Scenario simulator — PRN16 Melaka

21–25 / 28 seats · Government (BN+PN) · majority 100%

Base case · 1 seat flips vs 2021

Start from

BN+PN pact holds · 50/50 benchmark blend

Coalition configuration

Who contests whom — the model's primary input.

BN vs PH vs PN Three-corner fight BN+PH vs PN BN–PH understanding BN+PN vs PH BN–PN pact · base case BN vs PH+ally vs PN PH absorbs the third-force (Others) vote

Per-seat pins and swings work under every configuration — open a seat in the grid below.

Benchmark 50% GE15 PRN2021 (actual)GE15 (modelled) Which past result the run starts from: the real PRN2021 result, the modelled GE15 ordinary vote, or a blend of the two.

Voter-flow rule

Manual rule Pact transfer at (1 − leakage); sliders as set Observed pattern: Johor 2026 PRN2021 anchor replayed through the Johor 2026 transition pattern (one election, statewide shock included)

How the PRN2021 vote is carried into this run. Manual rule: the pact transfers the junior partner's vote at (1 − leakage) and every slider acts as set. Observed pattern: Johor 2026: the PRN2021 result (as shares of the 2021 electorate, including those who cast no valid vote) is first replayed through the transition pattern fitted on Johor's 2026 contests with the same signature — one election's realized pattern, statewide shock included; ecological, not individual transfers, and not a calibration (the pattern did not transfer between NS and Johor). It rewrites the PRN2021 benchmark only: inert at 100% GE15; seats the library does not identify keep the manual rule.

Statewide swing (pp)

Adds the percentage points to that bloc's share in every seat, before any pact merge.

BN 0 pp Observed 2026 (92 Johor+NS seats): 0 to +36.73, median +16.2 ppPH 0 pp Observed 2026 (92 Johor+NS seats): -14.19 to +25.59, median +2.5 ppPN 0 pp Observed 2026 (92 Johor+NS seats): -45.08 to +7.65, median -19.7 pp

Volatility, pact & turnout

Swing SD (correlated) 4 pp One random statewide swing shared by every seat in each simulation. Higher = wider seat range. Observed 2026 (each state's published default scenario vs the result): the statewide BN+PN-vs-PH miss was +6.09 pp in NS and +1.95 in Johor; PH -0.78 / +0.44. Inside the government side the BN↔PN split missed by +11.02 (NS, PN where it stood) and +12.72 / -18.28 (Johor) — a three-corner run (config A/D) carries that risk beyond this prior. Two elections cannot calibrate this slider; it stays at 4 pp. Local swing (per seat) 6 pp Independent per-seat noise on top of the statewide swing. Observed 2026: the seat-level residual around each state's mean had SD 6.84 pp (NS, 36 seats) and 8.44 (Johor, 56); robust 6.37 / 6.8. The default of 6 pp is calibrated on this (the inherited 3 covered less than half); the seat's archetype widens it further. Coalition leakage 15% Share of the junior pact partner's voters who stay home when their party stands down. They abstain, so turnout falls. Observed 2026: where PN stood aside, the carrier realized a median 95% of the notional BN+PN pool (p10 71%, 48 seats); where PN stood again it kept a median 59% of its prior vote mass (44 seats). Fitted transition pattern (Johor 2026, 13 seats where PN stood aside for BN, electorate frame): the carrier realized 0.89 of the 2021 PN vote (bootstrap p05–p95 0.67–1.00); 0.08 cast no valid vote; 0.02 went to PH. The default leak of 15% (carrier keeps 85%) is compatible with that band — one election's realized pattern, ecological, not a validation. Bersama vote-draw scale 100% How strongly Bersama's announced 15-seat run shaves PH's vote in those seats. 100% = each seat's baked shave: the midpoint of an analyst band set by the seat's Malay share (under 50%: 4.0–5.5 pp; 50–70%: 3.0–4.0 pp; over 70%: 0.8–2.5 pp), scaled around Bersama's Johor PRN 2026 result (15 seats, pooled 3.4% of valid votes, range 0.8–5.5 pp). Johor showed no Malay-share gradient and fielded no Bersama candidate in a seat over 70% Malay, so the bands are a registered assumption, not a measurement. The scale multiplies the midpoint — 70% of a 3.5 pp shave takes 2.45 pp off PH — and the shaved votes sit with LAIN as a spoiler, never returned to PH. 0 = off. Turnout multiplier ×1.00 Scales each seat's PRN2021 turnout on the Q4-2024 roll. Moves vote counts, not shares. Observed 2026: turnout moved +3.21 to +16.76, median +11.5 pp across the 92 seats; correlation with the BN swing +0.13 — volume changed, direction is not implied. New-voter turnout 60% Turnout of the post-benchmark new-voter cohort relative to everyone else.

