The Turnout Trap — AP Delimitation: Equal-Elector Plan vs EAC-PM Model

3.25 : 1

EAC model — heaviest vs lightest vote in the same state. Built-in institutional inequality.

1.36 : 1

Equal-Elector Plan — all 38 seats within ±16% of equal vote weight; 30 within ±10%.

10.87 L

The State Elector Quota: 4,12,97,733 voters ÷ 38 seats. What one fair seat holds.

18 / 25

PCs the EAC model leaves frozen at 14–18 lakh voters while minting 5.5-lakh seats next door.

25 / 25

PCs where the Equal-Elector outcome lands closer to the quota. No exceptions.

People vs predictions — the core conflict

Why the post-2026 delimitation has moved from administrative backrooms to centre stage

India's first delimitation in five decades will redraw the geometry of the republic itself. A widely publicised EAC-PM working paper (Ravi & Kapoor) proposes an algorithmic method: instead of dividing new seats evenly by population, a predictive statistical model targets specific constituencies — slicing high-density urban pockets into multi-way seats to maximise future turnout and ease booth congestion. Forward-thinking engineering, perhaps — but applied to a nation's sovereign franchise, it accidentally builds systemic structural inequality into the map. Andhra Pradesh, scaling from 25 to 38 seats, is the clearest case study.

EAC-PM turnout model Trains a complex algorithm on past elections to predict where turnout rises most if a PC is chopped into two or three. All 13 new seats go to 7 "high-gain" zones — tripling a few cities — while 18 of 25 constituencies are frozen untouched. Result: tiny hyper-fragmented urban seats beside giant overstretched rural ones; not a single seat lands in the fair zone.

Equal-Elector Plan Throws out predictive guesswork entirely. One sacred rule: every citizen's vote must carry equal weight. Compute the State Elector Quota (≈10.87 lakh voters per MP), blend neighbouring PCs into natural regional clusters, and distribute all 38 seats so every MP represents an almost identical slice of human lives. Every seat hugs the quota; anyone can verify it with a calculator.

Delimitation is not municipal traffic routing or behavioural nudging — it is the geometric layout of a republic's soul. Explore the AP Heatmap (watch the map flip from red to green as you switch plans) and look up your own constituency in PC-by-PC.

Three fatal flaws of the turnout model

Stripped of its mathematical presentation

Flaw A

It fractures "one person, one value"

Frozen Anantapur and Srikakulam stay above 17 lakh voters per MP while freshly split Kadapa and Rajampet daughters hold barely 5.5 lakh — a 3.25-to-1 disparity. One voter is handed nearly three times the political say of a neighbour. Not a rounding error; built-in inequality.

Flaw B

It rewards urban apathy, punishes rural discipline

The paper itself records that rural and Scheduled-Tribe voters post the country's highest turnout while metros suffer "urban apathy". Yet the algorithm showers extra seats on low-turnout cities to shorten their queues — and structurally freezes the disciplined rural and tribal belts that have faithfully stood in line for decades.

Flaw C

It opens a backdoor for opaque political steering

Urban and rural India vote differently. When a multi-variable black box — urban covariates, linguistic-diversity curves — decides which districts multiply and which freeze, the map inherits the politics of those settings. A map citizens cannot check is a map citizens cannot trust.

Andhra Pradesh fairness heatmap — 25 parliamentary constituencies

Colour = distance of each PC's resulting seats from the 10.87 L quota · map numbers follow the selected view

Equal-Elector Plan — PC table

Hover/tap a row ↔ highlights the map · table scrolls inside this panel

Setup & notation

The complete vocabulary of the model

E_i	Registered electors in PC i (ECI 2025 rolls); i = 1…N, N = 25
T	State total = Σ_i E_i = 4,12,97,733
S	Seats allotted to the state = 38
P	State Elector Quota = T/S = 10,86,782 ≈ 10.87 lakh
G_k	Cluster k of neighbouring PCs, k = 1…K, K = 11
B_k	Cluster electorate = Σ_{i∈G_k} E_i
q_k	Raw quota share = B_k/P (may be fractional)
n_k	Integer seats awarded to cluster k
A_k	Realised per-seat size = B_k/n_k
δ	Deviation tolerance, target 0.10 (±10%)
w_k	Vote weight = P/A_k (1 = exactly fair)
D	Disparity ratio = max_k(A_k) / min_k(A_k)

Partition constraints. The clusters must (i) cover all 25 PCs: ∪_kG_k = {1…25}; (ii) be disjoint: G_k∩G_j = ∅; (iii) be geographically contiguous, so daughter boundaries can be drawn from whole assembly segments inside each cluster. This is Hamilton's largest-remainder apportionment with a contiguity constraint — the same mathematical family the Constitution's Article 81(2)(b) "same ratio" command belongs to.

The Equal-Elector model, step by step

Five formulas + the optimisation it solves

1 · Quota

P = T / S = 4,12,97,733 / 38 = 10,86,782

divide the state's voters by its seats. ~10.87 lakh is what one fair seat holds.

2 · Cluster

B_k = Σ_{i∈G_k} E_i ; q_k = B_k / P

add up bordering PCs and ask how many quotas the cluster holds (e.g. Kadapa+Rajampet = 33.15 L = 3.05 quotas).

3 · Apportion

n_k = ⌊q_k⌋ + largest remainders, s.t. Σn_k = 38, n_k ≥ 1

every cluster first gets its whole-number share; leftover seats go to the biggest fractional remainders until the state hits exactly 38.

