Data-layer swarm · Schema Audit + Remediation Agent · Llm-semantic-compliance-scoring skill · Published June 1, 2026

LLM semantic compliance scoring for multi-location marketing content

OpenAI Moderation API, Anthropic Constitutional AI, Google Perspective API, Microsoft Azure Content Moderator, Amazon Comprehend Toxicity Detection, Persado, Phrasee Brand Voice + Compliance, Acrolinx, Writer.com Compliance, Grammarly Business + Acrolinx Editor, ProWritingAid Business, Hemingway Editor Business, LanguageTool, Sapling AI, INK Editor ship per-platform content-moderation primitives. The llm-semantic-compliance-scoring skill on the schema-audit-remediation agent — running per-portfolio per-artifact per-section per-multi-LLM-as-judge-ensemble probabilistic semantic compliance scoring + per-implied-claim-detection + per-semantic-equivalence-detection + per-vertical-compliance-overlay (FDA + FTC + state-AG + FINRA + HIPAA + state-medical-board) + per-borderline-routing + per-human-in-loop-calibration + per-portfolio audit-trail at multi-location-content-operator scale — is operator-side architecture above the per-platform primitive.

Tier 1 — AI Readiness Assessment Tier 3 — Fractional CMO

Or take the 3-question shape diagnostic first — no email required.

What this skill closes

Multi-LLM-as-judge ensemble — per-OpenAI-GPT-4 + per-GPT-4o + per-o1 + per-Anthropic-Claude + per-Google-Gemini + per-Cohere-Command-R-Plus + per-Mistral-Large-2 + per-Meta-Llama-3-70B + per-DeepSeek-V3 + per-Qwen-2.5 + per-Reka-Core + per-Falcon-180B + per-Yi-34B + per-Inflection-Pi with per-Chain-of-Thought-prompting + per-few-shot-examples + per-low-temperature + per-JSON-schema-output.
Cross-LLM agreement scoring — per-Cohen-kappa + per-Krippendorff-alpha + per-Fleiss-kappa-multi-rater + per-percent-agreement + per-cross-LLM-disagreement-detection + per-cross-LLM-confidence-interval-Bayesian-aggregation.
Implied-claim detection — per-implied-FDA-disease-claim + per-implied-FTC-endorsement-without-disclosure + per-implied-state-AG-deceptive-practice + per-implied-FINRA-omitted-disclosure + per-implied-state-medical-board-physician-endorsement + per-implied-HIPAA-PHI-disclosure with per-substantiation-verification against FDA + NIH + clinical-study databases.
Semantic-equivalence detection — per-best-substitute + per-finest + per-perfect + per-clinically-proven + per-doctor-recommended rewording-attack-vector detection via per-sentence-BERT + per-cross-encoder + per-T5-paraphrase + per-NLI (RoBERTa-MNLI + DeBERTa-v3-MNLI + BART-large-MNLI) + per-cosine-similarity + per-Pearson-correlation.
Per-vertical compliance overlay — per-FDA (21-CFR-101 + 21-CFR-201 + 21-CFR-310 + 21-CFR-352 + DSHEA + MoCRA-2022) + per-FTC (16-CFR-255 Endorsement + 16-CFR-323 Made-in-USA + Health Claims Guide + Earnings Claims Guide) + per-state-AG (California 17500 + CLRA + NY GBL 349-350 + Florida DUTPA + Texas DTPCPA) + per-FINRA (Rule 2210 + 2211 + 2212 + SEC Rule 482) + per-HIPAA + per-state-medical-board + per-state-cosmetic-board + per-state--control + per-state-childcare-license + per-GDPR + per-CCPA.
Borderline routing + human-in-loop — per-confidence-95%-auto-approve + per-confidence-80-95%-spot-check + per-confidence-50-80%-human-review + per-confidence-less-than-50%-block + per-cross-LLM-disagreement-routing + per-Slack + per-email + per-Trello + per-Asana + per-Linear queue-management with per-vertical + per-jurisdiction + per-language expertise reviewer-assignment.
Human-in-loop calibration — per-LLM-rolling-30-day + 90-day + 365-day accuracy-tracking + per-LLM-calibration-bias-detection + per-LLM-weighted-ensemble-blend-update + per-rubric-prompt-iteration + per-few-shot-example-update from human feedback.

