or no min/max → default score = 50.
• Layer 2: Today weight = 0.6; previous-day weight = 0.4. Dummy values: 0 and 100 (weight 1 each). Pad value for missing entries (non-security): 50. Security group (group 100): pad missing with 100 (conservative). Recovery per day: security 0.2 points, others 1.0; cap 95. Start without history: security 70, others 85.
• Layer 3: Conflict threshold = 70; malus factor = 1.0; conflict malus cap 35. Reference population = 45M; pop-neg multiplier cap 2.0. Small-country threshold = 5M; malus per log₁₀ step = 4.0; cap 25. Population bonus: factor 0.5, cap 4.0. WGI pull = 0.95. No-data score = 50. Fragility: WGI scale 0–10, factor 3.0, cap 15. Score floor = 1.
• Layer 4 (Inertia): Standard 80% previous day, 20% today; with many missing data 45% or 50%. Max daily change ±3. Security crisis (min security < 25): inertia off, immediate pass-through.

1. Input: Per data source, a table of rows (e.g. per country/date/time). One raw-value column is defined per source (or a computed severity).
2. Minimum and maximum: Over all rows of the source, min_raw and max_raw of raw values are computed. span = max_raw − min_raw.
3. Special cases: If span ≤ 0 (no spread) or min/max not available, score_row = 50 for all rows (neutral).
4. Normalization per row: normalized = (raw − min_raw) / span × 100. So 0 ≤ normalized ≤ 100.
5. Direction: If “higher raw = worse stability”, then score_row = 100 − normalized; else score_row = normalized.
6. Clamp: score_row = max(0, min(100, score_row)); round to two decimals. score_row is stored per row.

1. Aggregation per (source, country, date): All Layer-1 scores (score_row) for this triple are collected. If an entry is missing for a country/date, it is estimated from previous day and trend: estimated = previous × (avg_current / avg_previous), clamped to [0, 100].
2. Security group (group 100): Daily score = MIN(collected scores). Missing values are padded with 100 (“one bad event suffices”).
3. Other groups: Dummy 0 and dummy 100 plus pad value 50 for missing entries are used so array length per date is uniform. Daily score = arithmetic mean of this array.
4. Previous day: The last available Layer-2 score for (source, country, previous day) is used as yesterday. If missing, start value 70 (security) or 85 (others).
5. Smoothing: score_final = 0.6 × daily_score + 0.4 × yesterday. score_final = max(0, min(100, score_final)); two decimals.
6. Recovery: If a country has no entry on a day, recovery is applied: score = min(95, score + 0.2) for security, score = min(95, score + 1.0) for others; fill at most 90 days.

1. Contributors: Per (country, date), all Layer-2 scores per source (connector) are provided. Missing connector entries get score 50 (no_data) and no_data flag.
2. Effective weight: For each source n: eff_w_n = group_n × (scoreValue_n / 100) × updateMult_n. group_n ∈ {100, 85, 60, 50, 40, …}; scoreValue_n ∈ [0, 100]; updateMult_n e.g. 1.0 (annual) to 7.3 (multiple times daily). Sources with group < 0 are excluded.
3. Weighted average: sum_weighted = Σ_n (score_n × eff_w_n) + 0×1 + 100×1 (dummies). sum_weights = Σ_n eff_w_n + 2. baseScore = sum_weighted / sum_weights.
4. Security minimum: minSec = MIN(score_n over all n with group_n = 100 and not no_data). If no security source: minSec = 80 (default).
5. Conflict malus: If minSec < 70: conflictMalus = min(35, (70 − minSec) × 1.0). Else 0. rawScore = baseScore − conflictMalus; rawScore = max(0, min(100, rawScore)).
6. Population multiplier (negative effects): popLog = log₁₀(max(pop, 100,000)); refLog = log₁₀(45,000,000). popNegMultiplier = min(2, refLog / popLog). Used in fragility and optionally small-country malus.
7. Fragility malus: WGI_0_10 = est_total / 10 (governance 0–100 → 0–10). governanceGap = 10 − WGI_0_10. popSensitivity = 1 / log₁₀(pop). fragilityMalus = governanceGap × popSensitivity × 3.0 × popNegMultiplier; clamped to [0, 15]. rawScore = rawScore − fragilityMalus; clamp to [0, 100].
8. Small-country malus: If pop < 5,000,000: smallPopMalus = (log₁₀(5,000,000) − log₁₀(pop)) × 4.0; clamped to [0, 25]. rawScore = rawScore − smallPopMalus; [0, 100].
9. Population bonus: popBonus = min(4, log₁₀(pop) × 0.5). rawScore = rawScore + popBonus; [0, 100].
10. WGI pull: If est_total present (0–100): finalScore = rawScore + (100 − rawScore) × (est_total / 100) × 0.95. Else finalScore = rawScore. nfsi_raw = max(1, min(100, finalScore)); two decimals.

1. Previous day: prevScore = last stored NFSI value for (country, previous day).
2. Security crisis: If minSec < 25 (physical security critical), no smoothing: NFSI_Today = nfsi_raw (immediate pass-through).
3. Smoothing: If prevScore exists and no security crisis: no_data share = count of connectors with no_data / count of contributors. If no_data share ≥ 50%: inertiaWeight = 0.45 (55% today). If “score without L1/L2 data”: 0.50. Else: 0.80 (20% today). smoothed = prevScore × inertiaWeight + nfsi_raw × (1 − inertiaWeight).
4. Daily change cap: diff = smoothed − prevScore. If diff > 3: smoothed = prevScore + 3. If diff < −3: smoothed = prevScore − 3. NFSI_Today = smoothed (two decimals).

