Audit: QRS - Milk Thistle for Health & Longevity

Audit conducted on 22/09/2026 18:41 using AI4L / Opus 5

Iterations

Summary

Items	Count
Total	94
Passed	84
Failed	0
N/A	10
Pass Rate	100.00%

---

1. General Rules

#	Description	Result	Comments
1.1	Every claim, magnitude, label, recommendation, and statement in the QRS is literally supported by content in the source ER.	🟢	Spot-checked every populated span against the ER: protocol figures (ER 386, 390, 394), time-to-effect (ER 400, 447), benefit/risk headings (ER 166-238, 264-312), gate items (ER 336-362) and the full biomarker table (ER 481-490) all trace to ER text.
1.2	Where the ER uses cautious phrasing (“not formally studied”, “None documented in human trials to date”, “theoretical concern”, “data are limited”), the QRS uses the same phrasing.	🟢	Speculative tiers for the senotherapeutic, neuroprotection, MCT8 and ER-beta items mirror the ER’s own “Speculative” grading; “No effect on living longer or better has been shown” matches ER 524.
1.3	The QRS never strengthens an ER claim (e.g., “not formally studied” → “not required”) or softens one (e.g., “do not use during pregnancy” → “use with caution during pregnancy”).	🟢	No claim strengthened or softened; e.g. Child-Pugh Class C and the “unless drug levels are being actively monitored” qualifier are both carried over intact (ER 359, 362).
1.4	The QRS does not relabel an ER fact under a different decision category. A “Benefit-Modifying Factor” from ER section is not surfaced as a “Caution”; a “Risk-Modifying Factor” is not surfaced as a “Side Effect”; etc.	🟢	Benefit- and Risk-Modifying Factors (ER 243-255, 317-331) are not surfaced anywhere; gates draw only from ER Key Interactions & Contraindications.
1.5	PubMed IDs, study citations, expert names, clinical trial identifiers (NCT*), and brand names appear in the QRS only if they appear in the source ER for the same fact.	🟢	The QRS carries no PMIDs, study citations, expert names, NCT identifiers or brand names at all.
1.6	The QRS does not introduce new attributions.	🟢	No attributions of any kind are present.

2. Focus, Tone & Audience

#	Description	Result	Comments
2.1	The QRS follows the tone of the ER, which is determined by the ER’s own language, phrasing, and framing.	🟢	Measured, evidence-weighted register matching the ER.
2.2	The tone of the QRS is simultaneously expert, accessible, objective, and data-driven, but also empowering and encouraging	🟢	Expert but accessible; magnitudes are replaced by tiers and the lede states plainly where there is nothing to gain.
2.3	The QRS reads as a trusted, knowledgeable guide rather than a prescriptive doctor	🟢	Presents ranges and markers without instructing; no imperative voice.
2.4	The QRS avoids language that implies medical or clinical advice	🟢	No prescriptive verbs; the footer disclaimer is the unmodified template text.
2.5	The QRS “presents information” instead of “providing guidance”, “recommending”, or “advising”	🟢	No occurrence of “recommend”, “advise”, “should”, “must” or “consult” in document content.
2.6	The QRS never addresses “the reader” directly — it presents evidence, not guidance	🟢	No second-person pronouns anywhere in the document.
2.7	The QRS is written in plain language, avoiding unnecessary medical jargon	🟢	Plain language throughout; the technical terms that remain (silymarin, flavonolignans, Asteraceae, Child-Pugh) are the ER’s own and are load-bearing.
2.8	Information is presented in a concise and very compact manner	🟢	Benefits and Risks are reduced to bare tier labels; gate items carry no rationale.
2.9	It DOES NOT address the reader directly	🟢	Confirmed — no direct address.
2.10	The target audience is health- and longevity-oriented adults who are risk-aware, proactive, and actively seeking to optimize health or apply the intervention under review.	🟢	Framed for a proactive reader: baseline-dependent benefit, functional (not reference-range) targets, explicit stop point.
2.11	The target audience is willing to employ lifestyle and behavioral changes as well as follow protocols that may be inconvenient, costly, or require effort.	🟢	Presents a 12-week course with paired blood work, three-times-daily dosing and imaging as normal expectations.
2.12	The document is NOT written for the general population, who are unwilling to employ lifestyle and behavioral changes or follow protocols that may be inconvenient, costly, or require effort.	🟢	Functional targets (ALT under 25/20 U/L, hs-CRP under 0.5 mg/L) are well below conventional cut-offs.
2.13	Framing, takeaways, and risk/benefit weighting throughout the document reflect this audience, including where an intervention’s signal for the average person differs from its signal for this audience.	🟢	The lede states the signal for this audience explicitly: “Where numbers are already good, nothing to improve.”
2.14	The document’s own voice frames usage in longevity terms, not “anti-aging” (e.g., “anti-aging clinics”, “anti-aging community”, “anti-aging medicine”). Proper names that contain “anti-aging” (e.g., “American Academy of Anti-Aging Medicine”) are quoted verbatim.	🟢	“anti-aging” does not appear.
2.15	The document’s own voice uses formal clinical and scientific terminology, not colloquial or consumer-grade language (e.g., “oral medication” not “pill(s)”; “oral” / “administered orally” not “taken by mouth” / “given by mouth”; “injection” not “shot”; “adverse event” not “bad reaction”). This holds on EVERY surface, including the QRS lede — the stricter plain-language bar (see 2.7) does not license lay phrasing for route of administration. Direct quotes from sources are exempt.	🟢	Formal terminology throughout; route of administration is given as “Oral products” (Contraindications), not a lay equivalent.

