After activation
The assessment uses windows wholly after activation and separate from discovery. An archive received today keeps its actual measurement time.
THE METHOD / GIVING INTUITION A SHAPE
To find a trace,
define what to look for.
Trinity turns an idea into a traceable investigation: measure the texture of signals, declare the pattern being sought, then watch for its return.
Follow the three steps- 01Characterise 1/fβ
A slope, its frequency band and its limits.
- 02Freeze the pattern sought
Criteria retained before the observation.
- 03Test what comes next
New measurements, a traceable assessment.
01 / CHARACTERISE THE SIGNAL
The slope describes a texture.
β describes how signal power is distributed across the frequencies studied. Trinity publishes the fit, its physical frequency band and its stability across sub-bands: three ways to examine possible 1/fβ behaviour.
Loading spectral diagnostics…
| Source / state | β / r² | Periods studied | Coverage | Synthetic 95% reference |
|---|
Read the range with its model.
The reference range contains the 2.5% and 97.5% quantiles of slopes computed from finite-length Gaussian simulations with a 1/fβ spectrum, at the stated reference β. The report states the low-frequency cutoff; this finite model does not assume stationary 1/f behaviour at every scale. It is a conditional dispersion of the estimator, not a confidence interval for the real β. The model and the regime still need examination; r² alone does not validate them.
COMPARE / ONE SLOPE, SEVERAL EXPLANATIONS
Which model follows what comes next?
White noise, short-memory AR(1) and a finite Gaussian 1/fβ model examine the same window. Each fits its first 75%, then is evaluated on the following 25%, at their real positions without filling gaps.
Loading the model comparison…
Input snapshot and calculation time: reading the report…
| Model | Relative score | Fitted parameter | Calculation state |
|---|
Parameters, assumptions and fingerprints
The same mean and linear trend are fitted for all three models. Parameters are fitted on the prefix only; their uncertainty is not integrated into the score. AR(1): ρ from −0.99 to 0.99; finite 1/f: β from −2 to 4, extension eight times the window and a fixed low-frequency plateau.
Zero is the best score among these three models.
A negative score is the difference from the highest conditional score per point for this source. It is neither a probability, model acceptance nor a glitch detection. All three models may be inadequate; do not compare scores across sources.
A retrospective report computed from a coherent read-only SQLite snapshot, refreshed hourly. Snapshot time is separate from calculation time; reports older than 90 minutes are marked stale. Measurement freshness remains visible for each source. No detector or prospective criterion is changed.
METADATA / KEEPING THE CONTEXT
Where do measurements come from?
A separate archive retains instruments, the active state and codes received with NOAA products. Active identifies NOAA's selected spacecraft, without judging physical measurement quality. The codes below remain raw: their values are not translated into good or bad.
Loading NOAA metadata…
This collection does not change detector inputs. The archive start is Trinity's first receipt, even for an older measurement. Reports not updated for more than 45 minutes are marked stale.
02 / DECLARE WHAT TO LOOK FOR
The pattern comes before the result.
A spectral change or an overlap between sources becomes a tracked experiment once its criteria, sources and evaluation horizon are sealed. The registry retains the protocol fingerprint before examining the future period.
Loading the pattern registry…
03 / WAIT FOR A NEW OBSERVATION
A recurrence is measured
in what comes next.
With known exposure
Periods when a source could not detect remain visible in the coverage. Missing measurements do not become an absent pattern.
By examining connections
Different domains do not guarantee independent instruments. Shared causes, sensor quality and context are part of the investigation.
AT WHAT SCALE ARE WE LOOKING?
Every measurement has a clock.
The current spectral sources are sampled by the minute or the hour. Analysis windows span 6 hours or 7 days; peaks are placed on an hourly grid. The published frequency band also depends on the Welch segments.
A microsecond is not resolved by these measurements. A persistent consequence might leave a trace accessible to the investigation, without directly dating such a brief event.
- Sensor cadence
- Minute / hour
- Spectral window
- 6 h / 7 days
- Detection step
- 1 hour
An intuition, criteria, then the test of time.
Explore the investigations Contribute a check Déjà vu: follow the experiment