4 · Verify

A_k = B_k/n_k ; require |A_k−P|/P ≤ δ

the average seat in every cluster must sit within ~10% of the quota; shift seats between clusters until it nearly does.

5 · Objective

minimise max_k |A_k/P − 1| → D → 1 ; w_k = P/A_k

the plan minimises the worst vote-weight distortion in the state (a minimax rule). Achieved: D = 1.36, w ∈ [0.86, 1.17]. EAC model: D = 3.25, w ∈ [0.61, 1.99].

Contrast · EAC-PM objective

maximise Σ_i ΔT̂_i·E_i , T̂ = GAM(size, urban, SC, ST, language)

split wherever a fitted turnout surface predicts the biggest jump, under a seat budget — no parity constraint anywhere. Equality is absent from the formula, so it is absent from the map. Their own predicted gain swings +0.3 → +2.3 pp across four versions of the same model (8×), with per-PC validation error ≈8.6 pp — larger than the effect claimed.

Worked example & full cluster math

Cluster 10 by hand, then all eleven

Worked · Cluster 10

B₁₀ = 16.41 + 16.74 = 33.15 L

q₁₀ = 33.15 / 10.87 = 3.05 → n₁₀ = 3

A₁₀ = 33.15/3 = 11.05 L ; dev = +1.7% ✓

w₁₀ = 10.87/11.05 = 0.983

Kadapa + Rajampet → three near-perfect seats. The EAC model instead 3-way-splits both, flooding this one block with 6 seats of 5.5 L each (w = 1.97) — double the voting power of every neighbour.

k	Cluster	B_k (L)	q_k	n_k	A_k	Dev	w_k
1	Araku+Srikakulam+Vizianagaram	47.82	4.40	5	9.56	−12.0%	1.14
2	Visakhapatnam+Anakapalli	35.27	3.25	3	11.76	+8.2%	0.92
3	Kakinada+Amalapuram	31.67	2.91	3	10.56	−2.9%	1.03
4	Rajahmundry+Narasapuram+Eluru	46.38	4.27	5	9.28	−14.6%	1.17
5	Machilipatnam+Vijayawada	32.42	2.98	3	10.81	−0.6%	1.01
6	Guntur+Narasaraopet+Bapatla	50.38	4.64	4	12.59	+15.9%	0.86
7	Ongole+Nellore	33.21	3.06	3	11.07	+1.9%	0.98
8	Nandyal+Kurnool	34.60	3.18	3	11.53	+6.1%	0.94
9	Anantapur+Hindupur	34.41	3.17	3	11.47	+5.5%	0.95
10	Kadapa+Rajampet	33.15	3.05	3	11.05	+1.7%	0.98
11	Tirupati+Chittoor	33.67	3.10	3	11.22	+3.2%	0.97
	State total	412.98	38.01	38	9.28–12.59	D=1.36	0.86–1.17

Three clusters sit just outside ±10% (widest +15.9% at G6) — a deliberate trade-off to hold the state at exactly 38, curable at assembly-segment grain when real boundaries are drawn. Even the worst cell beats the EAC plan's best frozen PC (+27%). The 2025 rolls proxy population until the 2027 Census substitutes its figures; the clustering logic is unchanged by that swap.

All 38 seats at a glance

Each dot = one proposed seat · shaded band = quota ±10% (9.78–11.95 L)

EAC daughters (5.5–8.6 L) EAC frozen (13.8–17.8 L) Equal-Elector (all 38)

Two separate Andhra Pradeshes under the EAC plan — tiny new seats left, giant frozen seats right, zero inside the fair band. The Equal-Elector Plan puts 30 of 38 inside the band, all 38 within ±16%.

Every PC, both fates, one row

Filter:

PC	Constituency	2025 (L)	EAC turnout model			Equal-Elector Plan
PC	Constituency	2025 (L)	Action	/seat (L)	Dev	Cluster	/seat (L)	Dev

† inferred from the EAC paper's own criterion (it names Visakhapatnam, Kadapa, Rajampet, Nandyal among AP's six 3-way splits). Dev = deviation from quota 10.87 L. green ≤±10% · amber ±10–16% · red beyond ±20%. In all 25 rows the Equal-Elector figure is closer to the quota.

Equal-Elector clusters vs the ±10% band

Per-seat deviation from quota · 3 of 11 slightly outside, flagged honestly

Widest deviation +15.9% (G6) — held to keep the state at exactly 38; curable at assembly-segment grain. Even this worst cell beats the EAC plan's best frozen PC (+27%).

Why simple arithmetic must triumph over predictive modelling

Five reasons, in layman's words

Parity

Absolute voter parity

Every citizen's voice counts equally — vote weight held in a tight uniform band (0.86–1.17). The turnout model hands one voter ~3× another's say by design.

Law

Constitutional compliance

Article 81(2)(b): population-to-seat ratio "so far as practicable, the same throughout the State." The Equal-Elector Plan honours this command directly; the turnout model rejects it to optimise booth logistics — and its own paper concedes it "is not a boundary-drawing plan."

Stability

Mathematical stability

The EAC paper's predicted benefit swings by an eight-fold range (+0.3 → +2.3 pp) across versions of its own model. Long division yields the same answer on every computer, every time.

Neutrality

Complete neutrality

A rule that counts only people cannot be steered by any party or designer. It is perfectly blind to voting patterns, language and geography — no backdoor for political steering.

Transparency

Radical transparency

Replicating the government map needs data pipelines, statistical software and optimisation code. Replicating this plan needs public voter rolls and a basic calculator. A delimitation citizens cannot check is one they will not trust — and trust is the entire point.