Changelog

What was wrong. Many contributing sources — most state-forestry price lists among them — publish a price without ever stating when the wood sold. The public views filled that gap with coalesce(date_sale, current_date), and then republished the result as latest_sale. The effect was that a price with no known sale date was published as having sold today, and it re-dated itself to today on every refresh. On the day this was found, 24 of the index’s 30-day cells were stamped with the current date and the home page showed sixty cells all claiming to be “today”.

What the underlying intent was. The coalesce was deliberate and half right: it kept undated rows inside the rolling window instead of discarding them, which is correct — an undated price list is still evidence. The defect was reusing that same expression as the date the index then published. Window membership and reporting are two different questions and were sharing one expression.

What changed. They are now separate. latest_sale is null exactly when no source states a sale date, and a new latest_observation carries the day the price was recorded. Freshness everywhere is measured from the observation. The public-index contract now rejects any published view that lacks latest_observation or that reintroduces the old expression, so this specific mistake cannot come back silently.

What it did to the numbers. No price changed. Dates did: the 30-day window went from 24 published cells to 14, because ten of them only qualified on a date that was never real. Cells whose sources state no sale date now show an honest age instead of “today” — in some cases over a year. Nothing on the site got fresher; the site stopped claiming it was.

Migration: 20260930110000_tpi_honest_observation_dates

The index went public with 11 sources scraping successfully. The minimum-evidence gate at launch was five normalized observations per cell, from at least one public source; the original two-source rule was relaxed to one before launch and that relaxation is documented in the methodology. Everything before this date is pre-publication and is not logged here.