New-voter lean (pp) — flat assumption, not measured by race (A9). How the new-voter cohort splits relative to its seat. 0 = votes like the seat.

BN 0 pp PH 0 pp PN 0 pp

Weighting dials (advanced)

Age weighting — GE15 age→bloc layer. Reweights the GE15 vote mix by young (18–29) vs senior (50+) share; gross turnout is held fixed. Effect scales with the GE15 anchor weight.

Young weight (18–29) ×1.00 Senior weight (50+) ×1.00

Ethnic weighting — advanced hypothesis (weakly identified)

Same joint GE15 layer (Malay & Others are the baseline). Saluran do NOT separate ethnicity — treat as a registered assumption, not a measurement.

Chinese weight ×1.00 Indian weight ×1.00

How to read the weighting dials

What a weight dial asks: "what if this group is a bigger or smaller share of the votes cast than it was at GE15?" Chinese weight ×1.2 = their slice of the mix grows 20%. Nobody switches parties — only the mix of who shows up changes. That is why the voter field fades or densifies cells for these dials, and never recolours a dot.

How it works: each seat's GE15 vote is split into 12 persona cells (3 age brackets × 4 ethnic groups), each with its own BN/PH/PN split estimated from saluran-level data. A dial rescales its cells' mass and everything renormalises — total votes cast stay fixed. The dials act only on the GE15 benchmark: at 0% GE15 anchor they do nothing, and say so. A per-seat ethnic weight multiplies the statewide one in that seat's cells.

Why ethnicity is behind a warning: AGE is measured — saluran are age-sorted by construction, so the age→vote signal is well-identified. Saluran do NOT separate ethnicity, so the ethnic splits lean on much weaker evidence. The ethnic dials are registered assumptions, not measurements: they answer "IF this group's mobilisation shifts, THEN…" — they carry no evidence that it will. They are also powerful, and powerful + weakly identified is exactly what the warning is for. Treat them as scenarios, not readings.

Why twelve cells and not more: the grid is not a modelling preference — it is the widest split this data can identify. SPR fills saluran from an age-sorted roll, so age separates almost cleanly and the age→vote signal rests on measurement. Nothing else about a voter is sorted that way. Ethnicity is mixed within every saluran, which is why its splits are assumptions rather than readings — and sex, which the underlying GE15 roll does carry, is mixed in exactly the same way and would inherit exactly the same weakness. It is left out deliberately, not overlooked. Widening the grid with attributes the saluran structure cannot separate would add dials without adding evidence.

Simulation settings

Simulations (N) While a slider is moving the page shows a 1,000-draw preview, then re-runs at this N when you let go. Seed (RNG)

Per-seat win probability

Colour = favoured bloc in the model (win frequency per seat is on the cards below). Flipped seats are outlined in the colour of the bloc that won them in 2021 — rim = who held it, fill = who takes it here (the card also spells out 2021 → simulated). Grey = ceded (N1–N5) under the GE15 anchor.

seats won (favoured winner per seat)

Base case · 1 seat flips vs 2021

Loading map…

N01

KUALA LINGGI

BN+PN

N02

TANJUNG BIDARA

BN+PN

N03

AYER LIMAU

BN+PN

N04

LENDU

BN+PN

N05

TABOH NANING

BN+PN

N06

REMBIA

BN+PN

N07

GADEK

BN+PN

N08

MACHAP JAYA

BN+PN

N09

DURIAN TUNGGAL

BN+PN

N10

ASAHAN

BN+PN

N11

SUNGAI UDANG

BN+PN

N12

PANTAI KUNDOR

BN+PN

N13

PAYA RUMPUT

BN+PN

N14

KELEBANG

BN+PN

N15

PENGKALAN BATU

BN+PN

N16

AYER KEROH

N17

BUKIT KATIL

PH → BN+PN

N18

AYER MOLEK

BN+PN

N19

KESIDANG

N20

KOTA LAKSAMANA

N21

DUYONG

BN+PN

N22

BANDAR HILIR

N23

TELOK MAS

BN+PN

N24

BEMBAN

BN+PN

N25

RIM

BN+PN

N26

SERKAM

BN+PN

N27

MERLIMAU

BN+PN

N28

SUNGAI RAMBAI

BN+PN

Government (BN+PN) bloc seats

21–25 / 28 assessed

Majority probability (≥15): 100%

p05–p95 range across 5,000 simulations (not a single number).

