RAPI's sizes rise between 0.33% and 3.00% in a month

The window runs 2 September to 2 October on Rapaport's own daily series. Gahcho Kue realised 44.4% less a carat in its latest published quarter, a rough price the polished index does not carry.

By Carat Capital

3 October 2026

What changed

All five sizes rose. Rapaport's published daily series gained on every size across the 30 days to 2 October 2026. The 1-carat added 0.33% to $3,912.48, the least of the five, and the 0.30-carat 3.00%.

What it means

If you quote RAPI: Carry the window as well as the level. The page's percentage is a trailing-window figure.
If you price rough: The polished benchmark is not your counter. One rose over 30 days as the other realised 44.4% less.
What this is not: Evidence the decline narrative is wrong. The two instruments measure different goods over different periods.

Key figures

0.30-carat over the windowto $959.58, the largest of the five	+3.00%
1-carat over the windowto $3,912.48, the smallest	+0.33%
Sizes higher over the windowno size fell	5 of 5
Gahcho Kue rough a carat, Q2 2026year on year, CAD 50 against CAD 90	−44.4%

Source: Rapaport Diamond Price Index, embedded daily series, window 2 September to 2 October 2026, 31 points per size · page read direct, 305,223 bytes

The cheap end led again.

Thirty-pointer led. The 0.30-carat added 3.00% to $959.58 and the 1-carat 0.33% to $3,912.48. That is 9.1 times the move on the division 3.00 over 0.33.

All five rose. The 0.50-carat added 2.26% to $1,315.39, the 2-carat 0.82% to $10,659.16 and the 3-carat 1.96% to $18,208.08. No size fell across the window.

RAPI by size, 2 September to 2 October 2026Rapaport's embedded daily series, 31 points per size · USD a carat · change column computed

Size	2 September	2 October	Change, computed	Page's own %
0.30-carat	931.63	959.58	+3.00%	3
0.50-carat	1,286.32	1,315.39	+2.26%	2.26
1.00-carat	3,899.61	3,912.48	+0.33%	0.33
2.00-carat	10,572.46	10,659.16	+0.82%	0.82
3.00-carat	17,858.06	18,208.08	+1.96%	1.96

Notes on this table

The level columns are the first and last points of each size's embedded 31-point daily series, read out of the page's own JSON. The fourth column is Carat Capital's division, last over first. The fifth is the unlabelled percentage Rapaport prints beside each size. All 5 of 5 sizes rose across the window. The two columns agree on 5 sizes out of 5, which is what establishes the page's field as a trailing-window change rather than a daily or year-to-date move. Levels on this page are restated after publication, shown in Carat Capital's report of 22 September 2026, so a quote needs the date the series was read as well as the date of the point: these were read on 3 October 2026. The series lags one session and carries no colour, clarity or cut specification beside it.

The page's percentages are established.

Field equals window. Rapaport prints an unlabelled percentage beside each size whose window had not been pinned down. Recomputing first point to last reproduces it on five sizes out of five, to the hundredth.

Date the window. A figure read from that field is a 30-day trailing change and should be dated to the window, not to the day it was read. The series also lags one session.

Polished is not rough.

Two different goods. RAPI measures polished asking prices. Gahcho Kue sells rough, and realised CAD 50 a carat in the three months to 30 June 2026. That is against CAD 90 a year earlier, or 44.4% below.

Periods do not match. A 30-day polished window cannot be set against a year-on-year rough quarter and called a contradiction. Rapaport publishes only 30 trailing days here, so no 12-month polished comparison is available.

What the decline claim rests on.

The reason is stated. Mountain Province's chief executive attributed his company's restructuring to a decline in diamond prices over the past year. The attribution is in its release of 1 October 2026.

Not tested here. That claim covers twelve months of rough. Nothing in the 30 days of polished points above tests it, and the two should not be quoted against each other.

