A forecast must survive its history.

The previous winning party repeated in 23 of 60 seats in the first replay and 26 of 59 comparable seats in the second. That is a warning against treating historical strength as future certainty.

These are intentionally simple benchmarks, evaluated retrospectively after the outcomes were known. They establish a reference for future research; they are not a fitted model, a prospective validation record or probabilities for 2027.

THE HISTORICAL REPLAY / WINNING PARTY CODES

Start with a benchmark.
Keep the misses.

The statewide rule chooses INC, the largest party by seat count in 2017; this does not identify who formed government. AC 40 Jiribam is excluded from both comparison scores because its previous winner was IND. The statewide rule can still be evaluated over all 60 seats: 5/60; that different denominator is not mixed into the comparison.

60 of 60 seat observations shown · 2022 historical outcomes

PROTOCOL 1.0 / WHAT ENTERS THE TEST

Separate the inputs
from the answers.

2027 / EVIDENCE AND VALIDATION GATES

Readiness is evidence,
not a score we assign ourselves.

Historical benchmark research only; calibrated 2027 probability model unavailable.

No probability model is selected merely because one retrospective benchmark beats another. Further research needs time-stamped inputs, additional evaluation evidence, explicit candidate/alliance assumptions and tested uncertainty before numerical forecasts can be defended.

Reproduce the result.

↗ Official election sources are indexed in the source register ↗. Derived calculations by Mandate Terrain. The model protocol and data are versioned in GitHub; source claims retain their original scope.