Most people can feel when a field is narrowing before they can prove it. Stream is for that moment: what keeps getting repeated, who benefits from that repetition, what is shelf-driven, and what deserves to slow down before it hardens into common sense.
Built for people who make, buy, study, ship, or question media. Some lanes here are synthetic and clearly labeled. Some are public and checkable. The point is not to sound certain. It is to help you move from a hunch to a defensible read without losing the people underneath the data.
Resonance Surface
Signal, drift, consequence, proof.
This front door is meant to help people who do not live inside charts all day. The note field shows movement, repetition, and pressure in plain sight before you ever open a table.
Current tempo2026present-day slice
♩signal
♫drift
♪proof
♬memory
What to look forIf a pattern survives across lanes, it is usually worth more than a polished explanation of it.
Why music mattersHits feel intuitive, but the path around them often is not. Repetition, shelf placement, timing, and scale all leave a trail.
Where governance entersGovernance matters where a weak pattern starts getting repeated like it is neutral, natural, or inevitable.
Synthetic base clearly labeledReal YouTube music moduleStatic preview + methods notebook
People do not need another dashboard telling them what to believe. They need a place to see whether attention is broad or bottlenecked, whether a hit is being helped along by scale, and whether a system is repeating something because it is easy rather than because it is good. Governance matters here because someone has to keep those boundaries honest.
Published Work
Open research should be useful, legible, and sturdy enough to argue with.
Traceable Evidence
The synthetic lanes stay labeled. The public music lane stays separate. You should always know what kind of evidence you are looking at.
Useful Friction
A good read should sharpen judgement. You should leave with better questions, not just prettier numbers.
Capability Map
What Lives Where
Different questions deserve different surfaces.
Different people arrive here with different stakes. Use the lane that matches the decision you are trying to make, then cross-check it with the others when you need more proof.
Overview
If you only have two minutes, start here
The overview stitches the field map, the decision studio, and the governed lane together before the details take over.
Data Explorer
Check the slice before you repeat the story
Use this when you want the raw rows behind the summary and need to see whether the pattern survives context changes.
Findings
See what actually held up
This is where the live analysis turns into cautious conclusions that still have to survive disagreement.
Pattern Library
Learn the shapes numbers tend to hide
Use the library when you need language for what keeps repeating on screen, in markets, or across recommendation pressure.
Methods
Know what was counted, cleaned, and left out
This lane is for people who want to know what was measured, what was excluded, and how the claims were kept smaller than the evidence.
Decision Grade
Turn the surface into a practical read
The scorecard is the compressed view: useful when a team needs a decision posture, a consequence read, or a quick comparison path.
Real Music
Read public attention on its own terms
The music field is distinct on purpose so songs, channels, virality, and market exposure can be read without synthetic bleed-through.
Contribute
Improve the thin spots instead of repeating the obvious
The repo is open because public research gets stronger under pressure. If something feels brittle, incomplete, or too neat, that is real contribution surface.
Each card is drawn from the live backend or the local ledger. The point is not spectacle. It is to let you see, quickly, what is active, what is partial, and where the next useful proof belongs.
This surface reads a rolling stream of exposure, recommendation, and listener response events. When the local table is empty, it falls back to a marked demonstration window instead of pretending to be live.
Pressure band—
Waiting for the event window.
Current window—
No event window loaded yet.
Discovery breadth—
Genre entropy is still loading.
Independent lift—
Independent share is still loading.
Reading the operating window…
Waiting for live metrics.
Current watchpoints
Waiting for the current window.
Ingress contract
Loading batch limits, retention, and filter lanes.
Waiting for the event contract.
Read The Surface In Layers
Start with the shape of the field before you start hunting for a verdict. Stream is arranged so you can notice what looks broad, what looks staged, and what starts to feel overstated without having to force everything into one score.
The cleanest way to read it is by lane: modeled representation first, governed response second, public attention third. That separation is deliberate. It keeps the evidence honest, keeps the public music read distinct, and makes it easier to tell when a conclusion is carrying more confidence than the data earned.
Lane 01
Representation surface
Use the overview, explorer, findings, and pattern library to study who appears, who stays central, and where narrowness keeps repeating.
Lane 02
Governed response
Use the split-screen demo and decision studio when you need to see how evidence boundaries, risky output, and routing logic behave in motion.