Solid

Competitive

Toss-up

Tier: Solid ≥80% · Competitive ≥55% · Toss-up below that (probability the leading option wins).

Typology lens

structural lean analyst prior × this scenario

seats to review

core assumptions failed

Defensive core 14 seats

N01 N02 N03 N04 N05 N10 N11 N12 N18 N23 N25 N26 N27 N28

Three-corner 8 seats

N06 N07 N09 N13 N14 N15 N17 N24

Opposition-leaning 4 seats

N16 N19 N20 N22

True marginal 1 seat

N08

Candidate-driven 1 seat

N21

Typology is the analyst's structural prior — not re-derived from the result (margin alone is not the tier). Amber ring = stressed: this scenario contradicts a directional prior (a government-prior seat — defensive-core or three-corner — no longer led by the government bloc or dropped to Toss-up, or the government intruding on an opposition-leaning seat). Grey ring = on the review list without a directional break (the four candidate-watch seats, and the structurally-uncertain true-marginal & candidate-driven seats). Toggle structural lean to read each seat's Gate B2 prior strength — (BN%+PN%)−PH%, mean of the PRN 2021 & GE15 anchors, in points; positive = government-leaning, negative = opposition (hover for the worst-anchor floor). This is the fixed prior, not the live margin. Not a forecast.

28 DUN seats

Show Sort

N1 KUALA LINGGI

Solid

majority ~5,050 votes if won

Defensive Core local swing ±6pp turnout ~61% lead ~54.3pp wins in 100% of simulations

N2 TANJUNG BIDARA

Solid

majority ~7,910 votes if won

Defensive Core local swing ±6pp turnout ~65.1% lead ~84.5pp wins in 100% of simulations

N3 AYER LIMAU

Solid

majority ~7,300 votes if won

Defensive Core local swing ±6pp turnout ~61.2% lead ~78.7pp wins in 100% of simulations

N4 LENDU

Solid

majority ~5,770 votes if won

Defensive Core local swing ±6pp turnout ~62.9% lead ~62.3pp wins in 100% of simulations

N5 TABOH NANING

Solid

majority ~4,410 votes if won

Defensive Core local swing ±6pp turnout ~61.3% lead ~63.9pp wins in 100% of simulations

N6 REMBIA

Solid

majority ~4,080 votes if won

Three-Corner Volatile local swing ±7.8pp turnout ~62.1% lead ~28pp wins in 97% of simulations

N7 GADEK

Solid

majority ~2,840 votes if won

Three-Corner Volatile local swing ±7.8pp turnout ~59.6% lead ~27.6pp wins in 96% of simulations

N8 MACHAP JAYA

Solid

majority ~2,100 votes if won

True Marginal local swing ±6.9pp turnout ~62.9% lead ~18.8pp wins in 91% of simulations

N9 DURIAN TUNGGAL

Solid

majority ~3,250 votes if won

Three-Corner Volatile local swing ±7.8pp turnout ~65% lead ~22.6pp wins in 93% of simulations

N10 ASAHAN

Solid

majority ~4,810 votes if won

Defensive Core local swing ±6pp turnout ~62.5% lead ~37.8pp wins in 100% of simulations

N11 SUNGAI UDANG

Solid

majority ~10,660 votes if won

Defensive Core local swing ±6pp turnout ~64.2% lead ~57.9pp wins in 100% of simulations

N12 PANTAI KUNDOR

Solid

majority ~5,520 votes if won

Defensive Core local swing ±6pp turnout ~60.4% lead ~43.9pp wins in 100% of simulations

N13 PAYA RUMPUT

Solid

majority ~6,640 votes if won

Three-Corner Volatile local swing ±7.8pp turnout ~61.8% lead ~22pp wins in 92% of simulations

N14 KELEBANG

Solid

majority ~5,710 votes if won

Three-Corner Volatile local swing ±7.8pp turnout ~62.1% lead ~33.6pp wins in 99% of simulations

N15 PENGKALAN BATU

Solid

majority ~3,810 votes if won

Three-Corner Volatile local swing ±7.8pp turnout ~63.4% lead ~21.2pp wins in 92% of simulations

N16 AYER KEROH

Solid

majority ~4,950 votes if won

Opposition Lean local swing ±6pp turnout ~59.9% lead ~20pp wins in 94% of simulations

