Restricting every supplier comparison to buyer-sourced records is the sensitivity analysis the scorecards publish. On the current dataset it changes no ordering that was not already inside measurement noise.
multiple_compare
Twenty suppliers, sixty-six compounds, and the multiplicity problem.
Section 1
Contribution
Comments technically on records submitted by others. Annotations are signed and are not moderated for agreement.
Records submitted
0
n=0
Annotations
92
n=92
Median purity submitted
suppressed
n=0
n<3 — see individual records
Suppliers covered
0
n=0
This handle has submitted no records. Annotators comment on other people's submissions and are not required to submit.
Section 2
Annotations
Signed technical comments on records. The index does not moderate annotations for agreement with its own arithmetic, and several of these disagree with it. Where a handle has made the same point on several records, one instance is shown and the rest are reachable from the records themselves.
What this record does establish is on the page and it is substantial. What it does not establish is listed under the provenance block rather than left for the reader to infer.
The index records the absence rather than imputing a value. An unmeasured property is unmeasured, and the specification block above says so explicitly.
Sixty-six compounds and twenty suppliers is a lot of cells. Some records will sit below specification by chance alone, and a single one is not a supplier finding. The per-compound table on the scorecard is the correct denominator.
Worth flagging to whoever submits next on this lot.
No disagreement from me.
Fair.
Comparing supplier-supplied results across suppliers compares their sample-selection practices as much as their material. The buyer-sourced subsets are the comparable series.
Comparing suppliers on endotoxin using only the records that carry a figure compares testing practice as much as material. The index states the denominator on every such comparison.
Recovery varies by claim size, by compound and by salt form. Pooling all of those into a supplier median mixes four effects and the per-compound table is the correction.
Showing 10 distinct comments of 92 annotations by this handle.