Lane 03
Public attention field
Use the real music lane when the question is market pressure, virality, exposure concentration, or how attention converts into business consequence.
How To Use It
Start with the lane, not the chart: decide whether you are reading representation, governance behavior, or public attention first.
Move the year: watch one system change instead of comparing disconnected screenshots of progress.
Filter the field: ask whether a pattern is broad or tied to a platform, genre, or production culture.
Use methods with findings: the field knowledge and the math are meant to correct each other.
Switch to real music on purpose: use the public-data lane when the question is attention markets rather than modeled representation.
Change the slice, then check whether the story still holds.
Every adjustment recomputes the visible metrics. The main representation dataset is synthetic and seeded on purpose so the method stays inspectable. The separate real-music lane is built on public YouTube data and stays labeled as a different evidence surface everywhere it appears.
Wrap one request, then watch the difference between a policy reaction and a governed response path.
The public repo needed something people could use immediately. This lane does that without pretending to be a private runtime. Type a prompt, try one of the nuisance presets, and compare what a typical wrapper lets render versus what a deterministic guard can stop before the risky fragment lands.
Loading the public preset surface…
Immediate nuisance—The near-term headache this surface is designed to remove.
Why adopt it—A practical reason someone would drop this into an existing workflow today.
Package entry—The tiny public wrapper surface behind this demo.
Rendered risky tokens—How much risky material reached the screen before the block landed.
Reaction window—The delay between risky generation and policy reaction in this demo.
LOOPCHii Lite
Risk is frozen before the fragment resolves
Waiting for a prompt…
No simulation running yet.
Rendered risky tokens—This should stay at zero whenever the governed path blocks correctly.
Intercept window—The simulated cut-off point before the risky fragment would have landed.
Live command surface
Read the field before you name the failure.
This surface is here to make the logic legible. Instead of dropping you into five unrelated charts, it shows how the current slice behaves as one environment: attention pressure, representation balance, and dialogue allocation move together, or they do not. The point is to help a critical reader see the pattern before they carry it into a verdict.
What it is doingTracking the active platform, genre, year, and sample size while the dashboard recomputes its main equity surfaces.
How to read itStart broad with the field map, then test whether the same signal survives in networks, time, and role distribution before calling it structural.
Why procurement caresA claim that cannot explain its own conditions is expensive. This flow keeps the reasoning attached to the evidence.
Move left to right: map the gap, stress-test it, then collapse it into a judgment.
Use Stream like a system
Start with the decision you need, not the tab you happen to open first.
Most public analysis repos become hard to use the moment someone asks what to do next. This layer keeps Stream practical: choose who you are reading as, choose what you need next, and the surface will route you through the strongest evidence path already inside the repo.
This is not a chatbot and it is not guessing. It recomputes from the current slice you selected above, then sends you to the tabs and sections most likely to answer the question honestly.
Current sliceAll platforms · all genres · 2026The guidance changes as you change year, platform, genre, or focus.
Working modeEvidence before narrativeUseful when you need a route that can survive scrutiny.
Best fitResearch, procurement, and product reviewAlso routes creators and contributors into the real-data music and build surfaces.
Read as
Need next
Researcher · measure imbalance
Start with the broad field, then see whether the gap survives the breakdown.
Loading decision route…
SignalLoading…Preparing route context.
RoutePreparing the strongest path through the existing surfaces.
QuestionsLoading prompts…
Multi-Dimensional Bias Analysis
Visual Encoding System
Position: X-axis shows female lead percentage (0-100%), Y-axis shows diversity index (0-10)
Color: Each platform-genre combination has a unique color - platforms define the base hue, genres create shade variations
Pattern Analysis: Look for clustering by color family (platform bias) or by shade intensity (genre bias). Before you call a point biased, compare it against year, platform, genre, media type, and sample size. Bubbles in the upper-right quadrant represent the stronger cases after those checks.
Decision questionIs the imbalance broad across the field, or mostly tied to one platform-genre pocket?
What strengthens the caseClusters that stay coherent after year, platform, and genre filters are changed deserve more weight.
When to slow downA dramatic bubble with a small cohort can be directionally useful without being ready for a procurement claim.
