Situation Room
Karnataka, 224 assembly constituencies
DEMONSTRATION DATA
--:--:--
Colour Region
Congress standing, statewide
+0
—

Share of voice

7D

Seats by standing

224 SEATS

Seat detail

Click any constituency on the map.
What is moving

The story this week

Narratives ranked by reach and momentum, split by whether they help us or hurt us. Drawn from public social, news and vernacular media, plus the consented citizen network.

Live narratives

Issue salience

Statewide

Where it is landing

The other side

What they are pushing, and where it is working

Opposition messaging tracked the same way ours is. The useful question is not what they said, it is which of their lines is actually being repeated by people who are not them.

Their lines, by traction

Repeat rate outside party accounts
The party

The bench

Real people, real portfolios, as notified. Names, constituencies and departments on this page are factual. No performance score is shown against any named person, because inventing one would be indefensible. Those fields stay empty until they are populated from live listening.

Where this stands today

Us against them

Congress voice share
—
Net sentiment gap vs BJP
—
Seats warming
—
DeployOne ground layer

The voters social listening cannot hear

Disclosed automated research calls in Kannada and regional dialect. This is the half of Karnataka that does not post: older, rural, non-English. It is also where the social-only picture is most wrong.

Where voice and social disagree

Net sentiment
Each row is one issue. Left is what public social says. Right is what citizens said on the phone. The gap is the finding.

Sample reach

How the call runs

0600 hrs

Morning brief

Five things, in order. Written to be read in ninety seconds, or listened to on the drive in.

Karnataka,

Ownership

Every issue, who owns it, and where it is moving

Two things are measured for every issue. Salience is how loudly people are raising it. Sentiment is whether what they say helps you or hurts you. What matters is the combination of the two, which is what the board below shows.

All twelve issues on one board

SALIENCE ACROSS · SENTIMENT UP
Colour

Seat detail

Click a constituency to see what was asked there and what came back.

What the voice agent asks

The exact wording, the reason it is asked, and the rule attached to it. Questions are reviewed before every wave.

What came back

ANONYMISED · SEGMENT TAGGED
Responses are held against a segment and a constituency, never a person. This is the level at which it is useful and the level at which it is defensible.
Who we are actually hearing

Panel and coverage

You can only say Karnataka thinks this if the people you heard from look like Karnataka. This page proves they do. It shows who we heard from, how that compares to the real state, where we are still thin, and who we are calling next to fix it.

Does our sample look like Karnataka?

LOWER IS BETTER · UNDER 3 IS REPRESENTATIVE
Each number is how far the people we heard from sit from the real population, averaged across every group below and measured in percentage points. Zero would mean a perfect mirror of the state.
Online listening alone
—
points off
Over-weights urban, under-35, salaried. This is the picture most war rooms run on.
Voice calls alone
—
points off
Reaches the rural and older half, under-reaches students and urban salaried.
Both, weighted
—
points off
The two channels correct each other. This is the number to quote in the room.

Every group, us against the real state

The bar is the share of our sample this group makes up.
The pale line is the share of Karnataka they really are.
Bar short of the line means we under-heard them. Past it, we over-heard them.

Who we are still under-hearing

The groups furthest short of their real share. Until these close, a read on these people should not be quoted.

So who do we call next

Each gap above turns into a calling wave. Built from where we are deaf, not from a list of people to persuade.

Segments we do not hold

Channel reach by segment

WHICH CHANNEL FOUND THEM
The bars show where each group was heard. Anything heavily oxblood is a group that online listening would have missed entirely.
The shield

Coordinated narrative detection

Some things spread because people believe them. Some spread because a small number of accounts push them hard and fast. In a feed those look identical. Measured, they look nothing alike. This page shows how we tell them apart, and what we are allowed to do about it.

Each dot is one account that has posted or forwarded the claim. The tight cluster on the left is the origin group.
Lines are onward spread from that origin into the wider public, over the first seventy-two hours.
Gold is corrected, meaning a real consented citizen in that area shared a checkable fact and the claim stopped travelling.

Select a narrative

IDLE
Choose one of the tracked claims to replay how it spread.

What we will not do

The first seventy-two hours

HOW IT TRAVELLED

What happens after we detect it

SIX STEPS, ONE HUMAN GATE
Detection is automatic. Every step after it needs a named person to approve, and nothing is published that is not checkable against a public source.
Were we right

Track record

A flag is a written prediction. The system raises it, a named analyst signs it, it goes to a named recipient with a deadline, and it is never edited afterwards. When the deadline passes it is scored against what actually happened, including when we were wrong. A tool that never marks its own homework is a newsletter.

Raised by the system, from a specific signal crossing a specific threshold. The signal is shown on every flag.
Sent to a named role, because a finding nobody receives is not a finding.
Resolved on a date fixed when the flag was raised, so the scoring cannot be moved after the fact.

Flags raised, and what happened

OLDEST LAST

From finding to action

A finding is only worth something if somebody did something. This is who was told, what they did, and whether the number moved afterwards.

Where we were wrong

Kept deliberately visible. The misses are the reason to trust the hits.
↑↓ move ↵ open esc close ? shortcuts

Keyboard control

Command paletteCtrl / ⌘ K
Jump between views1 … 0
Cycle time windowW
Back to state mapEsc
Print the current viewP
This panel?
Press esc to close.
Listening
…
Try “situation room”, “go to Belagavi”, “read the brief”, “thirty days”, “stop reading”.