SagarmathaIQ · Analytica Nepal · Post-Election Model Audit · Falgun 2082

2082 Election: Model Post-Mortem

Published forecast (March 1, 2082) vs. official results · result.election.gov.np · All 165 seats declared
← Back to home View all 165 constituency results →
Analysis
SagarmathaIQ Team | March 9, 2026 | 8 min read
■ 82 correct (49.7%) 83 wrong ■

What happened

Nepal's 2082 election was not a close race. RSP won 125 of 165 FPTP seats — a majority government on its own, the first time any party has done that in the republic era. The result came from a near-total collapse of NC and UML outside their mountain strongholds, driven by the youth-led 2025 uprising that ended KP Sharma Oli's government. Balen Shah defeating Oli in his own constituency of Jhapa-5 was the symbolic headline. Rabi Lamichhane won Chitwan-2 for the third time.

NC fell from 89 seats (2079) to 18. UML fell from 78 to 9. Gagan Thapa's NC 2.0 pitch did not land — he lost Sarlahi-4, the Madhesh constituency he deliberately chose to contest in. NCP (Prachanda) held 8 seats, concentrated in Karnali and Rukum. Shram Sanskriti won 3 FPTP seats as an urban left alternative. The PR vote was even more decisive: RSP took 47.7% of the national PR vote.

Province accuracy

Province Correct / Total Accuracy
Bagmati 27 / 33
82%
Best
Gandaki 10 / 18
56%
Partial
Karnali 8 / 12
67%
Partial
Lumbini 13 / 26
50%
Partial
Sudurpashchim 6 / 16
38%
Failed
Koshi 13 / 28
46%
Partial
Madhesh 7 / 32
22%
Failed
Dhanusha-1 counted as NCP (certificate issued, under court challenge)

What the model got right

RSP as the dominant force in urban Nepal. Kathmandu Valley (14/15), Kaski, Chitwan, Jhapa — the model called the RSP sweep across all major urban centres correctly. These weren't marginal calls. The prior already showed RSP consolidating educated urban voters away from UML and NC, and win probabilities in those seats were high enough that even a model underestimating RSP's national magnitude still got them right. That's a meaningful distinction from a lucky call.
NCP's floor. Forecast 9, actual 8 — the model correctly identified that Prachanda and Barsha Man Pun would hold in Karnali and Rukum. The party did not collapse.
RSP as largest party. The model gave RSP a 67.6% probability of being the largest party. It won an outright majority — the right direction, just underestimated scale.

What went wrong

Madhesh was a complete miss. RSP won 31 of 32 seats. The model forecast NC, JSP and others in most of these, treating Madhesh as a multi-party competitive region with deep identity politics. It was not — RSP's wave overrode all of it. This is the single biggest structural failure: no Madhesh-specific wave term.
RSP scale was underestimated everywhere. The model forecast RSP at 56 FPTP seats. Actual was 125. Even the p97.5 upper bound of the simulation (78 seats) was 47 seats below reality. The 2079 PR baselines used to calibrate vote shares were built for a multi-party equilibrium that no longer existed.
JSP collapsed entirely. Forecast 6 seats, won 0. Parties under existential pressure need explicit collapse scenario branches — a Gaussian model cannot generate zero-seat outcomes from a 6-seat mean.
No constituency is 100% certain. Bhaktapur-1 was assigned 100% to NMKP. RSP won it. Every seat needs a non-zero probability floor for the wave party.

Our High-profile miss: Gagan Thapa (Sarlahi-4)

When Thapa won the NC presidency ~2 months before the election in 2082, our model — like most analysts — absorbed the NC Version 2.0 narrative: new generation leader, reform credentials, likely stops the NC's bleeding. The forecast gave NC a mean of 67 total seats and a 27.7% chance of being the largest party. Thapa himself made a bold territorial bet: instead of defending a safe hill seat, he contested Sarlahi-4 in Madhesh — a signal, he argued, that NC could speak to all of Nepal.

