Methodology

News tone (Tlatolli)

Tlatolli continuously reads statements from central banks and statistical offices, financial media and regional media. An AI assesses each story: how important it is (materiality), which currencies or markets it affects (relevance) and its tone for each, on a scale from −1 to 1, with a declared confidence from 0 to 1.

The 24-hour tone of a currency or market is a weighted average of that day’s stories that matter: materiality and relevance of at least 0.5. More important, more relevant, more confident and more recent stories weigh more. That is why the figure may not match a simple count of positive and negative stories.

In words, tone is positive above +0.05, negative below −0.05 and neutral in between. Confidence is shown as high (0.8 or more), medium (0.6 to 0.8) or low (below 0.6). It is the confidence declared by the AI and is not yet calibrated.

If no stories in the last 24 hours support the tone, the page shows no figure.

Relative tone of pairs (pair-tone:v1)

The tone of a BASE/QUOTE pair is half the difference between the tone of the base and that of the quote: (base tone − quote tone) / 2, so it stays on the same −1 to 1 scale. Both tones are taken to 4 decimals and the result is rounded to 4 decimals.

The pair’s confidence is the lower of the two. If either currency has no stories in 24 hours, the pair shows no figure. If the formula changes, it will be a new version (pair-tone:v2) with its own calibration from scratch.

Readings by factor (Machiyotl)

Each reading is an official series (rates, bond curves, macro, CFTC positioning, valuation, inventories or financial conditions) compared with its own history, usually 5 years (3 for the CFTC and 15 for the real exchange rate).

The percentile says what share of that history lies below the current value. The bands are: very low (10 or less), low (10 to 30), usual (30 to 70), high (70 to 90) and very high (90 or more). The change is over the stated window (for example, 1 week) and is flagged as unusual when it is twice the series’ typical change.

Each reading shows its source, the date of the data and whether it is stale, plus an evidence note: none of these families has shown the ability to anticipate prices in our pre-registered studies.

Sources and licenses

Readings use only public sources that allow reuse with attribution: ECB, BIS, the Fed (including the Chicago and St. Louis Feds), the US Treasury, the Bank of England, Japan’s Ministry of Finance, the CFTC, the EIA, BLS, BEA, Census, Eurostat, ONS, StatCan, e-Stat, FSO and the OECD.

For news we show only the headline, the link to the original outlet and our own summary, never the outlet’s text. For subscription outlets we show only the headline and the link.

We do not publish broker prices, spot prices from commercial providers or data from commercial providers’ free tiers.

Updates

News tone is updated every few minutes on weekdays and less often on weekends. Official readings are updated once a day, after the New York close. Each page shows when it was updated.

What Pohuant does not do

It does not forecast the direction of any price or combine the factors into a directional score. That score existed in Machiyotl and the pre-registered studies did not support it, so it is not published.

We measure how reliable the news tone is and will publish the results every month, including the bad ones.