78.4%
Overall Diversity Score
↑ 33.4% from 2015
Shannon entropy across all demographics
0.67
Gender Parity Index
↑ 0.32 from 2015
1.0 = perfect parity, current trajectory reaches parity by 2031
6.2
Ethnic Diversity Index
↑ 2.7 from 2015
Simpson's diversity index (0-10 scale)
3.8
Age Distribution Score
↓ 0.4 from 2015
Age representation declining for 50+ demographic
3D Representation Landscape
Three Dimensions of Equity
Drag to rotate, scroll to zoom. Each point is a platform-genre combination positioned by female lead share (x), diversity index (y), and dialogue equity (z). Marker size grows with overall equality, so you can tell whether a gap is broad, isolated, or only visible after another variable is removed from the frame.
What changes hereThe third axis reveals whether a point that looked healthy in 2D only stayed healthy because dialogue was hidden.
Reader lensFor technical and procurement readers, this is the fast check for whether the earlier story survives dimensional pressure.
Character Interaction Networks
Network Topology Analysis
This force-directed graph reveals demographic clustering in character interactions. Node size = screen time, color = gender (cyan/magenta/gold), links = interaction frequency. It helps answer whether the same groups keep being centered together, or whether the clustering changes once you slice by platform or genre.
Key Pattern: Strong homophily (similar characters interact more) can mean limited cross-demographic storytelling, but we still check whether it survives after sample-size and genre differences are accounted for.
Operational questionDoes the platform have one exclusion point, or a full social topology that keeps recycling the same centre?
Why it mattersA narrow network does not just change representation. It changes whose point of view becomes normal to the audience.
Operational Readouts
Live readout
Central character bias
Field map
73% of central characters in action genres are male.
Why it mattersIf the centre of the story keeps narrowing, top-line diversity can look better than the actual distribution of power.
Reader cueUse this as a staging signal, not the whole case. Then check whether the same bias appears in dialogue and network centrality.
Live readout
Dialogue imbalance
Voice share
Female characters receive 34% less dialogue time on selected platform.
Why it mattersA role can appear often and still be sidelined if it is structurally denied language, explanation, or memory.
Method noteChecked against platform and genre slices so a format effect is not mistaken for a universal one.
Live readout
Age representation gap
Temporal drift
Characters over 50 appear 68% less frequently in year 2024.
Why it mattersCatalogs can modernise on one dimension while quietly flattening generational range on another.
How to use itRead this beside the decade chart. A single year can be noisy; repeated decline is harder to ignore.
Live readout
Intersectional disparity
Combined lens
Women of color represent only 12% of lead roles.
Why it mattersTop-line parity can improve while the most compounded groups remain structurally distant from lead power.
Procurement noteThis is where superficial vendor claims often break, because combined categories expose what summary dashboards smooth away.
Temporal Evolution Analysis
Decade-Long Trends
Track four key metrics from 2015 to your selected year. Diverging trajectories reveal uneven progress - gender parity may improve while age representation stalls or declines, which is why we read trends alongside concentration and intersectional checks instead of treating one line as the whole story.
What to watchThe direction matters less than whether multiple lines improve together or one metric is masking another.
Enterprise useTrend stability is what makes an observation easier to budget around, publish, or take into an internal review cycle.
Role Distribution Patterns
Character Archetype Analysis
Radar chart reveals typecasting patterns. Male characters dominate protagonist/antagonist roles, females are overrepresented as romantic interests, and non-binary characters remain underrepresented across categories. The radar helps show whether a platform has one weak spot or a full pattern of narrow casting.
Why this mattersTypecasting is often the mechanism underneath a broader representation claim. The surface counts can improve while the role map stays old.
How to read itA balanced shape suggests range. One heavy spike suggests a platform is repeating one safe narrative assignment.
Technical Implementation — What Actually Runs
Metrics are computed live by the Python backend: Shannon entropy for diversity, a distance-from-parity index for gender balance, chi-square statistics for dialogue / age / racial bias, intersectional representation ratios against a uniform baseline, and assortativity coefficients on character interaction graphs (NetworkX). Before a claim lands, the app also checks whether the same signal survives when you slice the data by year, platform, genre, and media type. Visualizations use a real D3.js force simulation (charge, spring, and collision forces), Chart.js canvas rendering, and Plotly.js including WebGL-accelerated 3D. The demo dataset is synthetic, generated by a seeded, documented statistical model — see the Learn tab for the honesty note.
Explore in sequence
Start with rows, then compare format, then isolate genre.