How (World) Population is Calculated in Detail

Population figures per country and year used by NationFiles are not taken from a single source but computed from multiple public and statistical APIs. This section documents the sources, their weights, and the formulas and pseudocode of the implementation. Output: main table nf_calc_pop2 (year, iso2, pop, source_values); per source nf_calc_pop2_<source>.

Each source provides values per country (and optionally year). Weights are used in the weighted consensus.

Per year, gaps are filled by extrapolation; consensus is the weighted average; from 2024 a biological constraint is applied.

Extrapolation (years 2000 to target year)

Input: data = source[iso2] (year → population)
Sort data by year; minYear, maxYear
Growth rate rate = (data[maxYear]/data[minYear])^(1/(maxYear-minYear)) - 1 (if ≥2 points)
rate = CLAMP(rate, -0.04, 0.045)
FOR each year y from start to target: IF y not in data: IF y > maxYear: data[y] = ROUND(data[maxYear] * (1+rate)^(y-maxYear)) ELSE IF y < minYear: data[y] = ROUND(data[minYear] / (1+rate)^(minYear-y)) ELSE (interpolation): data[y] = ROUND(linear between prev and next)
Output: data for [start, target]

Weighted consensus per year

rawConsensus = ROUND( Σ(value × weight) / Σ(weight) )
Only sources with value > 0; weight = sourceWeights[source]

Biological constraint (from year 2024)

To smooth aggressive API forecasts: year-on-year growth rate is capped (e.g. max masterRate + 0.5%, min 1.5%).

allowedGrowth = MAX(0.015, masterRate + 0.005)
IF (rawConsensus - prevConsensus)/prevConsensus > allowedGrowth: rawConsensus = ROUND(prevConsensus * (1 + allowedGrowth))

Glossary — terms used in this report

This glossary explains key terms used in the report in plain language. It is intended for all readers — laypersons, auditors, and scientists.

NFSI

NationFiles Stability Index. A 0–100 normalised stability indicator per country and day. Higher value = more stable assessment; computed from multiple data sources across four processing layers (Layers 1–4).

Layer 1

First processing layer: raw values from each source are normalised to a row-level score 0–100 (min-max, with fixed direction ‘higher = better/worse’).

Layer 2

Second layer: row scores are aggregated per source, country, and date into a daily score and smoothed with the previous day (e.g. 60% today, 40% previous day).

Layer 3

Third layer: the country NFSI (raw score) is computed from all source scores: weighted average with effective weights, conflict malus, WGI pull, and other adjustments.

Layer 4

Fourth layer: daily smoothing (inertia). The final NFSI daily value combines previous day and current raw score (e.g. 80% previous, 20% today), with a cap on daily change (e.g. ±3).

Raw value (raw)

Original numeric value from a data source before conversion to a 0–100 score (e.g. conflict count, percentage, scale value).

Normalisation

Conversion of raw values into a common range (here 0–100). With min-max: (value − minimum) / (maximum − minimum) × 100, optionally inverted when ‘higher = worse’.

Score (row / daily / country)

A value bounded to 0–100: row score per record (Layer 1), daily score per source/country/date (Layer 2), country score = NFSI (Layers 3–4).

Goldstein scale

Scale used by GDELT from −10 (highly conflictual) to +10 (cooperative). Translated into stability scores in NFSI logic (higher Goldstein = better stability).

Dummy value

Fixed additive value (0 and 100) in aggregation so that averages and weighting remain stable even when few sources are available.

Inertia

Smoothing over time: today’s NFSI depends partly on the previous day (e.g. 80% previous, 20% new raw score) to dampen short-term noise.

Recovery

When no new data exists for a country/date, the score is gradually raised (up to cap 95) so that gaps are not penalised indefinitely.

Conflict malus

Deduction from the NFSI when the minimum of security indicators (e.g. conflicts, violence) falls below a threshold (70). Emphasises the role of security data.

WGI (Worldwide Governance Indicators)

World Bank indicators of government quality (rule of law, effectiveness, etc.). In NFSI: ‘WGI pull’ can raise the raw score when governance is strong.

Security group (group 100)

Thematic group of security-critical sources (e.g. conflicts, travel warnings). For daily aggregation the minimum is taken (one critical event suffices).

Connector / data source

An integrated data source (e.g. World Bank, GDELT, ACLED) that contributes to the NFSI in Layers 1–3. Each connector has a thematic group and a weight.

No-data

Missing value for a source on a given date. Treated with a neutral score (e.g. 50) or recovery logic so that the calculation does not fail.

VAR (vector autoregression)

Statistical method for the 7-day forecast: uses the history of several time series (e.g. NFSI, news risk) to estimate the next days.

Data provenance

Origin and traceability of data: which source, which licence, how often updated. Documented in this report’s data inventory and legal sources.

Audit trail

Traceable log of changes (e.g. data snapshot, model version) with timestamp and hash, for auditors and reproducibility.

Reproducibility

Property that the same result can be obtained from the same inputs and methodology. Pursued via documented formulas, code, and data provenance.