Audit: QRS - Perilla for Health & Longevity

Audit conducted on 20/08/2026 20:43 using AI4L / Opus 5

Iterations

Summary

Items	Count
Total	93
Passed	83
Failed	0
N/A	10
Pass Rate	100.00%

• Total = Passed + Failed + N/A
• Pass Rate = Passed / (Passed + Failed) × 100
• N/A items are excluded from the pass rate calculation

---

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.	🟢	Protocol doses (6–7 mL, 50–200 mg, 700 mg twice daily), time-to-effect values (8–12 weeks, 21 days), all gate items, all tier items, and all ten biomarker targets trace verbatim or near-verbatim to the ER.
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.	🟢	The ER contains no empty-state or hedged-absence phrasing that the QRS carries; at-a-glance “stay modest or unrepeated” mirrors the ER Conclusion’s hedging.
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”).	🟢	Contraindications retain absolute framing; “Poor conversion to eicosapentaenoic acid and docosahexaenoic acid” keeps the ER’s Medium-tier severity.
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.	🟢	Contraindications come from the ER “Populations who should avoid Perilla” list, Key Interactions from the ER interaction bullets; no modifying factor is promoted to a decision gate.
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.	🟢	No PMIDs, author names, NCT identifiers, or brand names appear anywhere in the QRS.
1.6	The QRS does not introduce new attributions.	🟢	Generic study references (“the 12-month perilla trial”, “the knee-pain trial”, “the gut-microbiome trial”, “the Japanese cohorts”) are all ER wording.

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, deflationary-where-warranted register matches the ER throughout.
2.2	The tone of the QRS is simultaneously expert, accessible, objective, and data-driven, but also empowering and encouraging	🟢	Quantified targets and doses paired with plain-language framing.
2.3	The QRS reads as a trusted, knowledgeable guide rather than a prescriptive doctor	🟢	Content is stated as evidence and protocol convention, not instruction.
2.4	The QRS avoids language that implies medical or clinical advice	🟢	No imperatives; protocol cells are descriptive labels and values.
2.5	The QRS “presents information” instead of “providing guidance”, “recommending”, or “advising”	🟢	Monitoring cadence and qualitative items are stated as noun phrases, not directives.
2.6	The QRS never addresses “the reader” directly — it presents evidence, not guidance	🟢	No second-person pronouns anywhere in the file.
2.7	The QRS is written in plain language, avoiding unnecessary medical jargon	🟢	Acronyms from the ER are expanded (ALA → α-linolenic acid, EPA/DHA → eicosapentaenoic/docosahexaenoic acid, TRACP-5b → tartrate-resistant acid phosphatase 5b, INR → international normalised ratio).
2.8	Information is presented in a concise and very compact manner	🟢	Gate and tier items are stripped to key facts; biomarker targets condensed.
2.9	It DOES NOT address the reader directly	🟢	Confirmed by full-text search for “you”/”your”: no matches.
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.	🟢	Functional-range targets (hs-CRP <0.5 mg/L, ApoB <80 mg/dL) address the optimizing reader, not the general population.
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.	🟢	Ten-marker monitoring panel and unheated-oil handling assume high willingness to comply.
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.	🟢	No simplified or general-population framing present.
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.	🟢	At-a-glance foregrounds the conversion ceiling and the complement-not-replace conclusion, the decision-relevant point for this audience.
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.	🟢	The string “anti-aging” does not occur in the file.
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)”; “injection” not “shot”; “adverse event” not “bad reaction”). Direct quotes from sources are exempt.	🟢	“gastrointestinal adverse events”, “adverse”, “anaphylaxis”, “international normalised ratio” used throughout; the plain-language wording in At-A-Glance is mandated by item 7.4.

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 headings, gate headings, tier labels, and the three table column headers match the template byte-for-byte (lines 446, 490, 538, 569, 582, 607, 634, 638–640).
3.2	All “…” from the [qrs_template] are present in the the QRS.	🟢	All 38 template variables present; the repeatable marker_#* and qualitative_item# variables are correctly instantiated as marker_1–10 and qualitative_item_1–6.
3.3	Spans that are not addressed in a checklist item are left unchanged	🟢	The non-variable spans website=”evidence_review”, website=”audit”, and website=”full_review” are unchanged; a structural diff against the template shows differences only inside data-qrs-var spans and the metadata block.