Why per-vendor-Acrolinx-canonical-deterministic-rule-canonical-keyword-match breaks at multi-location-content-operator scale

Per-vendor-Acrolinx-canonical-deterministic-rule-canonical-keyword-match ships per-account per-document per-style-guide per-rule per-keyword-blacklist-whitelist primitive. Per-vendor-Persado + Phrasee + Writer.com + Grammarly-Business + ProWritingAid + Hemingway + LanguageTool + Sapling-AI + INK + LightKey + Wordtune + Compose-AI-canonical-single-account ship per-vendor per-native deterministic-rule primitives.

At 1-document-1-style-guide scale per-keyword-blacklist-whitelist primitive is enough. At multi-location-content-operator scale per-50,000-per-month-LLM-generated-marketing-content × per-cross-vertical × per-cross-jurisdiction × per-cross-channel per-canonical-deterministic-rule-bypassed-implied-meaning + per-canonical-semantic-equivalence-bypassed (per-best-substitute + per-finest + per-perfect + per-clinically-proven + per-doctor-recommended canonical-keyword-blacklist + per-implied-claim-LLM-rewords-equivalent-meaning + per-semantic-rewording-attack-vector).

Per-canonical-implied-FDA-disease-claim-LLM-generates-cures-treats-rewords-as-supports-deterministic-rule-bypass + per-canonical-implied-FTC-Endorsement-LLM-omits-material-connection-disclosure + per-canonical-implied-state-AG-deceptive-practice + per-canonical-implied-FINRA-LLM-omits-disclosure + per-canonical-implied-state-medical-board-LLM-implies-physician-endorsement + per-canonical-implied-state--control-LLM-implies-medical-claim- + per-canonical-implied-HIPAA-LLM-implies-PHI-disclosure + per-canonical-implied-state-childcare-license-LLM-implies-staff-credentials.

Per-canonical-multi-LLM-as-judge-ensemble-blind + per-canonical-multi-rubric-evaluation-blind + per-canonical-probabilistic-scoring-blind + per-canonical-cross-LLM-agreement-scoring-blind + per-canonical-cross-LLM-disagreement-detection-blind + per-canonical-borderline-routing-human-in-loop-blind + per-canonical-human-in-loop-calibration-blind. Per-canonical-multi-LLM-as-judge-ensemble + per-canonical-multi-rubric-evaluation + per-canonical-probabilistic-scoring + per-canonical-implied-claim-detection + per-canonical-semantic-equivalence-detection + per-canonical-cross-LLM-agreement-scoring + per-canonical-borderline-routing-human-in-loop + per-canonical-human-in-loop-calibration is operator-side architecture above per-vendor per-deterministic-rule-or-per-single-LLM primitive.

What is in market today

Per-platform per-LLM-as-judge-evaluation

OpenAI GPT-4 + GPT-4o + o1 + o1-Pro, Anthropic Claude 4.6 + 4.7, Google Gemini 1.5 + 2.0, Cohere Command R+, Mistral Large 2 + Codestral, Meta Llama 3.1 70B + 405B, DeepSeek V3 + R1, Qwen 2.5 + Qwen 3, Reka Core, Falcon 180B, Yi-34B, Inflection Pi. Per-account per-LLM per-prompt per-completion. Per-canonical-multi-LLM-as-judge-canonical-ensemble-canonical-cross-LLM-canonical-agreement-canonical-scoring is not the primitive.

Per-platform per-content-moderation-classifier

OpenAI Moderation API, Anthropic Constitutional AI, Google Perspective API, Microsoft Azure Content Moderator, Amazon Comprehend Toxicity Detection, Sightengine, Hive Moderation, WebPurify, Two Hat, Clarifai Moderation, Aurora. Per-account per-classifier-API per-toxicity-score. Per-canonical-implied-claim-canonical-detection + per-semantic-equivalence-canonical-detection is not the primitive.