N17 BUKIT KATIL

Solid

majority ~4,560 votes if won

Three-Corner Volatile local swing ±7.8pp turnout ~62.6% lead ~17.9pp wins in 88% of simulations

N18 AYER MOLEK

Solid

majority ~9,020 votes if won

Defensive Core local swing ±6pp turnout ~65.9% lead ~56.7pp wins in 100% of simulations

N19 KESIDANG

Solid

majority ~9,170 votes if won

Opposition Lean local swing ±6pp turnout ~59.5% lead ~28.4pp wins in 99% of simulations

N20 KOTA LAKSAMANA

Solid

majority ~11,480 votes if won

Opposition Lean local swing ±6pp turnout ~58.2% lead ~57.7pp wins in 100% of simulations

N21 DUYONG

Solid

majority ~3,670 votes if won

Candidate/Locality Driven local swing ±9.6pp turnout ~63.8% lead ~17pp wins in 83% of simulations

N22 BANDAR HILIR

Solid

majority ~9,370 votes if won

Opposition Lean local swing ±6pp turnout ~57.3% lead ~69.6pp wins in 100% of simulations

N23 TELOK MAS

Solid

majority ~7,060 votes if won

Defensive Core local swing ±6pp turnout ~64.1% lead ~37.8pp wins in 100% of simulations

N24 BEMBAN

Solid

majority ~5,130 votes if won

Three-Corner Volatile local swing ±7.8pp turnout ~62.5% lead ~31.1pp wins in 98% of simulations

N25 RIM

Solid

majority ~4,440 votes if won

Defensive Core local swing ±6pp turnout ~62.4% lead ~40.1pp wins in 100% of simulations

N26 SERKAM

Solid

majority ~10,130 votes if won

Defensive Core local swing ±6pp turnout ~67% lead ~71.5pp wins in 100% of simulations

N27 MERLIMAU

Solid

majority ~5,320 votes if won

Defensive Core local swing ±6pp turnout ~65.7% lead ~42.4pp wins in 100% of simulations

N28 SUNGAI RAMBAI

Solid

majority ~5,810 votes if won

Defensive Core local swing ±6pp turnout ~63.4% lead ~57.5pp wins in 100% of simulations

Assumptions in this run

Base case (no changes)

The link carries every slider — shareable and re-checkable.

Engine melaka-mc-2026.09.03-r9 · register v0.6 · link 2026.09.03-r9-v0.6 · voter flow: Manual rule. Conditional sensitivity analysis — not a forecast.

Not a forecast

Conditional scenario — this is NOT a forecast. The tier outcomes depend entirely on the assumptions you choose.

Why — 6 assumption classes

Two PAST benchmarks that disagree (real PRN2021; modelled GE15) — not a range that contains the future (A8).
Vote flows (retained/switched/mobilised) CANNOT be identified from two aggregate elections — the sliders are assumptions, not measured flows (A7).
The new-voter cohort depends on the benchmark (A9): post-2021 at the PRN2021 anchor, post-GE15 at the GE15 anchor. Lean is a flat assumption, not a by-ethnicity measurement.
expTurnoutPct = gross turnout. Coalition-partner voters who refuse to cross over when their party withdraws from contesting (pactLeakage) ABSTAIN — they leave the voter pool, so gross turnout FALLS and the margin is computed over the remaining valid votes (03-next-plan §77).
The age & ethnic weighting dials reweight the GE15 vote mix over the joint (age×ethnicity) cells of the saluran-grain layer (melaka-age-ei-2026) and renormalise, so young/senior and Chinese/Indian shares compose jointly (a young-Chinese scenario is one cell, not two stacked tilts) while gross turnout is held fixed. Effect scales with anchorWeight (the layer is GE15-only — inert at the PRN2021 anchor). The signal is a saluran-level ASSOCIATION between mean registered age and bloc valid-vote share (mixing turnout and choice) — not an individual effect and not a turnout estimate. AGE is well-identified (saluran are age-sorted); ETHNICITY is WEAKLY identified (saluran do not separate ethnicity) — treat the ethnic dial as a registered assumption. Bands are age-at-GE15 (2022); a 2026-cohort reading shifts each band ~+4 years.
A3: only the Bersama vote effect (which shaves PH alone) is modelled — the only additional candidate announced so far. Under configuration D, the remaining OTHER votes are treated as backing PH; vote-direction controls for any further additional candidates are deferred until candidates are announced.