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 mapped into the QRS is empty.
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.	🟢	action_1_label “Standard seed-oil protocol”, action_2_label “Leaf-extract protocol”, action_3_label “Best time of day” match the ER Therapeutic Protocol bold labels verbatim.
4.3	Labels are not paraphrased, abbreviated, or invented.	🟢	Time-to-effect labels (“Membrane fatty-acid changes”, “Lipid and inflammatory marker changes”, “Allergic-rhinitis symptom relief”) come from the ER “Time to effect” bullet; marker names match the ER biomarker table.
4.4	The QRS DOES NOT use emoji indicators (no 🟩, 🟥, 🟨, etc.). Color and emphasis are conveyed through CSS and bold labels.	🟢	No emoji present; the ER’s “⚠️ Conflicted” markers are correctly dropped.
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 to key facts within the template’s single-sheet structure; biomarker targets (marker_1, marker_8) are trimmed to fit rather than carried in full.

---

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.	🟢	Comment opens at line 2, immediately after the doctype at line 1.
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.	🟢	Opening “—” at line 3, closing “—” at line 13.
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; no metadata value is echoed elsewhere except the header date/model, which are their own template variables.
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.	🟢	Only duration: "00:04" is quoted, correctly, 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: perilla_2026-0825-1753_Opus_ER.md.
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: 26.7.02, matching 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: 2026-0820-2037.
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” — single word, no version.
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” — nickname plus version, no qualifier.
5.12	The filename of the document is stated as “qrs_filename: [filename of this document]”	🟢	Line 9 matches the file’s actual name, perilla_2026-0825-1753_Opus_QRS.html.
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.	🟢	All nine values are trimmed and unquoted except the colon-bearing duration value.

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: “Perilla for Health & Longevity - Quick Reference Sheet”, matching ER canonical_topic with the ampersand 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: “Perilla for Health & Longevity”.
6.3	[header_subline_date] is set to [qrs_creation_date reformatted as MM/DD/YYYY]	🟢	Line 421: “08/20/2026”, correctly derived from qrs_creation_date 2026-0820-2037.
6.4	[header_subline_model] is set to [qrs_creator_ai_fullname]	🟢	Line 425: “Opus 5”, matching the frontmatter 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 title and the template’s fixed subline; the ER’s alternate_names list is not carried over.

7. At-A-Glance Section

#	Description	Result	Comments
7.1	[at_a_glance] is dense, execution-oriented summary of the ER Conclusion section	🟢	Condenses all four Conclusion paragraphs into the decision-relevant core: what is firm, what is thin, and the complement-not-replace consequence.
7.2	[at_a_glance] is no longer than 60 words	🟢	Exactly 60 words.
7.3	Every fact in [at_a_glance] is supported by a distinct passage in the ER.	🟢	Richest source claim, separate leaf compounds, membrane rise, modest/unrepeated downstream results, poor conversion, seed allergy, and rapid spoilage each map to a distinct ER Conclusion sentence.
7.4	It DOES NOT use acronyms or technical classifications that require specialist knowledge, uses plain-language terms instead	🟢	Uses “plant omega-3”, “blood-fat”, “thinking results”, “fish forms”, “marine oils”; no acronyms.
7.5	It DOES NOT cite specific trials (names, years, sample sizes, p-values)	🟢	No trial name, year, sample size, or p-value present.
7.6	It DOES NOT cite effect sizes, relative risks, or statistical results	🟢	No numeric effect estimates present.

8. Contraindications

#	Description	Result	Comments
8.1	The section is derived from the ER Key Interactions & Contraindications section	🟢	All six items come from the ER’s “Populations who should avoid Perilla” list within that section.
8.2	[stop_items] represent the Contraindications from the ER	🟢	All six ER avoidance populations are represented, none added.
8.3	Individual [stop_items] are formatted as <li></li>	🟢	Six <li> elements at lines 572–577.
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.	🟢	The ER’s trailing clauses (“— an absolute contraindication given anaphylaxis reports”, “in whom lipid handling and clotting factor synthesis are both impaired”, “the group over-represented in serious adverse-event reports”) are all stripped.
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).	🟢	“below 50,000/µL”, “Within 14 days”, “Child-Pugh Class C”, “under 3 years”, and the haemophilia/von Willebrand examples are all preserved.
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 in the contraindication list.
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 ER identifies six such populations and the section is correctly populated, not empty.
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 ten items map one-to-one to the ten ER interaction bullets.
9.2	[caution_items] represent the Key Interactions from the ER, excluding any that are already listed as Contraindications	🟢	No overlap with the six contraindication items; all ten ER interactions carried.
9.3	Individual [caution_items] are formatted as <li></li>	🟢	Ten <li> elements at lines 585–597.
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.	🟢	All ER rationale sentences (“additive antiplatelet effect…”, “Separation of at least four hours preserves uptake”, “Mitigation is dose restraint — perilla oil held below 3 g daily”) are stripped; only the labels and drug lists remain.
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 ER named-drug parenthetical is preserved, including the five-item supplement list with Ginkgo biloba italicised as in the ER.
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 in the interaction list.
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 ER identifies ten interactions and the section is correctly populated.
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 derive from the ER Therapeutic Protocol bullets.
10.2	The three sets of [action] items cover the three most important actionable implementation aspects from the ER Protocol section	🟢	Seed-oil dose, leaf-extract dose, and timing are the three actionable levers; the remaining ER bullets are context (competing approaches, attribution, half-life, modifiers).
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 provides at least three actionable aspects; 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 variables carry ER-sourced content; no placeholders remain.

