How a Vedic Labelled Event Set Is Built, the ground-truth list that makes calibration possible.

Tempora's calibration figures are derived from labelled event sets. A labelled event set is a list of confirmed historical events for a specific chart, each tagged with a date, an event type and at least one external source citation. The list functions as ground truth. The engine reads each event date, computes the Vedic configuration active on that date (mahadasha, antardasha, transit aspects, divisional positions) and asks which configurations recur across events of the same type. This piece walks through what a labelled event set is, the eight national charts Tempora maintains them for, the five event categories with their inclusion criteria, the source-citation discipline that keeps the set honest and the Monte Carlo 300-random-date baseline against which lift figures are measured. The engine sits behind Research Note 005 and the calibrated-lift figures published across Tempora's macro forward calls.

What a labelled event set is

A labelled event set is a list of confirmed historical events for a specific chart, each tagged with three required fields: the date the event happened, the event type (leadership transition, financial crisis, military escalation, economic contraction or social-policy shift) and at least one source citation that is independent of astrology literature. The set is stored as a JSON file in Tempora's data directory and is the input to every calibration run.

The word labelled is doing the heavy lifting. A list of dates is just a chronology. A list of dates tagged with event types is a labelled set: the engine can now ask configuration questions inside a type (which configurations recur across leadership transitions specifically) rather than against the undifferentiated whole. The labels are what make type-conditional measurement possible.

The set is also bounded. Tempora maintains the labelled event set for a chart from the chart foundation date to the current calibration cutoff. India's set runs from August 15 1947 to the current cutoff. The US set runs from July 4 1776. Inside that window every event meeting the inclusion criteria is captured. The boundedness matters because the Monte Carlo baseline (described below) draws random dates from the same window, which makes the chance-rate comparable to the event rate.

The eight national charts

Tempora maintains labelled event sets for eight national charts. The choice of charts is determined by two practical constraints. The foundation moment has to be credible (a defensible date and time for the chart) and the historical chronology has to be reliable (peer-reviewed academic histories, sovereign records, archived newspapers of record covering the chart's full life). The eight charts that meet both constraints are listed below.

Each chart's foundation moment is itself a contested choice in the broader literature. Tempora documents the foundation moment used, the source from which it was taken and any alternative moments rejected. The contested-moments record sits alongside the labelled event set in the same data directory. A calibration run that switched the India chart from midnight to the Nehru tryst-with-destiny speech moment (15 minutes earlier) would produce different lift figures and that switch would be recorded in the run metadata.

The five event categories

Each labelled event carries a type tag. The five tags Tempora uses cover the macro events that national charts are credibly read against. The categories are not arbitrary: each one corresponds to a Vedic house-or-karaka cluster that classical literature reads for national charts (10th house for leadership, 2nd and 11th for finance, 6th and 8th for military, 11th and 2nd for economic, 1st and 9th for social-policy). Confining events to these five tags keeps the labelled set tied to readable Vedic structure.

Leadership transitions

Heads of government changes. Inclusion criteria: elections that change the ruling party or coalition (not routine re-elections of the same coalition), assassinations of sitting heads, resignations under duress and coups. The criterion is whether the chart's 10th-house signature would credibly be expected to register the change. Routine party re-elections within a stable coalition are excluded because the chart signature is not expected to differ from background.

Financial crises

Stock-market breaks of more than 10 percent peak-to-trough within 30 days on the national exchange, currency depreciations of more than 15 percent against the dollar within 90 days, banking failures of more than 1 percent of national GDP in deposits at risk. The thresholds are explicit so the inclusion criterion is not a judgement call. The labelled set captures the 1987 US crash, the 1997 Asia crisis effects on relevant national exchanges, the 2008 global financial crisis dates by national onset, the 2020 March Covid crash and so on.

Military escalations

Declared wars (formal declaration date), large troop deployments above 50000 personnel to a foreign theatre and major terror attacks producing more than 50 fatalities. The deployment threshold matches the conventional definition of a major military operation; the terror threshold matches the academic definition of a high-casualty event. Routine border skirmishes and lower-casualty incidents are excluded.

Economic contractions

Quarters of GDP decline of more than 1.5 percent annualised, recession onsets defined by the two-consecutive-quarter rule, inflation spikes above 10 percent annualised within a 6 month window. The GDP threshold matches the academic definition of a meaningful contraction; the inflation threshold matches central bank crisis-classification practice.

Social-policy shifts

Constitutional amendments, major rights legislation (universal franchise, citizenship reform, sweeping civil-rights statutes) and regime-defining policy reversals (legalisation or criminalisation of previously settled categories, demonetisation events, currency replacement). The criterion is whether the legislation rewrites the chart's 1st-or-9th-house terms credibly enough to register against the labelled set.