New NC president
Generational signal
Voter confidence recovered
None of it materialized

The model called Sarlahi-4 for Gagan Thapa (NC) with a high-certainty wrong call. Thapa got 22,831 votes but lost overwhelmingly to RSP's Amaresh Kumar Singh. The defeat had several compounding factors: Balen Shah effect in Terai; the language barrier made direct outreach to Madhesi voters difficult; and internal party opponents reportedly worked against him too. Looking at the PR vote, NC got ~5k less PR vote than Gagan got in FTFP indicating upper level leaders didn't support newer faces in congress.

Thapa had deliberately chosen a Madhesh constituency rather than defend his own KTM-4 seat — a strategic bet that he could contest nationally while committing minimal time to any single constituency. That calculation failed. The model treated Sarlahi-4 as a historic NC hold and never accounted for the structural wave dynamics in Madhesh province or the hyper-local political dynamics. It was a confident wrong call on both counts.

For 2087: NC's baseline likely needs be anchored to their actual 2082 result of 18 FPTP seats, not a reversion to historical norms depending on the political landscape. Whether Thapa rebuilds the party over the next five years is an empirical question the next election will answer.

Proportional representation results

The PR vote was even more decisive than FPTP. RSP took 47.7% of the national PR vote — nearly half the electorate — which no single party has done in the republic era. NC and UML, despite decades of organisational depth, each finished around 13–16%.

Forecast vs. actual PR vote share — with 95% confidence intervals

Party Forecast CI [95%] Actual Verdict
RSP 26.5% [20.3–44.9]% 47.7% Above ceiling
NC 20.6% [17.4–25.7]% 16.2% Below forecast, inside CI
UML 20.5% [17.2–27.1]% 13.4% Below lower bound
NCP 14.1% [10.6–23.7]% 7.5% Below lower bound
JSP 6.9% [4.0–10.9]% 1.7% Eliminated (below 3%)
RPP 6.6% [3.3–14.1]% 3.1% Inside CI, barely above threshold

RSP's actual PR share of 47.7% sits above the p97.5 upper bound of the model's simulation (44.9%) — meaning the outcome was beyond the most optimistic scenario in 10,000 runs. The model was right that RSP would lead, but the scale was simply not in its prior. UML and NCP both fell below their lower bounds, suggesting the vote consolidation toward RSP was more total than any scenario anticipated. NC's result was the most accurately forecast — inside the CI and close to the lower tail. JSP's collapse to 1.7% and elimination from parliament was the starkest individual miss.

Final seat totals (FPTP + PR)

Party PR Votes PR Share PR Seats FPTP Total
RSP 51,63,493 47.7% 57 125 182
NC 17,51,172 16.2% 20 18 38
UML 14,55,884 13.4% 16 9 25
NCP 8,11,577 7.5% 9 8 17
SS 3,85,848 3.6% 4 3 7
RPP 3,30,684 3.1% 4 1 5
JSP 1,82,285 1.7% — (below 3%) 0 0
TOTAL 1,08,35,027 100% 110 165 275

RSP's combined total of 182 seats gives it a majority of 44 above the 138 threshold. The PR result mirrors the FPTP wave almost exactly — this was not a split-ticket quirk, it was a genuine national consolidation.

Changes for 2087

1
Build a Madhesh-specific structural model. Provincial identity politics and RSP wave dynamics need separate treatment.
2
Use 2082 FPTP results as the new baseline — not 2079 PR shares. RSP at 125 seats and 47.7% PR nationally is the new prior.
3
Add Shram Sanskriti as a modelled party. 3 FPTP + 4 PR in their first election is a real baseline.
4
Add collapse scenario branches for parties under existential pressure. JSP at 0 seats cannot come from a Gaussian distribution.
5
Cap constituency certainty at 95%. No seat is 100%.