3. Template Integrity

#	Description	Result	Comments
3.1	The following labels and headings on the QRS are fixed and not modified: • Card and section headings: “Protocol”, “Time to effect”, “Benefits”, “Risk & Side Effects”, “Monitoring”, “Qualitative Assessment” • Gate headings: “Contraindications”, “Key Interactions” • Tier labels: “High”, “Medium”, “Low”, “Speculative” • Table column headers in Monitoring: “Marker”, “Target”, “Why”	🟢	All fixed headings, gate headings, tier labels and the Marker/Target/Why column headers are byte-identical to [qrs_template].
3.2	All “…” from the [qrs_template] are present in the the QRS.	🟢	All 34 named template spans plus the repeatable marker_#* (10 rows) and qualitative_item# (5 items) spans are present.
3.3	Spans that are not addressed in a checklist item are left unchanged	🟢	Diff against [qrs_template] with variable regions masked shows no change outside the metadata block and the repeated monitoring/qualitative rows; the website=”evidence_review”, website=”audit” and website=”full_review” spans are untouched.

4. Formatting

#	Description	Result	Comments
4.1	When the source ER section is empty, the QRS uses the ER’s own empty-state phrasing verbatim. Typical phrasings are “None documented in human trials to date” and “Not formally studied”	N/A	No source ER section is empty — all four benefit and risk tiers, both gate lists, Protocol and Monitoring are populated.
4.2	Where the ER presents a bulleted item as “Label: content”, the QRS uses the ER’s bold label verbatim as the cell or row label.	🟢	Protocol cell labels “Standard dose”, “Split dosing” and “Timing” are the ER’s bold labels verbatim (ER 386, 390, 394); “Duration before judging response” likewise (ER 400).
4.3	Labels are not paraphrased, abbreviated, or invented.	🟢	No label paraphrased or abbreviated; the two time-to-effect aspect labels are derived per 11.4 rather than invented against an ER bold label.
4.4	The QRS DOES NOT use emoji indicators (no 🟩, 🟥, 🟨, etc.). Color and emphasis are conveyed through CSS and bold labels.	🟢	No emoji present in the file; the ER’s tier squares and the two “⚠️ Conflicted” markers are correctly stripped.
4.5	The QRS is designed to render on one A4 page. Any section that has more content in the ER than fits the per-section budget is condensed by the LLM, not extended onto a second page.	🟢	Every section is condensed rather than extended: benefits/risks reduced to headings only, gate items stripped of rationale, the ER’s Context/Notes column dropped from the biomarker table.

