How these numbers were built, and where they are weak

The core rule

Every number rendered on this site must be traceable to a named source with a URL and a verification date. This is enforced mechanically, not by good intentions: scripts/validate.ts runs before every build and fails it if any series lacks a definition, a unit, a resolvable source, or a confidence grade. A series that cannot satisfy that does not ship.

Gaps stay gaps

Where a figure could not be located, the period is stored as null and the chart shows a break in the line. Nothing is interpolated, smoothed, or back-filled with a plausible estimate. The defence exports series is missing FY2014-15 and FY2015-16 for exactly this reason — those two years are genuinely absent from the primary releases located during verification, and a smooth curve there would be a fabrication.

Confidence grades

Confidence describes the number, not the prestige of the publisher.

High (709 series)

Primary document, unambiguous definition, cross-checked against a second source.

Medium (40 series)

Reliable source, but the definition drifts between releases or the figure is routinely revised.

Low (18 series)

Estimate, single-sourced, or contested methodology. A low-confidence series is required to carry a note explaining the uncertainty — the validator enforces this.

Source tiers

Tier 1 is the primary record holder for that quantity — a PIB release, a ministry document, a SIPRI fact sheet. Tier 2 is a credible secondary compiler. Tier 3 is a press report of a primary claim. A series resting only on tier-3 sources cannot be graded above low confidence, and the validator raises this automatically.

Derived charts are labelled

Year-on-year, indexed, growth-rate and share views are computed here from the stored series, and every one says so beneath its title. Derived views are only generated where they are meaningful — a year-on-year chart is never drawn across non-adjacent periods, and an index chart is never drawn off a zero base. Of 4460 charts, 4380 are live on currently-loaded data and 80 await the pipeline backfill; the pending ones are shown with an explicit “awaiting data” state rather than an empty axis.

The two constructed indices

Every other number here reports something somebody else published. The Tonality Score and the Action Index do not. They are constructed on this site, they carry a judgement in their names, and they are graded low confidence for that reason. This section is here so you can decide whether to trust them, or reject them and keep the fatality counts underneath, which stand on their own.

Tonality Score, −100 to +100

A proxy for the state’s posture in a given year. Negative reads accommodative — talks first, threat downplayed. Positive reads decisive — security first, an eradication mandate. Five dimensions, each bounded to ±20:

1. Initiative — the adversary’s share of combatant deaths.
2. Civilian protection — inverse of the civilian share of all deaths.
3. Containment — direction of travel in total violence against the previous year.
4. Attrition — cadre taken alive by arrest and surrender, as a share of all cadre removed.
5. Escalation dominance — this year’s exchange ratio against its own three-year baseline.

Action Index, −1.6 to +1.6

A performance measure rather than a posture one. Four components, each bounded to ±0.4 and each measured against the series’ own average: neutralisations and arrests-plus-surrenders push it up, civilian deaths and total incident volume push it down.

Why nothing here is hand-scored

The obvious way to build a “willpower” index is to read the year’s policy documents and code doctrine and framing by judgement. That would track the stated posture more directly — and it would be unfalsifiable, unreproducible and impossible for you to check. So posture is inferred from what the state did, not from what it said: every dimension above is computed from published fatality counts, and you can recompute all of it from the series on this site.

That choice has a cost, and it is the first item below.

What these numbers cannot tell you

• —Posture is inferred from outcomes, not from statements. A government that talks tough while operations stall scores low, and one that talks softly while operations succeed scores high. The name says willpower; the arithmetic reads behaviour.
• —The inputs are compilations, not a register. SATP, MHA parliamentary answers and press tallies disagree by a few per cent on most years and by more on contested incidents. Differences under about 10 points on tonality are inside that noise.
• —'Neutralised' is the state's count of adversaries killed. Where an encounter is disputed, this index inherits the dispute without resolving it, and a year with contested encounters will read as more decisive than it may have been.
• —Fewer deaths can mean control or it can mean the adversary chose not to fight. The index cannot separate a defeated insurgency from a dormant one.
• —Arrests and surrenders are unevenly reported. Years missing them score zero on attrition rather than being penalised, so those years are measured on four dimensions while others are measured on five.
• —The index says nothing about whether the underlying grievance was addressed, which is the question most of the surrounding argument is actually about.

These are ordinal, not cardinal. A year at +60 is more security-first than one at +30; it is not twice anything. Differences under about 10 points are inside the noise of the underlying compilations.

Known weaknesses

• Nominal rupees. Indian financial series are stored as published — nominal, not inflation-adjusted. A large share of the apparent growth in defence exports, production and budget is rupee inflation and depreciation. Treat every rupee series as an upper bound on real growth.
• Definitional drift. Airport counts vary between releases (157, 159, 160, 163) because “operational” sometimes includes heliports and water aerodromes. Highway network length grows partly through re-designation of existing state roads. These are flagged per series.
• Estimates presented as data. Nuclear warhead counts are external inferences — India publishes nothing. That series is graded low and should be read as order-of-magnitude only.
• Commitments versus delivery. Investment figures for defence corridors are MoU commitments. Grounded investment runs at roughly 15% of that. Where both exist, both are shown.
• Schematic map geometry. Corridor lines on the map are straight segments between real geocoded endpoint cities, not surveyed alignments. The opening years are the load-bearing fact; the geometry is indicative.
• Fiscal versus calendar years. Indian data is mostly fiscal-year (April-March); World Bank data is calendar-year. They are never plotted on a shared axis.

The pipeline

Two connectors run on a schedule. The World Bank connector pulls full 2001-present history for 66 indicators across India and five comparator countries, rewriting each series in full rather than appending — the World Bank revises history, and appending would freeze stale revisions in. The news connector aggregates seven outlets purely as a change-detection signal.

Connectors never throw. A failed connector leaves the previous committed data in place and marks the run partial, so the site keeps serving last-known-good values rather than blanking a chart. Output is written only after validation passes, so a malformed upstream response cannot land in the repository.

What this site is not

Not affiliated with any government body, party or contractor. The assessment page carries editorial judgements and says so plainly; each grade ships with the strongest argument against itself. The data underneath is separable from those judgements — you can reject every grade and still use every number.