Intake
1940 observations were collected for the 26 Oct 2026 session. 143 of them reached the calculation, 7.4%. This page is the difference, itemised.
A row qualifies for a print only if it clears every rule of the unit definition: the right GPU, a provider the panel admits through the collector that saw it, an on-demand price rather than a term commitment, firm or list capacity rather than spot, a datacenter inside the block being priced, a node at or above the size floor, a currency there is a rate for, and a price inside the band. Most collected rows fail one of those, and most of the time that is the index working rather than failing: a broad catalogue holds a great many GPUs it does not price.
Each row below is attributed to the first rule it failed, so the counts add up to the session exactly. That also means a rule's count is not the number of rows that would print if it were loosened — a row stopped at the panel may be outside the block and spot-priced as well.
Where the session's rows stopped
Every rule, with what it holds and why.
| Rows | Share | Rule | Recorded as |
|---|---|---|---|
| 606 | 31.2% | a GPU the index does not price, or a variant that is not the reference unit of its class | not_a_reference_variant |
| 461 | 23.8% | the panel does not admit this provider, through this collector, in this class | not_in_panel |
| 364 | 18.8% | sold outside the region block being priced, or with no country recorded at all | outside_block |
| 280 | 14.4% | spot, interruptible or community capacity, which the unit excludes | tier_excluded |
| 143 | 7.4% | reached the calculation | admitted |
| 75 | 3.9% | a term-committed price, not the on-demand reference unit | term_not_reference |
| 11 | 0.6% | fewer GPUs than the node-size floor | below_node_floor |
What each collector contributed
A collector at zero is not necessarily broken. Several collect against region blocks that are not published yet, or sit outside the panel on purpose so they can accumulate the history a future constituent needs. Where nothing a collector produces can reach a print at all, that is marked, because the reason is a decision somebody should be making rather than a fact nobody has noticed.
| Collector | Collected | Admitted | Yield | Where the rest stopped |
|---|---|---|---|---|
| gpuhunt | 1351 | 107 | 7.9% | the panel does not admit this provider, through this collector, in this class |
| azure_retail | 196 | 22 | 11.2% | a GPU the index does not price, or a variant that is not the reference unit of its class |
| ovh | 110 | 3 | 2.7% | a GPU the index does not price, or a variant that is not the reference unit of its class |
| civo | 87 | 0 | 0.0% | a GPU the index does not price, or a variant that is not the reference unit of its class |
| coreweave | 34 | 0 | 0.0% | a GPU the index does not price, or a variant that is not the reference unit of its class |
| hyperstack | 33 | 0 | 0.0% | a GPU the index does not price, or a variant that is not the reference unit of its class |
| lambda_pricing no route to a print | 28 | 0 | 0.0% | the panel does not admit this provider, through this collector, in this class |
| digitalocean | 23 | 1 | 4.3% | a GPU the index does not price, or a variant that is not the reference unit of its class |
| vast_ai | 22 | 2 | 9.1% | spot, interruptible or community capacity, which the unit excludes |
| scaleway | 19 | 5 | 26.3% | a GPU the index does not price, or a variant that is not the reference unit of its class |
| latitude | 18 | 0 | 0.0% | a GPU the index does not price, or a variant that is not the reference unit of its class |
| together no route to a print | 8 | 0 | 0.0% | the panel does not admit this provider, through this collector, in this class |
| runpod | 5 | 2 | 40.0% | sold outside the region block being priced, or with no country recorded at all |
| seeweb | 4 | 1 | 25.0% | a term-committed price, not the on-demand reference unit |
| voltagepark no route to a print | 2 | 0 | 0.0% | sold outside the region block being priced, or with no country recorded at all |
What moved
A constituent that was qualifying and has stopped, or a rule holding a materially different number of rows than it held recently. A collector whose rows all fail one rule looks exactly like a collector that is working: the session succeeds, the stored row count rises, and the print survives on the constituents it already had. That is the failure this watches for.
Nothing moved
No constituent stopped qualifying on this session, and no rule is holding a materially different number of rows than it held recently.
Collected against admitted
Rows collected each session beside rows that reached the calculation. The two columns moving apart is the ordinary consequence of adding a collector, whose rows are stored for weeks before a methodology version admits them.
| Session | Collected | Admitted | Yield |
|---|---|---|---|
| 6 Oct 2026 | 2073 | 159 | 7.7% |
| 7 Oct 2026 | 2077 | 159 | 7.7% |
| 8 Oct 2026 | 2077 | 160 | 7.7% |
| 9 Oct 2026 | 2072 | 158 | 7.6% |
| 10 Oct 2026 | 2087 | 160 | 7.7% |
| 11 Oct 2026 | 2099 | 163 | 7.8% |
| 12 Oct 2026 | 2080 | 160 | 7.7% |
| 13 Oct 2026 | 2070 | 159 | 7.7% |
| 14 Oct 2026 | 2082 | 159 | 7.6% |
| 15 Oct 2026 | 2086 | 159 | 7.6% |
| 16 Oct 2026 | 2084 | 160 | 7.7% |
| 17 Oct 2026 | 2099 | 162 | 7.7% |
| 18 Oct 2026 | 2105 | 163 | 7.7% |
| 19 Oct 2026 | 2027 | 153 | 7.5% |
| 20 Oct 2026 | 2088 | 160 | 7.7% |
| 21 Oct 2026 | 2082 | 159 | 7.6% |
| 22 Oct 2026 | 2087 | 160 | 7.7% |
| 23 Oct 2026 | 2079 | 160 | 7.7% |
| 24 Oct 2026 | 2090 | 161 | 7.7% |
| 25 Oct 2026 | 2078 | 160 | 7.7% |
| 26 Oct 2026 | 1940 | 143 | 7.4% |