Below is an actual report as the client received it, published with the client's written permission. Name, firm, and every identifying number altered so the piece cannot be traced; the structure, register, and findings are exactly as delivered. The altered numbers still reconcile, because the disguise gets audited too.
Publish after two fixes: one bar re-rendered, one number reconciled. The text is exact everywhere I pushed.
All fifteen risk shares recomputed from the five printed spread widths, nothing taken from the chart on trust. Equal notional shares come out 4.6, 10.4, 15.0, 21.5, 48.5 percent and the chart's bars sit on all five. The claimed mirror between equal notional and inverse variance was tested rather than admired: it holds exactly, and for a reason the piece never stated, the five widths pair off to a constant product, the middle one to itself, so the symmetry is structural, not approximate. That is worth a sentence in the post; it is a strength you built without claiming it.
The inverse variance bar at the narrowest width is drawn near 33% where your own footer, and the requirement that shares sum to 100, both demand 48.5%. As drawn the bars total the mid-80s, and the chart contradicts the exact mirror symmetry the post teaches. The other four bars are exact to my recomputation; it is only that one, which looks like a stale value from an earlier render. Re-render before this goes anywhere.
The text says Sharpe 1.36 for the middle rule; the chart prints 1.37, twice. One of them is the median. Pick it and make them agree.
One piece this report; charged as one post. The batch stands at its agreed count, and a resend after your fixes verifies in minutes, the hashes do the work.
One finding of the forty-four counted in the public census. Batches arrive as per-piece sections like this one, under a headlines block that says up front what publishes untouched and what needs a hand. Prices and the census live on the referee desk →