Character Data Explorer
Live Dataset
Browse the character-level records behind every metric. Filters query the backend API directly.
The same equity metrics computed per media format — from series and film to anime, stand-up, talk shows, kids' content and sports docs. The donut shows the live catalog mix; the table shows how equity differs by format. Different formats have different production cultures, so bias patterns differ too.
Media Type
Diversity
Gender Parity
Avg Sentiment
Sample
Loading…
Platform × Media Type
Each streaming platform gets its own breakdown, so film, series, docuseries, animation, reality, and talk-show patterns can be compared side by side. This is the cleaner way to read whether a problem is platform-wide or format-specific.
Platform
Media Type
Diversity
Gender Parity
Avg Screen Time
Avg Sentiment
Lead Share
Sample
Loading…
Genre Analysis
Computed Per-Genre Metrics
Diversity, parity, male lead share, and the dialogue gap computed live from the current dataset.
Genre
Diversity
Gender Parity
Male Lead Share
Dialogue Gap
Avg Screen Time
Sample
Loading…
Read the claim in layers
First the generated signal, then the measured breakdown, then the combined-category gap.
Generated Insights
Computed From Live Data
These findings are generated by the analysis pipeline from the current dataset — re-run the analysis to regenerate them. The page does not stop at a single score; it compares platform, genre, media type, year, and concentration so the conclusion has a reason behind it.
What you getA plain-language readout that still stays attached to the structure that produced it.
What not to doDo not treat one sentence as the whole audit. Each insight is only as strong as the slices it survives.
Signal pressurePreparing live summary…The strongest live signals and their business or social consequence will appear here.
Loading insights…
Bias Breakdown
Four Bias Dimensions
Sentiment, screen time, age, and racial dialogue bias — each computed live with the methods described in the Learn tab. We use more than one lens so a “biased” label is not just a vibe; it is checked against effect size, sample size, and whether the gap still appears after slicing the data a different way.
LoadingPreparing measured dimensions…The live breakdown cards will describe the measured gap, its direction, and why it deserves follow-up.
Intersectional Representation
Ratio vs Population Baseline
Groups with a ratio below 1.0 are underrepresented relative to a uniform baseline. The gaps below only appear when gender, race, and age are analyzed together, which is why the dashboard treats intersectionality as a separate check instead of a footnote.
LoadingPreparing combined-group gaps…The underrepresented groups will surface here with ratio and baseline context.
Field library
Bias rarely arrives with one dramatic name. It arrives as a pattern teams have learned not to notice.
The library is here to widen perception, not just label mistakes. Some entries are measured live by Stream. Others matter because they explain how blind spots sneak in before anyone has enough numbers to feel confident.
Design teamsProcurement readersResearchers under deadline
Bias families—Distinct categories so people can separate identity, casting, language, and structural pressure.
Measured live—Entries already tied to a live signal inside the public Stream surface.
Visible now—The currently surfaced subset after category and search filters are applied.
Library modeBroad scan
Search lensAll categories visible
Reader cueMove from labels to examples, then test whether the same pattern shows up in the measured tabs.
Use the library in order
Scan the field, trace where the harm enters, narrow with search, then use the guide to decide what deserves a deeper check.
The Bias Library
50+ Named Biases, Explained
Every entry is a documented bias pattern studied in media-representation research — what it is, what it looks like, which assumptions it can distort, and why it matters. Entries marked measured here are quantified live by the Stream pipeline; the rest are field knowledge to sharpen what you look for. Filter by category or search.
Spot it
Do not begin with certainty.
Look for the repeating distortion first: who gets centred, who gets softened, who gets denied context, who gets treated as default.
Challenge it
Ask what would make the pattern disappear.
A useful bias label is one that survives alternative explanations like genre, sample size, period, or format.
Use it well
Some biases need metrics. Others need restraint.
The library is strongest when it sharpens human review, not when it encourages people to flatten every judgement into one score.
Bias Harmonics
Where a pattern enters, how it spreads, what someone can do
This layer turns a named bias into a systems trace. It does not claim hidden private instrumentation. It uses the documented bias library plus the live Stream metrics to show where a pattern usually enters a media pipeline, how repetition gives it force, and what creators, operators, buyers, and the public can actually do with the result.