5. Metadata

#	Description	Result	Comments
5.1	The metadata is placed inside a single HTML comment that is the first element after “<!doctype html>” and before any other comment, head, or body content.	🟢	The metadata comment opens on line 2, immediately after <!doctype html>, before any other comment or head content.
5.2	Inside that HTML comment the YAML block is delimited by a line “—” opening and a line “—” closing. Text before the opening “—” is permitted but is not parsed as YAML.	🟢	YAML opens at line 3 and closes at line 13 inside the comment.
5.3	The metadata is not visible in any rendered view of the QRS and is not surfaced by any other element on the sheet.	🟢	Enclosed in an HTML comment and not echoed by any visible element.
5.4	All frontmatter values are trimmed: no leading or trailing whitespace, no surrounding quotes unless the value contains a colon, bracket, or leading special character that requires YAML quoting.	🟢	All values trimmed; only duration: “00:02” is quoted, which is required because the value contains a colon.
5.5	The filename of the source ER is stated as “er_filename: [er_filename]”	🟢	Line 4: er_filename: milk_thistle_2026-0814-0607_Opus_ER.md — matches the ER on disk.
5.6	Version of the QRS.md file used to create the document is stated as “qrs_prompt_version: [Version of QRS.md]”	🟢	Line 5: qrs_prompt_version: 26.9.22 — matches the QRS.md version badge.
5.7	Creation date and time of the document is stated as “qrs_creation_date: [YYYY-MMDD-HHMM]” (e.g., 2026-0501-1430)	🟢	Line 6: qrs_creation_date: 2026-0922-1823 — correct YYYY-MMDD-HHMM form.
5.8	The nickname of the AI used to create the document is stated as “qrs_creator_ai_nickname: [qrs_creator_ai_nickname]”	🟢	Line 7: qrs_creator_ai_nickname: Opus.
5.9	The nickname of the AI is just a single word model name without version, etc. (e.g., Opus, Sonnet, Grok, Gemini, ChatGPT)	🟢	“Opus” is a single word with no version or qualifier.
5.10	The full name of the AI used to create the document is stated as “qrs_creator_ai_fullname: [qrs_creator_ai_fullname]”	🟢	Line 8: qrs_creator_ai_fullname: Opus 5.
5.11	The full name of the AI consists of the [qrs_creator_ai_nickname] and the model version number and no additional qualifier (e.g., Opus 4.6, Sonnet 3.2, Grok 4.5, Gemini 3.1, ChatGPT 5.4)	🟢	“Opus 5” is nickname plus version number with no trailing qualifier.
5.12	The filename of the document is stated as “qrs_filename: [filename of this document]”	🟢	Line 9: qrs_filename: milk_thistle_2026-0814-0607_Opus_QRS.html — matches the actual filename.
5.13	All frontmatter values are trimmed: no leading or trailing whitespace, no surrounding quotes unless the value contains a colon, bracket, or leading special character that requires YAML quoting.	🟢	Re-verified: no stray whitespace or unnecessary quoting on any of the nine keys.

6. Page Title & Header

#	Description	Result	Comments
6.1	[page_title] is set to the [canonical_topic] of the ER frontmatter followed by “ - Quick Reference Sheet” (e.g., “Intervention - Quick Reference Sheet”). The [canonical_topic] is HTML-entity-encoded as needed (e.g., & for &)	🟢	Line 22: “Milk Thistle for Health & Longevity - Quick Reference Sheet” — canonical_topic with the ampersand entity-encoded.
6.2	[header_topic] is set to the [canonical_topic] of the ER frontmatter, with HTML entities encoded as needed (e.g., & for &)	🟢	Line 417: “Milk Thistle for Health & Longevity”.
6.3	[header_subline_date] is set to [qrs_creation_date reformatted as MM/DD/YYYY]	🟢	Line 421: 09/22/2026, the MM/DD/YYYY form of qrs_creation_date 2026-0922.
6.4	[header_subline_model] is set to [qrs_creator_ai_fullname]	🟢	Line 425: “Opus 5”, matching qrs_creator_ai_fullname.
6.5	No additional header content appears: no badge, version stamp, AKA / alternate names line, source-AI attribution, audit date, or QRS variant marker.	🟢	Header contains only the template subline; no badge, alternate-names line, version stamp or audit date.

7. At-A-Glance Section

#	Description	Result	Comments
7.1	[at_a_glance] opens by saying what the intervention is — its kind (e.g., dietary supplement, prescription medication, peptide, plant, procedure, practice) — and what it is used for, in plain language, before any verdict on the evidence	🟢	Opens “Milk thistle is a plant seed extract sold as a dietary supplement for liver support” — kind and purpose stated before any verdict.
7.2	[at_a_glance] is dense, execution-oriented summary of the ER Conclusion section	🟢	Condenses ER 524-526: enzyme and glycemic effects, absent longevity benefit, no headroom when baseline is normal, and the two weighted cautions.
7.3	[at_a_glance] is no longer than 70 words	🟢	69 words, verified by word count.
7.4	Every fact in [at_a_glance] is supported by a distinct passage in the ER.	🟢	Each of the five clauses maps to a distinct passage in the ER Conclusion (ER 524, 526).
7.5	It DOES NOT use acronyms or technical classifications that require specialist knowledge, uses plain-language terms instead	🟢	No acronyms; uses “liver enzymes”, “high blood sugar”, “inflammation markers”, “daisy-family allergy” rather than ALT/AST, hyperglycemia, hs-CRP or Asteraceae.
7.6	It DOES NOT cite specific trials (names, years, sample sizes, p-values)	🟢	No trial names, years, sample sizes or p-values.
7.7	It DOES NOT cite effect sizes, relative risks, or statistical results	🟢	No effect sizes, risk ratios or statistics.

