The Gold Barometer

Corrections and revisions

If a number we already published changes, we say so here, dated, with the reason and the new value. A published page is never edited after the fact. The change is logged here instead. Each entry is dated, with what changed and what did not.

The rule

  1. Every number we publish stays public. If a score or a part of the score changes after release, this page shows both the old value and the new one. Each entry is dated.
  2. Changing the recipe means a new version number. The methodology page has a changelog: version number, date, and what changed. Changing how much a part counts, or which way it pushes the score, triggers a 90-day parallel run. Both the old and the new reading are published side by side. Only then does the new one become the standard.
  3. Data revisions are logged, not hidden. Sometimes the original data source revises a past number (for example the US government revises past inflation figures, or the IMF revises reserves data). If that changes a past part of our score, we note it here. We give the date we received the revision and the effect on the score.
  4. The daily archive is never overwritten. The public GitHub archive at github.com/thegoldbarometer/data keeps every daily record exactly as it was released. Any correction is added as a new dated entry. The old record is never rewritten.

The log

2026-09-08: two readings published three and two days late

The readings for September 6 and September 7 were computed on time, each night after the market close. Neither was published. On September 6 a monthly maintenance job stopped halfway, and a safety check then refused to publish anything until the job was completed. The check did its job. The site, the archive and the social accounts stayed on the September 5 reading until this entry.

What changed: both days are now in the archive, dated the day they describe, with a note saying when they were recorded. September 6 reads 31, Unfavorable, on 6 of 7 parts. September 7 reads 31, Unfavorable, on 6 of 7 parts. The September 6 record was replayed on September 8 with the same engine. Its score and zone match the original night run exactly. One part, the entry price, differs by a fraction of a point because one input was revised after that night. What did not change: the September 5 reading, and every reading before it.

2026-08-13: six early readings restated after the engine correction

The correction of August 11, described in the entry below, fixed three engine faults. The six readings published before the fix still showed what the faulty engine had said. The chart showed a dip to 16 on the 0 to 100 scale. It also showed two zone changes that were products of those faults, not facts about gold.

Each of the six days was replayed with the corrected engine, on the data available that day. Old and new: August 5, from 21 to 33. August 6, from 21 to 33. August 7, from 19 to 33. August 8, from 16 to 33. August 9, from 16 to 33. August 10, from 19 to 33. The zone changes dated August 7 and August 11 are removed from the record of events. Under the corrected engine, the reading stayed in the Unfavorable zone through that whole stretch.

The original readings stay in the public archive, unchanged, next to the restated values. Anyone can audit the difference there. Charts and event dates on this site now use the restated series.

2026-08-12: a launch-day revision, found late and logged late

On August 5, the first published reading went out at 22 around midday. That evening a second run changed the day's number to 21, once the entry price part had the day's closing data. The archive kept both versions, but this page did not record the change. That was wrong, and it is recorded now.

The rule that prevents it going forward: one reading per day, recorded once, after the market close. An outside reviewer caught the missed change, and this entry is the result. That is the system doing its job.

2026-08-11: three implementation faults in the score engine, corrected together

First, the real-rate part. When its source switched from the pre-2003 stand-in to market TIPS, the ranking window shrank to about three years. Readings near 2006 were ranked against almost no history, and the overall score crashed with no economic cause. The part now ranks every value inside one joined history, stand-in first, TIPS after.

Second, two smaller faults. The reconstructed monthly series ranked some yearly changes against data that only existed later. And the trend part's 12-month average included the month it was measuring. Both are fixed, in the daily engine and in the 1971 rerun alike.

What changed: 84 of the 667 rerun months moved to a different zone, most of them near 2006. The bottom zone shrank from 33 months to 27. What buying in those months returned five years later, after inflation, moved from +22.1% to -7.7%. Today's reading moved from 19 to 29, out of the bottom zone. The headline claim survived and strengthened. The Favorable zone beat the Unfavorable zone at five years in 100% of 10,000 reruns of the full record.

What did not change: the daily readings already published to the archive. Those are never rewritten. Every affected page was updated the same day, and this entry is the account.

2026-08-10: the dollar part used the wrong freshness limit

The dollar part reads a weekly release, the Fed's broad dollar index. Our freshness rule treated it as a daily source with a 7-day limit. A weekly series is routinely 8 to 13 days old just before its next release. So on 2026-08-08 and 2026-08-09 the dollar part was dropped. The score was computed on 5 of 7 parts, even though the dollar data was current by a weekly rule.

What changed: from 2026-08-10 the dollar part carries a 14-day limit, matching its weekly release rhythm. What did not change: the published readings. Both days keep their released score of 16 and their "5 of 7 parts" note. The move into the bottom zone happened on 2026-08-07, with 6 of 7 parts in use, so it was not caused by this rule.

Related