—Wave amplitude
—Spread
—Persistence
—Drift
Select a bias card to trace how it enters and travels.
Trace the distortion before it feels normal
The point is not only to name a bias. It is to notice where it entered, who feels it first, what keeps it alive, and which intervention arrives early enough to matter.
Loading bias route…
Preparing the propagation view…
Waiting for the bias propagation surface.
Why it matters will appear here.
Fundamental—The base signal this bias distorts.
Dissonance—How the distortion becomes legible.
PipelineLoading…Preparing route
Representation—Pressure score
Credibility—Pressure score
Action route
Loading role-specific action path…
Loading library…
Method studio
Metrics are useful when they sharpen judgment, and dangerous when they replace it.
The Learn tab is where Stream shows its hand. Each card is here to explain not just what a method calculates, but when it misleads, what it hides, and what you should check next before sounding more certain than the data deserves.
Explainability firstCheck the tradeoffsUse with human judgment
Method cards—Distinct statistical lenses, each mapped to what it notices and what it can miss.
Try-it prompts—Hands-on prompts designed to turn definitions into operational intuition.
Reasoning postureSmaller than the evidenceThe standard here is not bravado. It is clarity about where the metric stops being enough.
Current modeMethod constellation
What it favoursComparative judgment over single-score theatre
Reader cueUse these cards to pressure-test the dashboard instead of treating it like a finished answer.
How to learn from this surface
Start with posture, inspect the data platform, read the tradeoffs, then pressure-test the method cards.
Learn the Methods
Understand Every Metric
Each card explains one of the statistics Stream uses, but the point is not to rehearse definitions. It is to see where a metric is sturdy, where it overreaches, and when another lens says something more useful.
Read it
Ask what the metric notices first.
A useful statistic gives you a clean sensitivity: concentration, parity distance, assortativity, divergence, trend strength.
Pressure it
Ask what would fool it.
Small samples, category imbalance, structural confounders, and overconfident thresholds can make a method look smarter than it is.
Translate it
Turn the number back into a human question.
The best use of a metric is to improve the next review, the next procurement question, or the next design decision.
Working Data Platform
Contracts, lineage, and claim discipline
This section translates the repo into data-engineering language. The point is not to cosplay a warehouse. It is to show the real operating system underneath the visuals: what lands raw, what gets standardized, what becomes a gold metric, and what quality gates stop a persuasive chart from outrunning the data that produced it.
Data platform
Enterprise readers should be able to see the grain, the guardrails, and the serving boundary in one pass.
Stream now exposes a live contract surface for both the synthetic representation lane and the public music lane. That means row grain, partitions, quality checks, and serving posture are visible instead of being buried in code comments or left as assumption.
Hybrid ETL + ELTGold metrics only after checksStatic + API parity
Operating model—How the repo behaves as a working analytical system.
Datasets tracked—Contracted datasets surfaced into the public analysis layer.
Quality gates—Checks on keys, nulls, leakage, coverage, and freshness posture.
Rows under management—Observed row count across the contracted public lanes.
Landingraw cohorts
Standardizetype coercion + dedupe
Goldmetrics + confidence
ServeAPI + static + UI
Lineage stages
Loading stage topology…
Loading lineage…
Operational controls
Reading freshness, parity, and delivery posture…
Loading control plane…
Governance contract
Reading fairness, review posture, and public-boundary signals…
Loading governance contract…
Quality gates
Checking contract health…
Loading quality checks…
Contract explorer
Reading row grain, keys, partitions, and schema posture…
Loading active contract…
Reproducibility catalog
Reading deterministic seeds, public-scope boundaries, and reproducible artifacts…
Loading reproducibility catalog…
Loading content…
From evidence to decision
Grade the platforms, size the gaps, then test the rewrite.
The Verdict
From numbers to a grade you can act on
This is where the analysis becomes a decision. Every platform is graded A–F across four representation dimensions, but only after the app checks whether the same signal persists across year, genre, media type, and concentration. The concentration of screen time is measured with a real inequality curve, each bias is sized (not just flagged), and the decade-long trend is fitted with a regression line. Then you can rewrite the cast yourself in the What-If simulator. Every figure is computed live from the labeled synthetic dataset.