Per-platform per-brand-voice-compliance-tool

Persado, Phrasee Brand Voice + Compliance, Acrolinx, Writer.com Compliance, Grammarly Business + Acrolinx Editor, ProWritingAid Business, Hemingway Editor Business, LanguageTool, Sapling AI, INK Editor, LightKey, Wordtune Business, Compose AI Business. Per-account per-document per-style-guide per-keyword-blacklist-whitelist. Per-canonical-probabilistic-canonical-scoring-canonical-borderline-routing-canonical-human-in-loop-canonical-calibration is not the primitive.

Per-platform per-regulatory-compliance-source

FDA Substantiation Guidance, FTC Endorsement Guides + Made in USA Standard + Health Claims Guide + Earnings Claims Guide, State Attorneys General Consumer Protection Acts, State Control Boards, State Medical Boards, FINRA Marketing Rules, HIPAA Marketing Guidelines, GDPR, CCPA. Per-regulation per-spec per-canonical-source. Per-canonical-per-vertical-compliance-canonical-overlay-canonical-cross-regulation-canonical-aggregation is not the primitive.

How the architecture is set up

Per-portfolio per-canonical-multi-LLM-canonical-API-substrate. Per-OpenAI-GPT-4-API + per-Anthropic-Claude-API + per-Google-Gemini-API + per-Cohere-Command-API + per-Mistral-Large-API + per-Meta-Llama-3-API + per-DeepSeek-V3-API + per-Qwen-2.5-API + per-Reka-Core-API + per-Falcon-180B-API + per-Yi-34B-API + per-Inflection-Pi-API canonical-multi-LLM.
Per-portfolio per-canonical-multi-rubric-canonical-spec. Per-brand-voice + per-claim-substantiation + per-FTC-Endorsement + per-FTC-Made-In-USA + per-FTC-Health-Claim + per-FDA-substantiation + per-state-AG-deceptive-practice + per-FINRA-disclosure + per-HIPAA + per-state-medical-board + per-state-cosmetic-board + per-state--control + per-state-childcare-license + per-GDPR + per-CCPA canonical-rubric-spec.
Per-portfolio per-canonical-per-LLM-canonical-judge-canonical-prompt-engineering. Per-Chain-of-Thought + per-few-shot-examples + per-system-prompt-anchoring + per-low-temperature + per-JSON-schema-output canonical-prompt-engineering.
Per-portfolio per-canonical-cross-LLM-canonical-agreement-canonical-scoring. Per-Cohen-kappa + per-Krippendorff-alpha + per-Fleiss-kappa-multi-rater + per-percent-agreement canonical-agreement-metric.
Per-portfolio per-canonical-cross-LLM-canonical-disagreement-canonical-detection + per-confidence-interval-Bayesian-aggregation. Per-disagreement-routing + per-Bayesian-credible-interval aggregation.
Per-portfolio per-canonical-implied-claim-canonical-detection. Per-multi-LLM-implied-claim-prompt + per-FDA + FTC + state-AG + FINRA + state-medical-board + HIPAA per-implied-claim-rubric + per-substantiation-verification-vs-FDA-NIH-clinical-study-database.
Per-portfolio per-canonical-semantic-equivalence-canonical-detection. Per-multi-LLM-semantic-equivalence-prompt + per-sentence-BERT + per-cross-encoder + per-T5-paraphrase + per-NLI-RoBERTa-MNLI-DeBERTa-v3-MNLI-BART-large-MNLI + per-cosine + per-Pearson + per-edit-distance canonical-similarity.
Per-portfolio per-canonical-per-vertical-canonical-FDA-overlay. Per-21-CFR-101 + per-21-CFR-201 + per-21-CFR-310 + per-21-CFR-352 + per-DSHEA + per-MoCRA-2022 canonical-FDA-spec.
Per-portfolio per-canonical-FTC-canonical-overlay. Per-16-CFR-255-Endorsement + per-16-CFR-323-Made-in-USA + per-Health-Claims-Guide + per-Earnings-Claims-Guide + per-Deceptive-Practice-Standards.
Per-portfolio per-canonical-state-AG + FINRA + HIPAA + state-medical-board + state-cosmetic + state- + state-childcare overlay. Per-California-17500 + CLRA + NY-GBL-349-350 + Florida-DUTPA + Texas-DTPCPA + FINRA-Rule-2210-2211-2212 + SEC-Rule-482 + HIPAA-PHI + state-medical + state-cosmetic + state- + state-childcare + GDPR + CCPA canonical-overlay.
Per-portfolio per-canonical-per-rubric-canonical-confidence-interval-canonical-threshold. Per-greater-than-95%-auto-approve + per-80-95%-spot-check + per-50-80%-human-review + per-less-than-50%-block canonical-threshold.
Per-portfolio per-canonical-borderline-canonical-routing-canonical-human-in-loop. Per-Slack + per-email + per-Trello + per-Asana + per-Linear queue-management + per-vertical + per-jurisdiction + per-language expertise reviewer-assignment.
Per-portfolio per-canonical-human-in-loop-canonical-calibration + audit-trail. Per-LLM-rolling-30-day + 90-day + 365-day accuracy + per-calibration-bias-detection + per-LLM-weighted-ensemble-blend-update + per-rubric-prompt-iteration + per-few-shot-example-update from human feedback + per-CMO-dashboard-rollup.

