Methodology

NarrativeControl tracks the daily balance of political narrative in a territory: which issues dominate public conversation, how each one lands for the party or alliance in power against its challengers, and how that balance moves over time.

Coverage is India at national level, major Indian states, and twenty other democracies. Each territory is read on its own terms — its own incumbent, its own opposition structure, its own electoral system.

These readings describe the day's reporting and public conversation. They are not a poll, a seat projection or a measure of voting intention.

NILNet Incumbent Leverage

The headline daily figure. Positive means the day's narrative works for the party or alliance in power; negative means it works against. Scale −5.00 to +5.00.

MA77-day moving average of NIL

The mean of the last seven daily NIL values, today included. The structural baseline a single day is read against. Same scale as NIL.

MA14 / MA3014-day and 30-day moving averages

Longer smoothing windows on the same series, used in the Pro Terminal for trend work.

ICIncumbent Composite

How the day's issues land for the party or alliance in power, weighted by how much each issue was being talked about. Scale −5.00 to +5.00.

CCChallenger Composite

The same calculation for the opposition. Scale −5.00 to +5.00.

Δ (delta)Partisan differential

IC minus CC — the raw gap between the two sides before opposition structure is applied. Scale −10.00 to +10.00.

IOUIndex of Opposition Unity

The figure shown on the fourth headline card, expressed 0–100. Higher means the opposition vote is concentrated in fewer blocs.

ICCIndex of Challenger Concentration

The stored baseline the IOU card reports and the leverage formula consumes. Computed once per territory from its most recent election, not daily.

HHIHerfindahl–Hirschman Index

The concentration measure the unity index borrows: rescale shares to 100, square each, sum the squares.

ShareConversation share

The percentage of the day's political conversation an item accounts for. The twelve displayed themes and the eight daily spikes together total 100; themes below the twelfth are recorded on the same scale but sit outside that hundred.

Intake poolThe day's recorded issue set

Thirty issues are researched and stored each day; the leading twelve are displayed. The remaining eighteen keep a fading issue on the record instead of dropping it.

ContinuityCarried over, returned or new

Each issue is marked against the previous day's list, so a standing concern can be told apart from a story that has just entered the cycle.

7-day share lineWeekly conversation share trace

The small line under each headline issue: its conversation share on each of the last seven calendar days. A day with no reading breaks the line rather than being drawn as zero.

VelocitySocial velocity

HIGH, MED or LOW — how fast an issue is spreading in online conversation on the day.

Macro domainIssue category

One of ECONOMIC, SECURITY, SOCIAL, GOVERNANCE or PUBLIC_GOODS.

IntensitySpike intensity

0–100. How loud a short-life story was in its own right that day. Read alongside its conversation share, which is what actually weights the score.

EscalationEscalation likelihood

LOW, MED or HIGH — the chance a spike story hardens into a standing political theme. Reported for reading the cycle; it does not scale the score.

AttributionImpact attribution

Which side a story actually damages. Fixed by the target of the reporting and by who held office when the events took place, not by who holds office today.

FPTPFirst past the post

Electoral system label stored per territory; it records the competitive context a reading sits in.

The night begins with a separate wire scan of the last seventy-two hours, run on three directed angles: state action against the press and challenges to executive power in court; rulings, subpoenas, raids, regulatory action and resignations; and protests, unrest and the day's breaking wire stories. Whatever that scan finds is handed to the main reading as a mandatory checklist, so an acute event cannot be missed simply because the reading was busy confirming yesterday's standing themes. Anything of political weight must then appear either as a spike or inside the context sentence of the theme it belongs to.

Once per night, each territory gets one grounded research pass against current reporting. It returns the thirty issues dominating that territory's political conversation that day; the screens show the leading twelve, and the remaining eighteen sit just below the line so a fading issue can slip back without disappearing from the record. Each issue carries:

• Conversation share (S) — its percentage of the day's political conversation. The twelve displayed themes and the eight daily spikes together total exactly 100.
• Macro domain — economic, security, social, governance or public goods.
• Social velocity — how fast it is spreading: HIGH, MED or LOW.
• Incumbent impact (I) and challenger impact (C) — two independent ratings from −5 (badly damaging) through 0 (neutral) to +5 (strongly helpful), scored separately for each side.