Platform Report Card
Grading platforms…
What-If: rewrite the cast
Gender parity is 1 − |0.5 − female_share| × 2. Drag the slider to set a hypothetical female lead share and watch the parity index and grade recompute — the value is returned by the backend, exactly like the live metrics. It is a simple example on purpose: you can see how the decision changes when just one variable changes, then compare that with the larger pattern above.
Female 50%Male 50%
Source: GET /api/simulate/parity?female_ratio=
A+
Parity 1.00
At or near a 50/50 cast — parity is effectively balanced.
Inequality Lab — who gets the screen time?
A Lorenz curve plots the share of total screen time captured by the least-visible characters. The diagonal is perfect equality; the deeper the curve sags, the more screen time concentrates in a few roles. The Gini coefficient (0 = equal, 1 = one character takes everything) and the Theil index summarize that gap.
—
Screen-time Gini
—
Dialogue Gini
—
Screen-time Theil
—
Dialogue Theil
How big is each bias? (effect size)
A chi-square test tells you a dialogue skew is unlikely to be chance, but its value balloons with sample size. Cramér’s V rescales it to 0–1 so you can compare magnitudes and read a plain-language band (negligible → large).
The decade trend, fitted
Instead of comparing only two years, this fits an ordinary least-squares line across every year in the dataset. The slope is the average change per year; R² says how much of the movement the straight-line trend explains.
Real YouTube Data
Music Virality
A deep statistical read of the 100 most-watched music videos on YouTube in 2025, paired with a broader 636-song public context cohort so the structure is not mistaken for a leaderboard quirk. We look at more than the headline view number: channel size, duration, collaboration, language mix, publication timing, and tag structure to explain why a track rose or stalled. Every number below is computed from public source rows in the repository, not synthesised.
Source:Loading…
Loading the resonance field…
LoadingPreparing the field
Pulling the first decision cues from the real music dataset.
Read the attention curve before you read the winners.
Stage 1 of 12 · Attention curve
The first job is not to celebrate the leader board. It is to understand what kind of market the leader board came from. If attention is brutally concentrated, every later conclusion needs to be read through that constraint.
Start with the shape of the field01 / 12
What it measuresHow steeply attention collapses into a tiny winner set.
Why it mattersA concentrated market distorts discovery, money, and the stories that survive.
Research protocol
Inspect how the music lane was assembled before you repeat what it says.
The charts stay public-facing, but the method should still be inspectable. Open the audit to see cohort boundaries, source additions, cleaning decisions, and the guardrails that keep the model from flattering itself.
Loading audit surface…
—Core cohortThe leaderboard cohort used for the main virality charts.
—Enriched contextBroader structural context used when the question needs more than the top 100.
—Rigor checksCross-validation, bootstraps, and leakage blocks that keep the outputs honest.
Data auditCompiled from the current repository state
Loading audit detail…
01
The Power Law of Attention
View counts are not normally distributed — they follow a power law where a tiny elite captures most of the attention. The complementary cumulative distribution (CCDF) on log-log axes reveals the heavy tail: a straight line means pure power-law behaviour. We care because discovery systems, ad revenue, and cultural memory all behave differently when attention is this concentrated.
—
Power-law exponent (α)
95% CI: —
—
Kolmogorov-Smirnov fit
Lower = closer fit
—
Videos in tail
—
Tail threshold (xmin)
02
The Inequality of Attention
A Lorenz curve shows how unevenly views are distributed across channels. The diagonal is perfect equality; the deeper the bow, the more a few mega-channels dominate. Gini (0 = equal, 1 = one takes all) and the Theil index quantify the gap, which matters because a supposedly “open” music field can still be governed by a very small set of accounts.
—
Gini coefficient
—
Theil entropy index
—
Top 3 channel share
—
Top 10 channel share
03
What Actually Moves with Reach
What actually correlates with more views? Spearman rank correlations show raw associations; partial correlations strip out the effect of channel size so you can see what matters independently of how big the channel already is. That helps separate reach from structure, which is the difference between “popular because it is everywhere” and “popular because something about it travels.”
04
Viral Archetypes
K-means clustering on views, virality (views-per-subscriber), channel size and duration groups the 100 songs into five distinct strategies for capturing attention. It is useful because it turns a noisy leaderboard into a few recognizable release patterns you can compare.
05
How Tags Cluster
When two tags appear on the same video they share an edge. The resulting co-occurrence graph reveals genre neighbourhoods, cross-over bridges, and the tags that sit at the centre of the music ecosystem. That makes it easier to see whether discovery is actually broad or just moving inside a small loop of familiar labels.
