Review before your audience reviews you.

The errors that survive AI passes and self-review are the ones this desk catches.

An independent referee for quantitative work: LinkedIn and X posts, white papers, journal submissions, theses, backtests, and the code underneath them. Every number is recomputed from the material alone, every chart is checked against its own text, every citation against its source. Findings come back in writing, in your register, ready to apply, and the corrected version is checked again once you have applied them.

No post or paper yet? Code alone counts. Send the script that will not run, or the one that runs and gives a number you do not trust, and it gets the same treatment: read line by line, the bug named, the fix written out.

What those look like:

• the chart that contradicts its own caption
• the stale version that undoes last week's fix
• the last sentence that lost its ending
• the off-by-one that shifts every return by a day and still prints a Sharpe

The method was built on graduate and post-doctoral work. See the deliverable before you write: a delivered report, redacted, published with the client's written permission.

Citation checks ride along free: references verified against their sources, paraphrases checked for honest distance from the original. All reviews and materials remain confidential; only the client decides if anyone ever sees one. Ongoing volume is quoted as a bundle. Independent referee, affiliated with no journal. Not investment advice; I check the artifact, not the strategy.

What a week of this actually catches

One client, 35 works, every number recomputed: 44 findings across 22 of the 35, and 13 clean passes. Counts only; no names, no content. Running total for the same client, 14 August to 8 September: 45 works, 71 findings across 30 of them, 15 clean passes.

And see the deliverable itself: a delivered report, redacted · download it as a PDF. Published with the client's written permission, every identifying number altered and still reconciling.