1. System Objective
Narrative Metrics, part of the OddsAccuracy framework, is not a sentiment toy, a generic news feed, or a tipster layer. Its purpose is narrower and more useful:
identify which football entities currently have real narrative force, measure how abnormal that force is, and estimate whether it is likely to persist.
In practical terms, every active window asks three questions:
- How large is the current narrative footprint of an entity?
- How abnormal is that footprint relative to expectation?
- Does the shape of the current window suggest continuation or decay?
2. Ranking Philosophy
The default public ranking used by Narrative Metrics is xN, not raw mention volume. That is deliberate. Raw counts systematically over-reward entities with large baseline coverage.
xN is designed to prioritize narratives with a stronger continuation signature.
Interpretation rule. A high xN does not mean an entity is objectively more important. It means the current narrative configuration shows stronger evidence of continuation than peers in the same window.
Practical tip. One of the fastest ways to understand how Narrative Metrics works is to take screenshots of the tables and run them through an AI assistant for interpretation.
This simple workflow (screenshot → AI → explanation) often reveals patterns, metric interactions, and narrative dynamics much faster than manual reading.
3. Input and Aggregation Model
The engine starts from timestamped media items. Each item contributes evidence to one or more football entities. The public methodology uses entity-window aggregation as the core analytical layer.
| Layer | What happens |
|---|
| Item layer | Timestamped media items are ingested with publication time, source context, and content metadata. |
| Entity mapping | Items are linked to football entities such as teams and matches. |
| Window aggregation | For each active window, item activity is converted into entity-level signals such as mentions, recency, persistence, and diversity. |
| Structural layer | Cluster-level processing generates coherence and quality signals used by CQS. |
| Ranking layer | Window metrics are normalized and blended into xN before publication in the ranked table. |
4. Window Model
Narrative Metrics is explicitly window-native. Every metric is evaluated inside a defined time slice rather than across an undifferentiated historical pool.
The current public build uses 3h, 6h, 24h, 7d, and 30d.
This matters because the same entity can look completely different across windows. A club can dominate the 30-day discussion while showing weak short-term continuation, or it can flash sharply over 3 hours without demonstrating durable persistence.
5. Snapshot Principle
The public UI is intended to represent a coherent snapshot state, not a patchwork of partial updates. That principle matters because narrative rankings lose meaning when window outputs are temporally misaligned.
In other words, the methodology is not only about formulas. It is also about publishing rankings from a synchronized export state that preserves interpretability.
6. Metric Definitions
xNExpected Narrative
xN is the primary ranking signal. It estimates the expected continuation of the current narrative profile.
In practice, xN blends normalized persistence, freshness, structural quality, deviation, source diversity, and positive acceleration.
xTExpected Trajectory
xT measures the short-vs-long horizon trajectory of the narrative.
In practice, xT is used to distinguish acceleration, stabilization, or cooling pressure inside the current narrative state.
StateNarrative State
State is the backend narrative classification assigned to the entity in the current window.
In practice, State is derived from the interaction between xN and xT, not from a separate standalone score.
NDSNarrative Deviation Score
NDS measures how abnormal the current activity is relative to structural expectation.
NPSNarrative Persistence Score
NPS measures how continuously active the narrative remains inside the current window.
The window is split into time buckets; NPS rises when activity persists across the window rather than clustering into a narrow burst.
FreshFreshness Score
Fresh measures how much of the window activity is concentrated in the recent tail.
CQSCluster Quality Score
CQS captures the structural coherence of the current narrative cluster.
Higher CQS indicates that the active narrative is not merely noisy repetition, but shows stronger internal structure and clustering quality.
6A. Insights and Badges
InsightsInterpretive Summary
Insights are the downstream interpretive layer attached to the ranked output.
In practice, they turn the current metric configuration into compact explanatory text about what is driving the narrative in the selected view.
BadgesVisual Signal Flags
Badges are quick UI markers that surface important contextual signals without replacing the underlying metrics.
They are designed to help the reader spot active narrative conditions faster, then verify them against the table and insight text.
DriverInsight Driver Badge
The Driver badge shows the sort-local interpretive driver for the currently selected metric and window.
It is not a standalone score term and it should not be read as the same thing as backend State; it is a downstream explanation surface.
7. What the System Rewards
- Persistent activity across the current window, not just one sharp spike.
- Abnormal activity relative to baseline expectation.
- Recent evidence that the narrative is still actively developing.
- Better structural coherence and source diversity.
8. What the System Does Not Claim
- It does not claim objective truth or real-world importance.
- It does not directly measure sentiment or correctness.
- It does not guarantee future attention; it estimates continuation probability from current evidence.
- It does not replace human interpretation. It sharpens it.
9. Methodological Position
Narrative Metrics should be read as a football narrative intelligence framework. The goal is not to summarize all coverage. The goal is to create a disciplined ranking system that isolates where narrative energy is accumulating, how unusual it is, and whether it is likely to persist.
Bottom line. The public table is most useful when read comparatively. It is designed to answer: Which entities are driving the strongest active narratives right now, and which of those narratives look most likely to continue?