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	🟢	Membrane fatty-acid change, lipid/inflammatory marker change, and allergic-rhinitis relief are taken from the ER “Time to effect” bullet; the Low-tier bone marker window is correctly omitted for budget.
11.2	The sets are picked and ordered by the magnitude of the related benefit	🟢	time_1 maps to the High-tier omega-6/omega-3 correction; time_2 and time_3 map to Medium-tier lipid/inflammation and allergic rhinoconjunctivitis benefits.
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	The ER lists four time-to-effect aspects; all three sets are used.
11.4	All used [time_#label], [time#value], [time#_sub] items are filled with meaningful content derived from the ER.	🟢	All nine time variables carry ER-sourced content (12-week reassessment, baseline-lipid/CRP dependence, 8–12 week seasonal leaf-extract run).
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 provides time-to-effect data in Practical Considerations, so the section is correctly retained.

12. Benefits

#	Description	Result	Comments
12.1	The section is derived from the ER Expected Benefits section	🟢	All twelve benefit items correspond to ER benefit headings.
12.2	Key variables are [benefits_high], [benefits_medium], [benefits_low], [benefits_speculative]	🟢	All four variables present and populated at lines 540–562.
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.	🟢	Each item is the ER heading alone; all Magnitude figures, trial names, and sponsor notes are dropped.
12.4	Parenthetical content — including effect sizes, sample notes, mechanistic hints, and example studies — is stripped, NOT preserved.	🟢	No parentheses appear in any benefits variable; the ER’s “⚠️ Conflicted” markers are also removed.
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 ER benefit tiers contain items, 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 eight risk items correspond to ER risk headings at the same tiers.
13.2	Key variables are [risks_high], [risks_medium], [risks_low], [risks_speculative]	🟢	All four variables present and populated at lines 609–628.
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.	🟢	Each item is the ER heading alone; the 28.6% anaphylaxis rate, thirteen-fold oxidation figure, and RR 1.06 cancer estimate are all dropped.
13.4	Parenthetical content — including frequencies, severity grades, sample notes, mechanistic hints, and example studies — is stripped, NOT preserved.	🟢	No parentheses appear in any risks variable.
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 ER risk tiers contain items, so no span needs hiding.

14. Monitoring

#	Description	Result	Comments
14.1	The section is derived from the ER Monitoring section	🟢	All ten rows and the cadence line derive from the ER Monitoring Protocol & Defining Success section.
14.2	All measurable/quantifiable biomarkers from the Monitoring section are listed	🟢	All ten ER biomarker-table rows are present: red-cell ALA, Omega-3 Index, triglycerides, ApoB, hs-CRP, platelets/CBC, INR, TRACP-5b, PSA, ALT.
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.	🟢	Line 796–800 reproduces the ER’s ongoing-monitoring paragraph: 12-week panel, 6–12-month repeat, INR at 2 weeks and after dose changes, PSA and liver enzymes annually.

15. Qualitative Assessment

#	Description	Result	Comments
15.1	The section is derived from the ER Monitoring section	🟢	All six items come from the ER’s “Qualitative markers worth tracking alongside the laboratory set” list.
15.2	All subjective/qualitative biomarkers from the Monitoring section are listed	🟢	All six ER qualitative markers are carried verbatim: allergy symptom burden, bruising/bleeding, bowel regularity, joint comfort, energy/mood/clarity, and oil taste and smell.

Issues 20/08/2026 20:43

Pass rate 100.00%. No issues found.