8. Contraindications

#	Description	Result	Comments
8.1	The section is derived from the ER Key Interactions & Contraindications section	🟢	All six items trace to ER 336-362.
8.2	[stop_items] represent the Contraindications from the ER	🟢	Carries the five populations at ER 358-362 plus the absolute contraindication on substituting oral products for intravenous silibinin (ER 354).
8.3	Individual [stop_items] are formatted as <li></li>	🟢	Six <li> elements inside the [stop_items] span.
8.4	Items are as concise as possible. No trailing explanations, no elaborations, no mechanistic rationale, no attributions, no citations, no study details. No content after an em-dash, en-dash, or hyphen-dash (e.g., “— dose reduction required”, “— reduced efficacy”) — these trailing clauses are stripped. Just the key fact.	🟢	Trailing rationale is stripped throughout — “the most severe grade”, “where no benefit is established and enzyme monitoring is uninterpretable”, “pending human data on estrogen receptor beta activation” and “where controlled safety data remain limited to a single small trial” are all absent; no dash-trailing clauses remain.
8.5	Parenthetical qualifiers from the ER bullet — time windows, severity classes, threshold values, clinical staging — ARE preserved as part of the item, kept as concise as possible (shortened or trimmed where needed to fit the one-page budget, but never dropped entirely).	🟢	Qualifiers preserved: the Asteraceae example list, “Child-Pugh Class C”, “outside a supervised trial” and “unless drug levels are being actively monitored”.
8.6	When the ER uses ranking notation inside parens (e.g., “>” for severity ordering) that depends on an explanatory phrase to interpret, normalize the items to a plain comma-separated list rather than carrying through the bare symbol.	N/A	The ER uses no ranking notation inside parentheses in this section.
8.7	The section is left empty ONLY IF the ER’s Key Interactions & Contraindications section identifies no population, condition, or scenario for which the intervention should be avoided, deferred, or used only under specialist supervision.	🟢	The section is populated, which is correct — the ER names five populations that should avoid the intervention (ER 358-362).
8.8	If the section is left empty, an HTML comment inside the [stop_items] span records the ER basis for the absence, naming or quoting the ER text relied on. E.g.	N/A	The section is not empty.

9. Key Interactions

#	Description	Result	Comments
9.1	The section is derived from the ER Key Interactions & Contraindications section	🟢	All seven items trace to the bullets at ER 336-352.
9.2	[caution_items] represent the Key Interactions from the ER, excluding any that are already listed as Contraindications	🟢	Includes the seven usage-changing interactions and correctly omits the anti-rejection drugs and intravenous silibinin bullets (already contraindications) and the cytochrome P450 bullet, for which the ER states no interaction was demonstrated (ER 342).
9.3	Individual [caution_items] are formatted as <li></li>	🟢	Seven <li> elements inside the [caution_items] span.
9.4	Items are as concise as possible. No trailing explanations, no elaborations, no mechanistic rationale, no attributions, no citations, no study details. No content after an em-dash, en-dash, or hyphen-dash (e.g., “— dose reduction required”, “— reduced efficacy”) — these trailing clauses are stripped. Just the key fact.	🟢	Only the ER bold labels are carried; consequences and mitigations are stripped, as is the em-dash gloss “cholesterol drugs that bind bile” (ER 346).
9.5	Parenthetical qualifiers from the ER bullet — example drug lists, time windows, severity classes, threshold values, clinical staging — ARE preserved as part of the item, kept as concise as possible (shortened or trimmed where needed to fit the one-page budget, but never dropped entirely).	🟢	Every example drug list is preserved: glipizide/glyburide, raloxifene/mycophenolate/ezetimibe, levothyroxine, cholestyramine, berberine/chromium/cinnamon/alpha-lipoic acid, and NAC/TUDCA/choline/vitamin E.
9.6	When the ER uses ranking notation inside parens (e.g., “>” for severity ordering) that depends on an explanatory phrase to interpret, normalize the items to a plain comma-separated list rather than carrying through the bare symbol.	N/A	The ER uses no ranking notation inside parentheses in this section.
9.7	The section is left empty ONLY IF the ER’s Key Interactions & Contraindications section identifies no interaction, additive effect, or exposure that changes how the intervention is used.	🟢	The section is populated, which is correct — the ER names multiple usage-changing interactions.
9.8	If the section is left empty, an HTML comment inside the [caution_items] span records the ER basis for the absence, naming or quoting the ER text relied on. E.g.	N/A	The section is not empty.