The agenda list holds standing themes, not headlines. A theme is a continuing fault line that would still be recognisable a month later — the cost of living, jobs, the farm economy, internal security, public services, corruption and institutional integrity, federal relations, reservation policy, foreign and trade policy. A single incident — a court order, an FIR, an arrest, a death, a resignation, a rally or a statement — never takes a row of its own on the agenda.

Breaking events are recorded separately as the day's eight spike issues, and are absorbed into the theme they belong to: the theme keeps its name, its velocity and impact move, and the event is named in its context line. The same story is never listed in both places. So a fresh agency case against a leader moves the standing integrity theme rather than replacing it, and the case itself sits in the spike register.

Two balance rules stop one loud day from rewriting the agenda. At least one of the leading twelve must come from each of the economic, public goods and social domains, and no domain may hold more than five of the twelve. No single theme may take more than a fifth of a day's conversation, and living-standard themes are not pushed below three per cent by one news cycle.

Nothing is invented: a figure only enters the model when the underlying reporting supports it. The pass also records the talking points each side actually pushed that day.

The pass is not blind to yesterday. The previous day's list is supplied to the research step, which must reuse the same wording for an issue still live and keep a standing concern on the list at a lower share rather than dropping it because a bigger story broke. Each issue is marked as carried over, returned or new, so a long-running concern such as inflation shows a continuous weekly line instead of blinking in and out.

Wording is stabilised on the way in. When the research step rewords an issue that is still the same story, the previous day's wording is kept as the canonical title, so the weekly line stays joined rather than splitting into two half-series.

The screens show twelve; the record keeps all thirty. The 7-day share line under each headline issue reads the whole pool, so an issue that slipped to rank twenty on a busy day still shows its real share for that day instead of a gap.

An impact rating is only meaningful if the right side is held responsible. Before any issue is scored, the reading establishes who the story targets and who was in office when the events took place, working from the territory's recorded roster: the governing side and each opposition bloc, with their parties and vote shares from the last national or state election on file.

Investigations and enforcement. A case, arrest, raid or inquiry is not treated as a failure of the government simply because it is a governance story. When an investigative or enforcement agency acts against opposition figures or an opposition-run administration, the opposition is scored as being on the defensive and the governing side as being on the offensive. The signs reverse when the same action reaches serving ministers, officials or allies of the government.

Backfire. Enforcement action counts against the government only where reporting shows a concrete backfire — a court quashing the case or naming political motive, the evidence collapsing, or broad public resistance to the action itself. The context line has to say so.

Historical responsibility. When a story revives an event or decision from an earlier period, liability sits with the party or coalition that held office at the time, even if it now sits in opposition.

National action in a state reading. A national agency acting inside a state is treated as an external intervention by the national governing party, scored on its local effect: an asset where the state government is aligned with it, a source of sympathy for a regional party where reporting shows resentment of central overreach.

Incumbent accountability. Prices, jobs, public services, exam leaks, infrastructure failures, law and order, and misconduct by serving ministers stay with the incumbent, scaled by how directly reporting ties the failure to the government.

The two ratings are not mirrored by reflex: both sides can lose on the same day, and a story can be close to neutral for one of them. Each context line names the target of the story and which side is pressing the attack, so the sign of every rating is checkable against the reporting.

Outlets are not interchangeable. Every publication cited in a reading is placed in one of three groups by its reported editorial posture towards the government of the day:

• Pro-government — outlets aligned with or sympathetic to the governing side, including party organs and ideological platforms.
• Neutral — wire services, mass-circulation dailies and broadsheets of record that carry the consensus account.
• Opposition-leaning — outlets aligned with or sympathetic to the opposition, including opposition party organs and adversarial platforms.

This is a description of editorial posture, not a judgement of accuracy. Alongside it, each outlet carries a reach tier: national mass reach, regional reach, or niche and partisan. A story alive only on party platforms is cadre conversation, not mass public conversation, and it is given a materially smaller conversation share than a story on the national front pages.

Consensus anchor rule. The strategic assessment of every issue must open with a neutral, mass-reach outlet. A party organ or niche ideological platform may appear in the source list, but it never speaks for the fact itself. Where no neutral outlet carried a claim, the assessment says so and the claim is treated as contested.

