Methodology · precision and limitations
Know what a signal measures before you trust it.
A search, a view, a hashtag rank and a purchase measure different behaviour. TrendFalcon keeps those differences visible and does not treat any signal as proof of demand.
What you are looking at
Three layers of information.
- Collected observations: values returned by a source, with their units, dates, geography and precision.
- Proposed analysis: methods we intend to test. They are not yet validated production scoring.
- The preview: Wikipedia reading observations and estimates with their methods, editorial product ideas, and sample data that is invented and marked “Sample data”. Sample figures are not calculated from the reading data.
Proposed scores
Attention, readiness, evidence: three questions, kept apart.
Attention asks how strongly interest is growing against its own recent baseline, and whether the growth continues across several observations. A sustained rise should weigh more than a single noisy spike. Sparse histories, very small baselines and incompatible measurements can make a score unavailable. Weights and thresholds still need calibration; a score of 80 is not an 80% probability of anything.
What do we need before ordering? This applies to a specific offer in a specific market: buying evidence, realistic unit economics, differentiation, fulfilment and timing. Views and searches cannot establish conversion, profit, supplier reliability or delivery time. Missing evidence stays an open question.
How much evidence do we have? This asks how much support an interpretation has: source coverage, freshness, topic and market fit, comparable history and precision. Several platforms repeating one campaign are not several independent proofs, and several editions of one source, such as Wikipedia in two languages, are one source family, not independent sources. It describes the support available, not the probability that a trend is real or profitable.
The three are never combined into one number: a single score would hide which of the three questions is weak.
Counts, ranks, indices and bands
Each number keeps its own unit.
Counts keep the unit and reporting window the provider supplies. Views are not unique people. A rank is a position in one list, not a measure of total popularity. Rounded counts and lower bounds such as “500+” cannot support exact percentage growth.
The Google Trending Now feed gives approximate traffic bands; they are distinct from Google Trends' relative 0–100 interest index, and equal index values in two markets do not mean equal volume. Google's explanation of Trends data.
A relative 0–100 index, from any source, shows interest within the returned series, not a number of searches, and separately normalised keywords cannot be compared as absolute demand.
A collection timestamp records when we obtained an observation, not necessarily when the activity happened. Weekly histories are not interpolated into invented daily evidence. Comparisons need the same topic, metric, market and window, and language, search market and audience residence are different facts.
Missing observations stay missing. A topic leaving a limited feed does not mean interest became zero. Percentage movement needs comparable, suitably precise values and a non-zero baseline; otherwise it is unavailable. Wikipedia reading changes in the demo are calculated from counts and labelled with their method; percentages on sample data are invented.
Limits
What these signals cannot establish.
- Attention does not establish sentiment, purchase intent, audience fit or future sales.
- News, advertising, seasonality, bots and coverage changes can produce spikes.
- A limited preview or top-keyword list is not a census of a whole platform.
- Similar names and hashtags can describe different topics; cross-platform matches need review.
- No platform is assumed to show a trend first. Timing depends on the topic, the market and the collection window.
- The proposed scores have not been validated against sales. Numerical detail does not create statistical certainty.
Our own test of whether a rise holds is described, with its plan and its limits, on the research page. For source status see Sources. If something looks wrong, tell us which topic, market and date you were viewing.