10. Protocol

#	Description	Result	Comments
10.1	The section is derived from the ER Protocol section	🟢	All three cells trace to ER 386-410.
10.2	The three sets of [action] items cover the three most important actionable implementation aspects from the ER Protocol section	🟢	Standard dose, split dosing and timing are the three executable levers; the remaining Protocol bullets are pharmacology, competing branded formulations or explicit non-differentiation by sex, age and genotype.
10.3	If less that three distinct actionable implementation aspects are mentioned in the ER the unused sets are left empty and made invisible, not filled with placeholder text or empty-state phrasing.	N/A	The ER supplies well over three actionable implementation aspects, so all three sets are used.
10.4	All used [action_#label], [action#value], [action#_sub] items are filled with meaningful content derived from the ER Protocol section.	🟢	All nine action fields carry ER-derived content (ER 386, 390, 394).

11. Time to Effect

#	Description	Result	Comments
11.1	The three sets of [time] items cover the three most important time-to-effect aspects from the ER	🟢	Enzyme response window, evaluation duration and absence of a perceptible effect are the only time-to-effect aspects the ER supplies (ER 400, 447, 449, 477).
11.2	The sets are picked and ordered by the magnitude of the related benefit	🟢	Liver enzymes — the ER’s sole High-tier benefit — is placed first, followed by the evaluation window and then the null perceptual signal.
11.3	If less that three distinct time-to-effect aspects are mentioned in the ER the unused sets are left empty and made invisible, not filled with placeholder text or empty-state phrasing.	N/A	Three distinct time-to-effect aspects exist in the ER, so no set is unused.
11.4	All used [time_#label], [time#value], [time#_sub] items are filled with meaningful content derived from the ER.	🟢	All nine time fields are ER-derived: “4 to 12 weeks” and “pooled effects strongest at durations of two months or less” (ER 447), the 12-week evaluation window (ER 400, 477), and “nothing perceptible happens within days” / “no felt effect to titrate against” (ER 447, 449).
11.5	If the ER does not provide any information on time to effect, the section is removed completely from the Protocol Panel	N/A	The ER does provide time-to-effect information, so the section is correctly retained.

12. Benefits

#	Description	Result	Comments
12.1	The section is derived from the ER Expected Benefits section	🟢	All items trace to the ER headings at ER 166-238.
12.2	Key variables are [benefits_high], [benefits_medium], [benefits_low], [benefits_speculative]	🟢	All four tier spans are present and populated.
12.3	Items are as concise as possible. No explanations, no elaborations, no effect sizes, no qualifiers, no attributions, no citations, no study details, no mechanistic explanations, etc. Just the key fact.	🟢	Bare benefit headings only; no magnitudes, confidence intervals, study names or mechanisms carried over.
12.4	Parenthetical content — including effect sizes, sample notes, mechanistic hints, and example studies — is stripped, NOT preserved.	🟢	No parenthetical content appears in any tier.
12.5	If no items of a specific sub-section (high, medium, low, speculative) are present the respective is set to “display=none”, not filled with “None documented in human trials to date” or similar empty-state phrasing.	N/A	All four tiers have at least one ER item, so no span needs hiding.

13. Risks

#	Description	Result	Comments
13.1	The section is derived from the ER Potential Risks & Side Effects section	🟢	All items trace to the ER headings at ER 264-312.
13.2	Key variables are [risks_high], [risks_medium], [risks_low], [risks_speculative]	🟢	All four tier spans are present and populated.
13.3	Items are as concise as possible. No explanations, no elaborations, no effect sizes, no qualifiers, no attributions, no citations, no study details, no mechanistic explanations, etc. Just the key fact.	🟢	Bare risk headings only; risk ratios, the 3,700% content spread and the contamination detail are all left out.
13.4	Parenthetical content — including frequencies, severity grades, sample notes, mechanistic hints, and example studies — is stripped, NOT preserved.	🟢	No parenthetical content appears in any tier.
13.5	If no items of a specific sub-section (high, medium, low, speculative) are present the respective is set to “display=none”, not filled with “None documented in human trials to date” or similar empty-state phrasing.	N/A	All four tiers have at least one ER item, so no span needs hiding.

