07 · Proof
How we test whether our method sees topics early
A retrospective blind test: at six past dates, our method recommended topics using only sources published before that date; twelve months later, public measures decided which ones materialized.
of predictions materialized
in the control group
median head start over the mainstream
Retrospective blind test 2024–2025 · six validated industries pooled · strict measurement
This quarter only. All six quarters for Energy & Utilities: 37 % vs 20 % in the control group.
Study design in four steps
Study design in four steps
Travel back
Six cutoffs between January 2024 and April 2025. The method sees only what was public at that date – no later news, no prices, no announcements.
Recommend
From those sources, the method recommends topics with a reasoning and a list of sources. The control group is drawn from the same sources but without the method's selection – this keeps any model hindsight symmetric across both groups.
Score with fixed criteria
Twelve months later, a second routine measures two of three public indicators (Google Trends, GDELT news volume, Hacker News). Materialization must show in at least two of three channels.
Let reality decide
Hit rate and lead are published without after-editing. Topics without a measurable signal count as “not measurable” and are excluded from the rate.
The blind test, hands-on
Six industries, six cutoffs. Switch the industry and the quarter to see the method work on a concrete example.
Four more industries are in exploratory testing.
Choose a topic to see its card.
Control topics come from the same model and the same sources, so any hindsight the model might carry helps both sides equally – the gap comes from the selection method.
Results by industry
| Industry | plinio® TL | Control | Lead | Median head start | |
|---|---|---|---|---|---|
| Technology & Software | 45 % | 23 % | +22 pts | 10 months | View industry → |
| Insurance & Risk | 51 % | 32 % | +19 pts | 8 months | View industry → |
| Manufacturing & Industry | 45 % | 26 % | +19 pts | 9 months | View industry → |
| Banking & Fintech | 44 % | 26 % | +18 pts | 9+ months | View industry → |
| Energy & Utilities | 37 % | 20 % | +17 pts | 9 months | View industry → |
| Healthcare & Life Sciences | 46 % | 39 % | +7 pts | 8 months | View industry → |
| Six industries pooled | 44 % | 28 % | +16 pts | ~9 months |
Validated in six industries. Four more are in exploratory testing because our historical sources underrepresent them.
Why 44 % is strong
A hit rate is only informative if you know the base rate. The control group drawn from the same sources and the same model materializes only 28 % of the time – more than random plausibility, but clearly less than 44 %. The gap is the selection method.
Base rate, not a coin flip
A single hit rate without a comparison is worthless. The control group from the same sources and the same model is what makes the 44 % readable: sixteen points above what the model produces without the selection method. That is what the method adds to the model – not the model alone.
Asymmetric economics
A topic that materializes early opens markets, analyst conversations and category leadership. Topics that do not materialize are still no loss: they remain sourced expert pieces with communication value, published in a field with little competition. The leverage is on the side of materialization – which is why a rate well above the base rate is worth more over time than a high rate bought with arbitrary recommendations.
More than an analyst can cover
A human analyst would have neither the time nor the means to evaluate this many sources, follow them daily and interpret them. plinio TL does exactly that, every day and for each of your markets – and only puts in front of you the topics that pass the checks.
Measured strictly on purpose – the real rate is likely higher
- Strict success definition: a topic only counts as materialized if it rises in two of three channels at the same time – Google Trends, GDELT, Hacker News. A topic that breaks through in only one channel counts as a miss. That makes 44 % a lower bound.
- Measurement uses broad public channels. Many B2B topics take hold first in trade media, analyst reports and conference programmes – where your customers read. The test does not capture those wins.
- In the test, the method works from a typical persona per industry. In use, plinio TL knows your company, your competitors and your sources, and selects more precisely.
- Fair control: comparison topics come from the same model and the same sources. Any hindsight the model carries helps both sides equally; the gap comes from the selection.
- Newer cutoffs score better, in both groups. What counts is therefore the gap to the control group, not the absolute rate.
- Six industries are validated; four more are in exploratory testing.
Our public predictions are timestamped in the Bitcoin blockchain. They can be neither changed nor backdated later – and anyone can verify that.