Our 90-Day Premium Data Audit: Coverage, Gaps, and Recorded Ranges

This report uses the data returned by this site's own /api/premium/history?days=90 endpoint on September 27, 2026. It is an audit of what the site actually retained, not a reconstruction of what the markets should have shown.

What was available

The requested window ran from June 28 to September 26. The cache contained 74 dated rows for both gold and silver, beginning June 29 and ending September 24. A 90-day request therefore did not produce 90 complete daily observations.

Series	Observed rows	Average	Minimum	Maximum
Gold SGE	0	—	—	—
Gold MCX	72	14.34%	12.42% (Jul 28)	16.54% (Jul 2)
Gold LBMA-related	72	0.10%	−1.78% (Aug 12)	2.42% (Sep 1)
Silver SGE	72	10.82%	8.42% (Aug 24)	12.67% (Sep 24)
Silver MCX	72	20.86%	13.63% (Aug 30)	28.30% (Jul 2)
Silver LBMA-related	72	0.09%	−2.95% (Aug 12)	5.84% (Sep 2)

The most important finding is the missing gold SGE series

No observed gold SGE row was present in this retained window. It would be misleading to draw a gold SGE trend line, calculate an average, or replace the missing series with zeros. The dashboard may still display a current estimate or another status depending on availability, but that is not the same as a retained observed history.

Silver SGE did have 72 observations. This difference is a useful reminder that coverage must be checked by commodity and market rather than assumed from the existence of a market card.

What the recorded ranges do and do not say

Within this retained sample, MCX-related differences were materially larger for silver than for gold. The silver MCX average was 20.86%, compared with 14.34% for gold. The LBMA-related averages were close to zero for both metals, but individual silver observations ranged from −2.95% to 5.84%.

These are properties of this dataset, not explanations of market behavior. The endpoint combines values recorded on available dates, and the market fields can differ in contract selection, timestamp, currency conversion, and vendor methodology. Repeated values across adjacent dates can also reflect a source that did not refresh at the same time as the benchmark.

How the figures were calculated

For each market, we counted rows containing an observed market value and calculated the arithmetic mean, minimum, and maximum of the stored premium percentage. Rows with no market observation were excluded rather than treated as zero. Derived benchmark-and-FX estimates are not written into the observed history.

The audit can be reproduced from the public JSON endpoint using commodity=gold or commodity=silver, market=all, and days=90. Because Redis retention and scheduled snapshots change over time, a later request may return a different window or sample count.

What we will change because of this audit

Future reports will state the actual first and last dates and the number of observations before discussing an average. A selected period will never be described as complete unless the stored dates support that statement. Missing markets will remain missing, and no causal explanation will be assigned from a premium percentage alone.