14. Monitoring

#	Description	Result	Comments
14.1	The section is derived from the ER Monitoring section	🟢	Derived from ER Monitoring Protocol & Defining Success (ER 475-490).
14.2	All measurable/quantifiable biomarkers from the Monitoring section are listed	🟢	All ten rows of the ER biomarker table are present — ALT, AST, GGT, ALP, total bilirubin, HbA1c, fasting insulin, hs-CRP, liver fat fraction and TSH — with targets and rationales matching the ER verbatim.
14.3	[monitoring_cadence] is populated with the monitoring cadence/frequency derived from the ER Monitoring section. It is not left with placeholder text or empty.	🟢	Populated from ER 475, 477 and the liver-fat imaging note at ER 489; baseline, 12-week, 6-month and 6-to-12-month intervals plus the first-month glucose self-monitoring are all carried.

15. Qualitative Assessment

#	Description	Result	Comments
15.1	The section is derived from the ER Monitoring section	🟢	Derived from the qualitative marker list at ER 494-498.
15.2	All subjective/qualitative biomarkers from the Monitoring section are listed	🟢	All five qualitative markers are present and unaltered.

Issues 22/09/2026 18:41

Pass rate 100.00%. No issues found.

Issues 22/09/2026 18:39

1. 9.2 — Anti-rejection interaction duplicates gate: The Key Interactions entry “Anti-rejection drugs with a narrow safety margin (sirolimus, tacrolimus, cyclosporine)” (line 578) restates the Contraindications entry for organ transplant recipients on calcineurin- or mTOR-acting anti-rejection drugs (line 564), which 9.2 requires to be excluded.

Fixes 22/09/2026 18:39

1. 9.2 — Anti-rejection interaction duplicates gate: Removed the Key Interactions item “Anti-rejection drugs with a narrow safety margin (sirolimus, tacrolimus, cyclosporine)” from [caution_items], since the same population is already covered by the Contraindications item on organ transplant recipients taking calcineurin- or mTOR-acting anti-rejection drugs.

Issues 22/09/2026 18:34

1. 1.2 — Hedge dropped in time-to-effect cell: [time_2_sub] (QRS lines 509–510) states “if nothing has moved by then, it is doing nothing measurable in that individual”, while the ER (line 477) frames this as an interpretation — “the reasonable reading is that it is doing nothing measurable in that individual”.

Fixes 22/09/2026 18:34

1. 1.2 — Restored ER hedge in time cell: [time_2_sub] changed from “if nothing has moved by then, it is doing nothing measurable in that individual” to “if nothing has moved by then, the reasonable reading is that it is doing nothing measurable in that individual”, matching the ER’s cautious phrasing.

Issues 22/09/2026 18:28

1. 9.2 — CYP drugs listed as an interaction: The ninth [caution_items] entry lists “Drugs broken down by cytochrome P450 enzymes (midazolam, warfarin, dextromethorphan)” as a Key Interaction, but ER line 342 states “No interaction demonstrated in controlled human dosing at 140 mg silymarin three times daily for 14 days; no routine dose adjustment applies.”
2. 9.4 — Trailing explanatory clauses on two items: Two [caution_items] carry elaborations after a colon — “Other liver-targeted supplements (…): additive rather than hazardous” and “Drugs broken down by cytochrome P450 enzymes (…): no interaction demonstrated” — where 9.4 requires the key fact alone.

Fixes 22/09/2026 18:28

1. 9.2 — CYP drugs removed from Key Interactions: Deleted the [caution_items] entry “Drugs broken down by cytochrome P450 enzymes (midazolam, warfarin, dextromethorphan): no interaction demonstrated”, since the ER states no interaction was demonstrated and no dose adjustment applies.
2. 9.4 — Trailing explanatory clause stripped: Changed “Other liver-targeted supplements (N-acetylcysteine, tauroursodeoxycholic acid, choline, vitamin E): additive rather than hazardous” to end at the parenthetical drug list, leaving the key fact alone.