AI lab · The Wanderer’s Notes

Yaroslav Nagul · MRICS, REV

Contract No. 1 · 7 October 2026

The Shapeshifter

A paraphrase of a standard passing itself off as a quotation.

A hooded figure, one half made of neat parchment pages, the other dissolving into ink; it holds a scroll whose lines turn into scribbles
Illustration generated by AI (GPT Image) from my brief
  1. Notice

    An executing agent’s test handover contained four references to standards, complete with standard, paragraph and content. The executor had checked its own work and reported ‘6 of 6, 0 discrepancies’. The script that matches quotations against the primary-source text found none of the four.

  2. Tracks

    Zero out of four meant not four errors in the script but one problem in the handover: the table held descriptions of the requirements, not quotations. The opposing agents checked the substance of the descriptions: two references cited IVS 104 and IVS 105 for points that sit in IVS 102 and IVS 103, and one cited an outdated edition of EVS.

  3. The creature

    I named it the Shapeshifter: a paraphrase that passes itself off as a quotation. It sounds like the standard, but a script has nothing to match it against, so a wrong standard or edition hides behind plausible text.

  4. The hunt

    The cure lay in the template, not the agent. The handover template gained a separate column, ‘quotation of the requirement’, which takes only verbatim text from the primary source with its location. A converter turns this column into input for the script, and rows without a verbatim quotation are skipped, not counted as claims.

  5. Trophy

    During the end-to-end test, the script caught a quotation from the definition of market value attached to the wrong place in the standard. The IVS glossary says ‘an asset or liability’; the appendix to IVS 102 says ‘an asset and/or liability’. ‘A quotation, not a description’ is one of the eight rules of my method.

Sources: IVS 2025, Glossary, para. 10.18 (PDF p. 10); IVS 102, Appendix, para. A10.01 (PDF p. 25); the lab’s decision log, 7 October 2026.