Opening the atlas…
FORECAST RESEARCH / A THREE-RULE COMPARISON
Test a prior Parliamentary segment leading-party rule against two prior-Assembly reference rules. Inspect the same-seat comparison, every error and the source limits before drawing conclusions.
The Parliamentary-context rule adds three exact party matches over repeating the prior Assembly party in the first window and two in the second. It still misses more than half of the comparable seats in both elections. In 2017, it also trails the statewide reference rule by two matches.
These are event-prior retrospective reconstructions after the historical outcomes were inspected. They do not prove a calibrated forecasting edge or identify a 2027 favourite. Constituencies are correlated observations within two elections, not 118 independent validation experiments.
↗ The three-rule protocol was written before implementing this rule, after past outcomes and earlier reference scores were known. It is not pre-election preregistration. Original publication-version timestamps for the recovered sources remain unverified.
ONE ADDITIONAL CONTEXT RULE / THE SAME COMPARISON SEATS
Prior Parliamentary segment leader · 2022 Assembly outcomes. All three main scores use exactly 58 seats. Display filters do not recalculate these comparison scores.
Prior Assembly input: 2017. Prior Parliamentary context: 2019. Target: 2022 Assembly winning-party code. Predictions are computed before target labels are joined. Original historical source-version availability remains unverified; this is an event-prior reconstruction after outcomes were known.
| Compared with | Both match | Only context matches | Only reference matches | Both miss | Net extra matches |
|---|---|---|---|---|---|
| Prior seat-winning party | 21 | 6 | 4 | 27 | +2 |
| Prior statewide seat leader | 5 | 22 | 0 | 31 | +22 |
| Rule | Correct / eligible | Abstentions |
|---|---|---|
| Prior Assembly seat-winning party | 26/59 | 1 |
| Prior statewide Assembly seat leader | 5/60 | 0 |
| Prior Parliamentary segment leader | 27/59 | 1 |
| Predicted party | Observed party | Seats |
|---|---|---|
| BJP | BJP | 17 |
| INC | BJP | 9 |
| INC | INC | 5 |
| NPF | NPF | 5 |
| INC | JD(U) | 4 |
| BJP | NPP | 3 |
| NPF | BJP | 3 |
| BJP | KPA | 2 |
| CPI | BJP | 2 |
| INC | NPP | 2 |
| NPF | NPP | 2 |
| BJP | IND | 1 |
| BJP | JD(U) | 1 |
| CPI | IND | 1 |
| NPF | IND | 1 |
THE PROTOCOL / INPUTS ARE NOT TARGETS
Copy the prior ordinary Parliamentary segment leading-party code. Compare it with the previous Assembly seat party and previous statewide Assembly seat leader. No fitted blend, winner-selected rule or probability model is added.
2012 Assembly and 2014 Parliament supply the first window; 2017 Assembly and 2019 Parliament supply the second. Observed Assembly winners are joined only after predictions are frozen. 2024 Parliamentary results and 2026 directory observations are excluded from both replays.
A prior independent is not a persistent political party or a verified person match. Missing, withheld, tied or unresolved Parliamentary leads abstain. This protocol also abstains on statewide ties; neither actual prior Assembly year has one.
For 2022, AC40 Jiribam is excluded from all three main scores because its prior Assembly winner was independent. AC57 Henglep is excluded because the 2019 official sources conflict. Its combined-sheet row is not silently substituted. Separate full-eligibility scores retain their different denominators.
2014 context comes from sixty reconciled combined-sheet rows, with five selected scanned-footer comparisons and no complete individual-station audit. 2019 uses fifty-nine released individual segment tables and leaves Henglep withheld.
Source codes remain election-scoped. Chingai’s selected original 2019 header prints NPEP, while other headers use NPP; that literal difference is preserved in the versioned column register. No merger alias or cross-election candidate identity is inferred.
The net gains are small and the error rates remain high. A few additional exact party matches do not establish stable uncertainty or a prospective forecasting record. Historical scores are not probabilities of winning in 2027.
Next research needs additional compatible Assembly history, evidence about historical source availability and dated current nomination/alliance inputs. Missing evidence remains unavailable.
data/processed/winners.json · SHA-256 cba5b2aec5f6018cb7ce3a3fd2ddca30f66c8b74de9bc2a6ffb8852897432803
data/research/parliament-segments-2014.json · SHA-256 b7a71a9fe03404e194eee352f8384cd183e4a3cffacbbf0e52d75834f8f582b3
data/research/parliament-segments-2019.json · SHA-256 8dcdedbfb518724d7e0f4f5cb154191fdd79760f9f177e95f97cac27e6f2d4cf
data/research/parliament-2019-party-columns.json · SHA-256 11fe77add378eb9ee05b79bfc1793dca9d811e751c33418338452c44ccc4969e
src/lib/forecast-context.mjs · SHA-256 65c440b7ad00937898cd9a2fa390a06bfab4b0f1b39475bca85b4fa4663dcf93
docs/FORECAST_CONTEXT_PROTOCOL_2026-10-21.md · SHA-256 8560be27f1ed2487c83e4f38f78dc27546772e219f3d06d3fe1598cb77420687
↗ Evaluated 21 October 2026. Every site build recomputes the committed context report from the versioned inputs, implementation and protocol. No numerical 2027 forecast, causal voter explanation or inferred sensitive identity is published.