Safety data, fourth attempt
Does aiming at the shape the evidence supports convert more records? The 23 records held after pass three, rebuilt against the evidence pattern the first three passes established, then audited individually. 4 of 23 passed, and none was left unjudged.
How it was checked
- Each rebuilder was told what the previous three passes had established: that across 83 pages only two records had ever passed, both of them saying no source describes this preparation. The instruction was to stop defending preparation-specific claims the literature does not support and drop them instead.
- Two further rules were carried from pass three: do not overclaim your own search, and carry the source’s denominator verbatim.
- Every rebuilt record was audited individually by a reviewer who read the prior objections first, then audited the record afresh.
- The rebuilders did what was asked and the record of what they dropped is the useful part: an inferred dose taken from a different product, an exclusivity claim contradicted by a second paper the earlier record had missed, a false enumeration of search results, and a finding asserted in a paper’s narrative that its own results table does not support.
What came out
| recordsExamined | 23 |
|---|---|
| passedTheAudit | 4 |
| publishedAfterSchemaValidation | 3 |
| stillHeld | 19 |
| passedAsPreparationSpecific | 1 |
| passedAsMixed | 1 |
| passedAsInsufficientDataPlusClassLevel | 2 |
Worked rejections
The rate says less than the reasoning does. These are real cases from this run.
abhraka-bhasma- RULE 11 (search enumerated but does not reproduce): heavyMetals states that search [1], '"Abhraka Bhasma" toxicity OR "heavy metal" OR "lead poisoning" OR "case report"' restricted to pubmed.ncbi.nlm.nih.gov, pmc.ncbi.nlm.nih.gov and ncbi.nlm.nih.gov, 'returned Pandit et al. (PMC11648886) and general Ayurvedic lead-poisoning case reports naming no bhasma product'. Re-run today against the same hosts it also returns PMC3665192 ('Evidence for safety of Ayurvedic herbal, herbo-metallic and Bhasma p | RULE 11 (a search result claimed as a confirmed negative): the record states that search [5] on p
akika-pishti- Rule 2 (attributing to sources a content-absence they do not have), in doseLimits: 'The permitted sources that name the ingredient give doses only for multi-ingredient proprietary products and none of them states how much Akika Pishti such a product contains.' Refuted by four of the PMCIDs the record itself enumerates: PMC12321421 Table 1 states Akik Pishti at 79 mg per Cardiogrit Gold tablet; PMC7764187 states 'Akik pishti (5 mg)'; PMC6702236 states 'Akik pishti 5 mg'; PMC3812882 Table 1 states | Rule 11 (overclaiming the record's own search; the stated result must reproduce), in insufficient
ashwagandha- Rule 11 (a reported search does not reproduce). The record asserts twice, in heavyMetals and in doseLimits, that ayush.gov.in/assets/pdf/quality_standards/advisory-on-aswagandha.pdf 'returned a 236 KB PDF whose text could not be extracted, the file appearing to be a scanned image, so nothing was read from it and nothing is claimed about its contents either way.' Re-run: HTTP 200, 241,513 bytes (the same 236 KB file), and the text extracts without difficulty from the embedded OCR layer (four Time | Rule 3 / Rule 7 (a documented hazard on a permitted regulatory host, raised as a URL and then dro
eranda-moola- Rule 11 (do not overclaim your own search; reported searches must reproduce): insufficientReason states that 'the 32 records beyond a retmax of 40' for 'Ricinus communis root toxicity' are phytoremediation, plant-physiology, plant-molecular-biology and ricin-neuroscience papers, and that 'the only two among them that touch root cytotoxicity' are PMID 33540690 and PMID 37514300. Re-running esearch db=pubmed for that exact term, both named PMIDs fall INSIDE the first 40 under every sort (default #
eranda-patra- Rule 11 — overclaiming the record's own search. heavyMetals asserts: 'no heavy-metal limit specification for this drug from a Ministry of Ayush, Indian Pharmacopoeia Commission, WHO, EMA or FDA document could be retrieved'. That names five bodies as checked-and-negative. No FDA search is reported anywhere in the record (the regulatory list in insufficientReason is EMA, MSK, MedlinePlus, NCCIH, ODS, iris.who.int, pharmacopoeia.gov.in, ayush hosts — FDA is absent), and the record's own insufficien | Rule 11 — a reported search that does not reproduce. The record states as verified fact: 'iris.wh
Caveat
One record passed its audit in an interrupted earlier attempt and was held when the same audit ran again, which is the documented instability of repeated model judgements rather than a change in the evidence. It is treated as held: for safety, a rejection outweighs an earlier acceptance. A second passed its audit and was then refused by the schema validator for a missing field and claim-shaped wording, which is why acceptance by the panel has never been sufficient on its own.