What to watch

Monday 5 October 2026The next daily point, and whether the 1-carat holds above $3,912.48. The series publishes on weekdays and lags one session.
Q3 2026, date not publishedMountain Province's next realised rough price. In August the company said its most recent sale showed an improvement it was too early to call sustained.
Each read of the pageWhether published points move again. All five points read on 21 September 2026 were restated by 12 to 27 cents when read a day later.

The story so far

30 September 2026The 1-carat added 0.43% in a month, the 0.30-carat 3.3%
22 September 2026RAPI restated all five daily points read a day earlier
18 August 2026Gahcho Kue sold 111% more carats at 44% less a carat

Go deeper

What would change this call

A window chosen by the page, not by the desk. Every percentage above is a 30-day trailing change, and the page publishes no longer series, so the window's start date is whatever the page happens to carry on the morning it is read. A month that begins just after a dip will show a rise on any series that has stopped falling, which is the honest reading of a 0.33% move on the 1-carat across 31 points. The reading that would change the desk's view is a longer published series: if Rapaport restores a 12-month view, a polished figure could be set against the rough claim directly instead of being declared untestable.

Why polished and rough diverge

RAPI is an index of asking prices for polished stones of a stated size, in the D to H colour and IF to VS2 clarity range, and it moves when the people holding polished inventory change what they ask. A mine's realised price a carat is the average of what it was actually paid for rough at its sales, across whatever mix of stones that parcel contained. The two can move in opposite directions for reasons that have nothing to do with either being wrong. Rough is bought by cutters who must fund polishing and financing before they sell, so a rough price carries their view of polished demand several months forward and their cost of capital now. A mine's average also moves with the grade and size mix it happens to recover, which is why Gahcho Kue's volume more than doubling in a quarter can pull its average down even if nothing repriced. Polished asking prices, by contrast, can hold or firm while rough falls, because holders of finished goods are not forced sellers in the way a producer with a reclamation bill is. For a reader pricing either, the practical rule is that the polished benchmark sets what a counter can ask and the realised rough price sets what a mine can survive on, and neither substitutes for the other.

Method

Rapaport's index page was fetched with a browser identity and saved as raw bytes before any figure was written: 200, 305,223 bytes. Levels were read out of the page's own embedded JSON block rather than from any rendered summary, 31 points per size across the window the page itself declares as 2 September to 2 October 2026. The figures printed are the first and last points of each series and are in the saved bytes by construction: $931.63, $959.58, $1,286.32, $1,315.39, $3,899.61, $3,912.48, $10,572.46, $10,659.16, $17,858.06 and $18,208.08. The change column is Carat Capital's arithmetic, last over first, and is owed as arithmetic rather than as a grep. The claim that the page's own unlabelled percentage equals the trailing-window change was tested rather than assumed: recomputation reproduces the page's printed 3, 2.26, 0.33, 0.82 and 1.96 on all five sizes, which is why the sentence establishing it is stated as a finding and the field is no longer described as unestablished in this paper's tooling. The rough figures are from Mountain Province's own Q2 2026 release of 13 August 2026, 200, 143,437 bytes, where '$50 per carat' returns 4 hits and '$90 per carat' 2, and the attribution of the restructuring to a price decline is in the company's release of 1 October 2026, 200, 74,501 bytes, where the phrase 'material decline in diamond prices over the past year' returns 1 hit. No comparative percentage is taken from any upstream desk: every division above was recomputed here and printed as a pair, share of and below. ARCHIVE, periods on both sides: RAPI by size over a trailing window, ours 2026-09-30 'point-four-three-on-the-one-carat' covers the window 30 August to 30 September 2026 (1-carat +0.43%, 0.30-carat +3.30%) | this item covers the window 2 September to 2 October 2026 (1-carat +0.33%, 0.30-carat +3.00%). DIFFERENT PERIOD, so ADVANCED, not a repeat. Nine archive hits for RAPI were opened and read, not counted, and the three closest were read in full: the 30 September window piece, the 21 September five-session piece and the 22 September restatement piece. None establishes what the page's own percentage field measures and none sets the polished window against a realised rough price, which is what is new here. The window percentages themselves are close to the 30 September figures and are printed as the supporting table rather than as the headline for that reason.

Sources3