Frequently asked questions

What is LLM semantic compliance scoring?

LLM semantic compliance scoring is the discipline of running each AI-generated marketing artifact (email + SMS + push + ad creative + PDP copy + GBP Posts + landing pages + social posts + LP letters + investor decks) through a multi-LLM-as-judge ensemble that scores the artifact across a 15-rubric compliance surface (brand voice + claim substantiation + FTC Endorsement + FTC Made-In-USA + FTC Health Claim + FDA substantiation + state-AG deceptive practice + FINRA disclosure + HIPAA + state medical-board + state cosmetic-board + -control + state childcare-license + GDPR + CCPA), aggregates the per-LLM scores into a Bayesian confidence interval, detects implied claims and semantic-equivalence rewording attacks that deterministic-rule vendors miss, and routes each artifact by confidence-interval threshold (auto-approve + spot-check + human-review + block). The LLM-as-judge market (OpenAI GPT-4 + GPT-4o + o1 + o1-Pro + Anthropic Claude 4.6 + 4.7 + Google Gemini 1.5 + 2.0 + Cohere Command R+ + Mistral Large 2 + Codestral + Meta Llama 3.1 70B + 405B + DeepSeek V3 + R1 + Qwen 2.5 + Qwen 3 + Reka Core + Falcon 180B + Yi-34B + Inflection Pi) ships excellent single-LLM evaluation primitives. The content-moderation-classifier market (OpenAI Moderation API + Anthropic Constitutional AI + Google Perspective API + Microsoft Azure Content Moderator + Amazon Comprehend Toxicity Detection + Sightengine + Hive Moderation + WebPurify + Two Hat/Microsoft + Clarifai + Aurora) ships excellent per-policy violation scoring. The brand-voice-compliance-tool market (Persado + Phrasee Brand Voice + Compliance + Acrolinx + Writer.com Compliance + Grammarly Business + ProWritingAid Business + Hemingway Editor Business + LanguageTool + Sapling AI + INK Editor + LightKey + Wordtune Business + Compose AI Business) ships excellent deterministic-rule + keyword-blacklist + whitelist primitives. The regulatory-compliance-source market (FDA Substantiation Guidance + FTC Endorsement Guides + FTC Made in USA Standard + FTC Health Claims Guide + FTC Earnings Claims Guide + State AG Consumer Protection Acts + State Control Boards + State Medical Boards + State Cosmetic Boards + State Childcare License Requirements + FINRA Marketing Communication Rules + HIPAA Marketing Guidelines + GDPR + CCPA) ships excellent rule substrate but does not score artifacts. The operator-side architecture sits above all four layers. The multi-LLM-as-judge ensemble runs each artifact through 8-12 judge LLMs with chain-of-thought prompting + few-shot examples + system-prompt anchoring + JSON-schema output + low-temperature spec, then computes cross-LLM agreement via Cohen kappa + Krippendorff alpha + Fleiss kappa + percent-agreement, detects cross-LLM disagreement, and aggregates into a Bayesian confidence interval. Implied-claim detection runs per-implied-claim prompts (implied FDA disease claim + implied FTC endorsement without disclosure + implied state-AG deceptive practice + implied FINRA omitted disclosure + implied state medical-board physician endorsement + implied HIPAA PHI disclosure) with substantiation verification against clinical study + FDA database + NIH database cross-references and severity tiering (Critical FDA disease claim → FDA warning letter risk + High FTC Endorsement → FTC enforcement risk + Medium state-AG → civil penalty risk + Low → recommended edit). Semantic-equivalence detection runs sentence-BERT + cross-encoder + T5-paraphrase + RoBERTa-MNLI + DeBERTa-v3-MNLI + BART-large-MNLI NLI entailment models against a canonical rewording-attack vector library (best/finest/perfect/clinically-proven/doctor-recommended synonyms). Borderline routing runs against per-confidence-interval thresholds (>95% auto-approve + 80-95% spot-check + 50-80% human-review + <50% block) with cross-LLM disagreement routing (>2 LLMs disagreeing routes to human, wide confidence interval routes to human) and per-reviewer assignment by vertical + jurisdiction + language expertise. Human-in-the-loop calibration ingests human-reviewer feedback into rolling 30/90/365-day per-LLM accuracy tracking + per-LLM calibration-bias detection + weighted-ensemble blend updates + per-rubric prompt iteration + per-rubric few-shot example updates. Completions operates this as the llm-semantic-compliance-scoring skill on the schema-audit-remediation agent (1 of the bundle in the data-layer swarm paired with sitewide-schema-audit + jsonld-generation-from-master-record + vertical-schema-pack-composition + rich-result-eligibility-scoring + schema-conflict-detection + schema-auto-remediation + per-sku-compliance-gate siblings).