Spin decomposition. Each issue also records how the same event is framed on each side of the press: the neutral factual baseline, the pro-government framing and the opposition-leaning framing, each attributed to the outlet that used it. Partisan wording lives there, never in the assessment. Where one side of the press did not cover a story, that frame is left empty rather than invented. On the radar, the source chips under each assessment show the stance of every outlet, and “Compare spin” opens the three framings side by side.

Each side's composite is the share-weighted mean of its own impact ratings across the twenty scored items of the day — the twelve displayed themes and the eight spike stories — so a loud item counts for more than a quiet one. Those twenty shares total exactly 100, which is why the two readings are directly comparable across days and across territories. Themes ranked below the twelfth are a watchlist: they are recorded and charted, but they sit outside the hundred and outside the composites.

Each share is first deflated by the reach of the outlets that actually carried the story (section 5): mass-reach national reporting counts at full weight, regional reporting at 0.60, party organs and ideological niche platforms at 0.25. The composites are then normalised by the sum of those effective shares, so a story alive only in a cadre weekly cannot swing a national reading the way a front-page story does. When every item of a day is carried at the same reach, the weighting cancels out and the figures are identical to the plain share-weighted mean.

Eᵢ = Sᵢ × Wᵢ
IC = Σ (Eᵢ × Iᵢ) / Σ Eᵢ
CC = Σ (Eᵢ × Cᵢ) / Σ Eᵢ

i  = each of the 12 themes and 8 spikes
Sᵢ = conversation share of item i  (the 20 shares total 100)
Wᵢ = reach weight of item i        (1.00 mass | 0.60 regional | 0.25 partisan)
Iᵢ = incumbent impact of item i    (−5 … +5)
Cᵢ = challenger impact of item i   (−5 … +5)
Range: −5.00 … +5.00

Item                      S     I     C
Inflation      (theme)   45   −4    +4
Border row     (theme)   25   +3    −2
Welfare push   (theme)   20   +2    +1
Agency raid    (spike)   10   +2    −3
                        ———
                        100

IC = (45×−4 + 25×+3 + 20×+2 + 10×+2) / 100
   = (−180 + 75 + 40 + 20) / 100  =  −0.45
CC = (45×+4 + 25×−2 + 20×+1 + 10×−3) / 100
   = (180 − 50 + 20 − 30) / 100   =  +1.20

The raw gap between the two composites, before opposition structure is applied.

Δ = IC − CC
Range: −10.00 … +10.00

Continuing the example:  Δ = −0.45 − 1.20 = −1.65

The eight daily spike issues are not a surcharge added after the fact. They sit inside the same hundred points of conversation as the twelve displayed themes and enter the composites on exactly the same terms, weighted by their own conversation share. A spike that owns eight per cent of the day's conversation moves the reading as much as a standing theme on eight per cent would.

Each spike still carries its own intensity, how many consecutive days it has run and its escalation likelihood — the chance it hardens into a standing theme. Those are reported for reading the cycle; they no longer scale the score. A seven-day line under each spike traces its conversation share across the week it ran, so a story that is fading can be told apart from one that is building.

The dashboard reports a spike contribution figure for transparency: the difference between the differential computed on all twenty items and the differential of the twelve themes alone.

spike contribution = Δ(12 themes + 8 spikes) − Δ(12 themes)

A fragmented opposition cannot convert a bad day for the government into pressure as effectively as a consolidated one. The index measures how concentrated the challenger field is — and it is measured at alliance level, not party level, because parties contesting together function as one bloc.

1. Take every party above 1% of the vote at the territory's most recent election.
2. Group parties into the alliance they contested under; a party contesting alone is its own bloc of one.
3. Remove the governing alliance.
4. Rescale the remaining bloc shares so they total 100.
5. Square each rescaled share and sum the squares. That sum, 0–100, is the index.