Source-citation discipline

Every event in the set carries at least one external citation. The citations come from a defined hierarchy of sources. First-tier sources are sovereign government records (constitutional amendment registers, central bank releases, election commission certifications, official military communiques). Second-tier sources are peer-reviewed academic histories with university press publishers. Third-tier sources are newspapers of record (Reuters, BBC, New York Times, The Hindu, Times of London, Pravda for Russia, People's Daily for China). The citation set is recorded in the event JSON with source name, publication date and permalink or page reference.

Astrological commentary on the same event is explicitly excluded from the citation set. The reason is structural. The point of calibration is to measure astrology against history. If astrology literature is itself citing historical events, including those citations would be measuring astrology against astrology and the lift figure would be circular. Historical claims need to come from outside the astrology corpus or the entire calibration loses its meaning.

Disputed events (different sources giving different dates) are handled by recording all candidate dates and either selecting the consensus date (when more than half the cited sources agree on a single date) or excluding the event from the labelled set entirely (when no consensus exists). The exclusion bias is documented: removing a credibly-dated event because its date is disputed introduces a selection effect and Tempora records which events were excluded for this reason.

Why the event set determines what calibration can measure

Calibration is the measurement of signal lift: how much more often a Vedic configuration appears on event dates than on random dates. The lift figure is meaningful only against the specific list of events the calibration uses. If the event set is biased (only includes hits, only includes events near transits, only includes events in periods researchers happened to study), the lift figure measures the bias rather than the signal.

This is why Tempora publishes both the configuration tested and the labelled event set against which it was tested. A claim of "1.74x lift on Saturn returns over the India chart" is meaningless without the event set the 1.74x was computed against. The published Research Note 005 documents the labelled India set in full, allowing any researcher to recompute the same lift against the same set or to apply the same configuration test against a different set and see whether the figure replicates.

The implication is uncomfortable but honest. Two researchers using different labelled event sets will produce different lift figures for the same configuration even with identical methodology. The lift figure is a joint measurement and both halves must be published. Tempora's choice has been to publish the event sets in JSON in the public data directory and treat the lift figures as derivative.

The Monte Carlo 300-random-date baseline

The Monte Carlo baseline is the comparison set against which event-set frequencies are measured. For each labelled event set, the engine generates 300 random dates inside the same historical window the labelled events span and computes the same configuration metric on each random date. The random-date distribution gives the background rate at which any configuration appears purely by chance, independent of whether anything historically interesting happened on those dates.

The lift figure is the ratio of configuration frequency on event dates to configuration frequency on random dates. A lift of 1.0 means the configuration appears no more often on event dates than chance produces, so there is no detectable signal. A lift of 2.0 means the configuration appears twice as often on event dates as random chance produces, so the configuration is recurring with the event type. Tempora's published lift figures for national-chart configurations sit in the 1.12x to 5.46x range across event categories, with the strongest signatures on leadership transitions and the weaker signatures on economic contractions.

The 300-draw count is the classical Monte Carlo trade-off between baseline stability and computation cost. At 300 draws, the random-date frequency for typical configurations stabilises within a 5 percent confidence interval. Going higher (1000 draws) tightens the interval marginally and increases computation cost substantially. Tempora's calibration runs document the random-date seed, so the same baseline can be regenerated exactly for independent verification.

Framework limits

A labelled event set is not a complete history. It captures the events that meet defined inclusion criteria within a chosen window. Events that fall just below thresholds (GDP contraction at 1.4 percent, military deployment at 48000 personnel) are excluded by design, even when they look qualitatively similar to included events. The threshold-bound nature of the set introduces a known selection effect that the published methodology notes openly.

The set is also limited to the events the citation hierarchy reaches. Events poorly documented in sovereign records, academic histories or newspapers of record do not enter the set, even when they would qualitatively meet inclusion criteria. This biases the labelled set toward English-language and major-power events. Tempora's Iran set, for example, is sparser than its US set because the academic-history coverage of post-1979 Iran is sparser. The bias is documented in the per-chart event-set README.

Finally, the labelled set is updated as new events occur and as historical reassessment changes the status of older events. The 2007 to 2009 global financial crisis dates, for example, are still being revised in the academic literature; Tempora's set tracks the current consensus and re-runs calibration when the consensus shifts materially. The full event-set version history is recorded so prior calibration runs remain reproducible against the set version they used.

Frequently asked questions

This article was prepared by Tempora Research as a methodology piece in the Method cluster. Calibrated lift figures and labelled event sets are documented in Research Note 005 and reproducible against the public engine. Internal audit log maintained. This article does not constitute medical, financial, legal or professional advice. First published 2026-06-04 by Tempora Research.