—
Tags (nodes)
—
Co-occurrences (edges)
—
Graph density
—
Communities
06
How Predictable Is Virality?
An ensemble of RandomForest and GradientBoosting regressors, cross-validated on real data, tries to predict (log) view counts from pre-publication features alone. The honest R² shows how much is skill versus timing and luck, and the feature importances show which levers still mattered after the rest of the context was held constant.
—
Cross-validated ensemble R²
07
Repeat Pressure Map
This panel treats repeating attention as a public proxy for oscillation. It does not inspect audio waveforms or private hardware traces. It looks at the visible shape of the catalog - duration fit, attention concentration, release timing, and tag/title echo - and turns that into a live resonance map you can click through.
—Oscillation
—Friction
—Stability
—Gate window
Loading resonance map…
Interception Trace
Click a track to focus the scene. The highest-resonance tracks are the ones whose attention pattern stays narrow, repeated, and durable.
08
How Songs Travel Through Culture
We do not pretend public rows can tell us heart rate, respiration, or hidden waveform physics. What they can show is whether the field keeps rewarding the same bodily invitation, the same corridor of circulation, and the same inherited distribution path. This section keeps those proxies explicit so the analysis can go deeper without becoming vague.
—
Preparing embodied summary
Loading embodied boundary…
Vitality signatures
Preparing signatures…
Selected signature
Cultural corridors
Preparing corridors…
Research provocations
Loading provocations…
09
Release Scenario Lab
Set the parameters of a hypothetical music upload and the trained model predicts where it would rank among the real top 100. All predictions use the same RandomForest model trained on actual YouTube data, so you can see how a change in one variable shifts the whole ranking story.
—
Percentile: —
Predicted views: —
Adjust the sliders to simulate a hypothetical upload.
10
Bias in the Top 100
Who accumulates the front row, and by what pattern? We measure gender representation, genre gatekeeping, collaboration patterns, language dominance, engagement disparities, and attention concentration across 636 real songs from 3 public datasets. This section is less interested in blame than structure: where exposure gathers, which genre families keep inheriting the advantage, and which variables still tilt the field after the surface story changes.
—
Overall Equity Grade
Compositing gender parity, views parity, genre diversity, and concentration
L
Role Lens
The facts do not change when the reader changes, but the consequence does. A creator reads this for room to maneuver. A buyer reads it for repeatable demand. A listener reads it for breadth. A researcher reads it for falsifiable structure. Change the lens; keep the evidence.
Artist lens
How does a creator use this?
Loading role lens…
Actionable opening—Calibrating from the live field.
Evidence confidence—Coverage and classification depth will appear here.
Loading
Preparing role-specific factors—
The underlying drivers for this role lens will appear here.
Waiting for coverage, classification, and stability notes from the live field.
Preparing the question this lens should force before anyone repeats the conclusion.
This is where note, chord, and key evidence will be joined back to the public reach data, but only when the repository actually contains notation-bearing files.
Current stateWaking upThe score-linked layer is waiting for the public music field to finish loading.
G
Genre Breakdown
Genre
Songs
View Share
Avg Views
Avg Dur.
Collab Share
Loading…
—
Known-gender parity
—
Views parity (F/M ratio)
—
Known-genre diversity
—
Top 5 concentration
—
Coverage confidence
—
Year coverage
Data Sources
11
3D Virality Landscape
The same archetype clusters rendered in three dimensions — views, virality, and channel size form the axes. Drag to rotate, scroll to zoom. Each sphere is a real song; colour encodes the cluster. This helps distinguish a true strategy from a one-off spike.
12
Real Songs Explorer
This is the public row layer behind the music lane. It now carries the broader structural context as well as the core leaderboard, so you can search, sort, filter, and page through the evidence before deciding whether a neat summary still feels earned.
S
Score layer
When a score, MIDI, ABC file, tab, or chord chart exists in the repo, this layer folds it into the song record and lets the note and harmony patterns speak for themselves. When the file is missing, it leaves the gap visible and invites the next addition instead of pretending.
—
Catalog songs
—
Local music files
—
Score-linked songs
—
Notation link rate
—
Parsed note events
—
Parsed chord events
If you want the note lane to deepen next, start here.
