Mission
Make candidate evidence inspectable.
The Observatory organizes anonymous interaction observations into transparent audience-pattern indicators for research engagement, momentum, breadth, and evidentiary support.
CSI Research Observatory
A delayed, privacy-preserving prototype showing how aggregate audience interaction observations can be transformed into indicators and candidate evidence for the CSI research program.
The public Observatory reports aggregate audience-pattern indicators—not visitor analytics or validated CSI collective states. Individual sessions, visitor identifiers, raw IP addresses, exact traffic counts, and demographic claims are not displayed.
Public instrumentation path
Anonymous observations→Aggregate indicators→Candidate evidence
A qualifying CSI inference would additionally require a theoretically defensible collective boundary, collective-level composition logic, and validation.
Scope of this prototype
Research instrumentation, not a CSI collective-state result
Co-observed website visitors are not assumed to constitute a bounded collective merely because they interacted with the same research site. The Observatory therefore demonstrates observation, evidence construction, temporal comparison, evidentiary support, and privacy-preserving publication—not a qualifying or validated CSI collective-state inference.
Current reading
Audience-pattern indicators
Heuristic signals derived from the current evidence window; Momentum may use shorter matched periods while evidence accumulates.
Shared metadata Snapshot methodology Evidence window, calculation timestamp, schema, and publication delay. View methodology
- Method
- —
- Evidence window
- —
- Calculated
- —
- Evidence completeness
- —
- Publication timing
- Published after a 24-hour delay
The calculation and thresholds vary by indicator; the metadata above applies to every indicator in this snapshot.
Research engagement indicator — Depth and continuity of aggregate interaction with the research program. View derivation
Definition: A behavioral indicator of substantive interaction with the research program across depth, duration, repeat visitation, and research-oriented behavior. It is not a validated collective-engagement construct.
Engagement indicator = 0.30(Reading depth) + 0.25(Duration) + 0.25(Returning-session rate) + 0.20(Research-oriented share)
How the inputs are scaled: Each input is normalized to 0–100 before weighting. Average reading depth maps one page to 0 and six or more pages to 100. Average observed duration maps zero minutes to 0 and 15 or more minutes to 100. Returning-session rate and research-oriented share use their aggregate percentages directly.
How to interpret it: The weights are shares of the final indicator, not observed percentages. Reading depth can contribute 30 points; duration and returning behavior can contribute 25 points each; research orientation can contribute 20 points. The weighted result is bounded to 0–100.
Illustrative example: Normalized inputs of 60 for depth, 40 for duration, 20 for returning sessions, and 50 for research orientation produce 0.30(60) + 0.25(40) + 0.25(20) + 0.20(50) = 43, classified as Moderate.
Current derivation: —
Label bands: Very high ≥72; High 54–<72; Moderate 34–<54; otherwise Low.
Momentum indicator — Direction of change in aggregate activity relative to the preceding evidence period. View derivation
Definition: Direction of movement in aggregate activity and reading depth across the two most recent adjacent, equal-duration evidence periods.
Momentum indicator = clamp 0–100 [50 + 30(Session change) + 20(Reading-depth change)]
How the inputs are scaled: The model uses the longest two adjacent matched periods currently supported, beginning at 12 hours per period and expanding automatically to seven days. Session and reading-depth changes are proportional differences between those periods; 40% growth is entered as 0.40 and a 20% decline as −0.20.
How to interpret it: The score begins at a neutral baseline of 50 when both measures are unchanged. Session change has 60% of the change-sensitive influence and reading-depth change has 40%. The final score is bounded to 0–100. Shorter comparisons are labeled Provisional with Low or Developing support; two fully retained seven-day periods are Established. “Strong growth” describes a strong two-window increase, not mathematical acceleration.
Illustrative example: If sessions rise 40% and average reading depth rises 50%, the result is 50 + 30(0.40) + 20(0.50) = 72, classified as Strong growth.
Current derivation: —
Label bands: Strong growth ≥72; Growing 56–<72; Stable 42–<56; otherwise Cooling.
Audience breadth indicator — Breadth of aggregate discovery paths and audience-orientation patterns. View derivation
Definition: Distributional breadth across heuristic audience orientations and aggregate discovery sources.
Breadth indicator = 0.55(Audience-orientation entropy) + 0.45(Discovery-source entropy)
How the inputs are scaled: Both inputs use normalized Shannon entropy on a 0–100 scale. A value near 0 means evidence is concentrated in one category; a value near 100 means it is distributed evenly across the available categories.
How to interpret it: Audience-orientation breadth contributes 55% of the score and discovery-source breadth contributes 45%. Entropy measures distributional evenness—not demographic diversity, visitor identity, or the number of visitors by itself.
Illustrative example: Audience-orientation entropy of 80 and source entropy of 70 produce 0.55(80) + 0.45(70) = 75.5, rounded to 76 and classified as Broad.
Current derivation: —
Label bands: Broad ≥72; Developing 48–<72; otherwise Concentrated.
Interpretive support — Strength of the available aggregate evidence for the heuristic audience-pattern interpretation. View derivation
Definition: Evidentiary support for the dominant heuristic audience-orientation interpretation. It is not construct-validation confidence for a CSI collective state.
Interpretive support = 0.75(Sample support) + 0.25(Leading-orientation separation)
How the inputs are scaled: Sample support rises linearly from 0 to 100 and reaches 100 at 30 sessions in the current seven-day window. Leading-orientation separation reaches 100 when the top heuristic audience-orientation share leads the second-ranked share by 25 percentage points; larger gaps remain capped at 100.
How to interpret it: Sample support supplies 75% of the score and separation supplies 25%. The five-session minimum that permits public disclosure is only a privacy gate; it is not the 30-session target used for full sample support and does not imply high support.
Illustrative example: Sample support of 60 and leading-orientation separation of 40 produce 0.75(60) + 0.25(40) = 55, classified as Moderate.
Current derivation: —
Label bands: High ≥72; Moderate 45–<72; otherwise Low.
Current audience-pattern interpretation
—
Temporal context
Historical interpretation timeline
Published audience-pattern interpretations across recent evidence windows.
Public safeguards
What remains outside this presentation
No individual analyticsSessions and visitor-level journeys are not shown.
No exact traffic countsPublic interpretation remains qualitative.
No demographic inferenceIdentity and protected-trait claims are excluded.
No CSI collective-state claimSite visitors are not assumed to constitute a qualified bounded collective.
Private operations
Access CSI Observatory Operations
Access to the private Observatory is by invitation only.
Continue exploring