ICC = Σ (bᵢ)²   where  bᵢ = blocShareᵢ / Σ blocShare
Range: 0 … 100

100  = one opposition bloc holds the entire challenger vote
low  = the challenger vote is split across many blocs

Opposition blocs (raw vote share):
  Bloc A  37%      Bloc B  12%      Bloc C   6%
Total opposition = 55%

Rescaled:  A = 67.3   B = 21.8   C = 10.9
ICC = 0.673² + 0.218² + 0.109²  (×100)
    = 45.3 + 4.8 + 1.2  =  51.3

The index is recomputed only when a territory's election data is refreshed, not every night — it describes structure, not the news cycle.

Leverage converts the day's differential into an incumbent-facing figure, adjusted by opposition concentration. The adjustment is deliberately asymmetric.

ICCn = ICC / 100                    (0 … 1)

Δ ≤ 0 :  NIL = (Δ / 2) × ((1 + ICCn) / 2)    a united opposition punishes harder
Δ >  0 :  NIL = (Δ / 2) × (1 − ICCn / 2)     a united opposition cushions a good day

Result clamped to  −5.00 … +5.00

In words: on a bad day, a concentrated opposition amplifies the damage, while a fragmented one absorbs it — the government pays less for the same narrative because no single challenger can bank it. On a good day the logic reverses: a united opposition limits how much the government can convert.

Δ = −1.65,  ICC = 51.3  →  ICCn = 0.513

Δ ≤ 0, so:
NIL = (−1.65 / 2) × ((1 + 0.513) / 2)
    = −0.825 × 0.7565
    = −0.62

Same day with a badly split opposition, ICC = 15:
NIL = −0.825 × ((1 + 0.15) / 2) = −0.47

A single day can be loud without being structural. The moving average is the plain arithmetic mean of the most recent daily leverage values, today included.

MA7 = (1 / n) × Σ NIL over the last n days,  n = min(7, days on record)

Ordered oldest to newest; today is the last term.
A partial window averages the days that exist rather than padding with zeroes.
MA14 and MA30 use the same formula with windows of 14 and 30.

Last seven daily NIL values:
  −0.42  −0.31  −0.55  −0.18  −0.62  −0.47  −1.10

MA7 = (−0.42 − 0.31 − 0.55 − 0.18 − 0.62 − 0.47 − 1.10) / 7
    = −3.65 / 7
    = −0.52

Whenever history is filled in or rescored, the whole series is recomputed from the stored daily values, so the averages always match the days on record.

• Sign. Positive favours the incumbent, negative favours the challengers. Zero is a balanced day, not an absence of news.
• Typical range. Most days sit between −2.0 and +2.0. A move of 0.5 against the weekly average is a meaningful shift; a move beyond 1.0 usually means a single dominant story.
• Compare over time, within a territory. Because the unity index and electoral structure differ by country, the figures are built for tracking a trend inside one territory rather than ranking territories against each other.
• Read the day alongside the agenda. Each headline card opens a written read naming the issues pushing it up and down, paired with what each side actually said that day.

Research is grounded in current reporting. Every strategic assessment and every spike note opens by naming the media organisation that reported the claim, and every agenda topic and spike story also carries its list of reporting outlets, printed beneath the assessment on the dashboard. Outlets are written out in full — never reference numbers, never an unnamed "reports say". Where a story moved during the day — an announced strike, march or resignation that was later called off, deferred or settled — the reading records the latest verified state for that date, not the earlier announcement, and the same event never appears twice in conflicting states. Readings are written once and stored, so every user sees the same figures for the same day.

Every reading is also checked for currency. Before anyone is named, their office is verified for that date, so a leader who has since been elevated or replaced is never described by an old title. The recorded election result is treated as a baseline for vote shares only, not as a statement of who governs today, because coalitions shift and heads of government change between elections. A storyline from an earlier cycle — a settled leadership contest, a withdrawn resignation offer, a patched-up coalition rift — may only appear if reporting from the week before that date shows it is live again.

04:00 IST   daily reading for every territory
04:45 IST   gap fill — any territory or day missing data
05:15 IST   written score explanations
05:30 IST   catch-up pass — re-runs anything still missing

The day is complete before 06:00 IST.

The public radar shows a rolling 15-day window. Pro accounts read the full archive, longer horizons, territory comparison and data export in the Terminal.

A trailing window can also be re-researched after the fact. Those days are re-read one at a time, oldest first, so each inherits the corrected day before it, and the moving averages are resynced once the pass finishes — a corrected history stays internally consistent.