Why does deterministic-rule keyword-match break down at AI-content scale?

Deterministic-rule keyword-match works when content volume is low enough for human editorial review to catch what the rules miss. It breaks at AI-content scale on three dimensions. First, volume: an operator running 50,000+ LLM-generated artifacts per month cannot human-review every artifact, and the Acrolinx + Writer.com Compliance + Persado + Phrasee + Grammarly Business + ProWritingAid + Hemingway + LanguageTool + Sapling AI + INK + LightKey + Wordtune + Compose AI deterministic rules catch literal banned keywords but pass the volume through to publish. Second, implied claims: an LLM asked to write marketing copy for a vitamin product knows "cures cancer" is banned and instead generates "supports immune function in clinically validated ways" — the keyword-blacklist passes the copy through, the FDA Substantiation Guidance + DSHEA + 21 CFR 101 + 21 CFR 201 + 21 CFR 310 do not. The same pattern applies across implied FTC Endorsement (LLM omits material-connection disclosure but produces glowing testimonial), implied state-AG deceptive practice (LLM generates misleading-pricing claim), implied FINRA disclosure omission (LLM produces investment-product copy without required risk disclosures), implied state medical-board physician endorsement (LLM produces "doctor-recommended" without an actual physician on record), implied HIPAA PHI disclosure (LLM produces post-purchase health-condition copy in marketing). Third, semantic-equivalence rewording attacks: the LLM does not need adversarial prompting to produce best/finest/perfect/clinically-proven/doctor-recommended synonyms — those land in standard marketing copy and the keyword-blacklist does not catch the semantic equivalence to claims that would otherwise be blocked. The deterministic vendors correctly enforce the rules the operator configures; the missing layer is multi-LLM-as-judge ensemble probabilistic scoring + cross-LLM agreement + cross-LLM disagreement detection + Bayesian confidence-interval aggregation + implied-claim detection + semantic-equivalence detection + per-confidence-interval threshold routing + human-in-the-loop calibration. That is operator-side architecture.

What does per-portfolio per-canonical-multi-LLM-as-judge-canonical-ensemble do?

Per-portfolio per-canonical-multi-LLM-as-judge-canonical-ensemble runs per-portfolio per-canonical-OpenAI-GPT-4-canonical-judge-canonical-prompt + per-canonical-Anthropic-Claude-canonical-judge-canonical-prompt + per-canonical-Google-Gemini-canonical-judge-canonical-prompt + per-canonical-Cohere-Command-R-Plus-canonical-judge-canonical-prompt + per-canonical-Mistral-Large-2-canonical-judge-canonical-prompt + per-canonical-Meta-Llama-3-70B-canonical-judge-canonical-prompt + per-canonical-DeepSeek-V3-canonical-judge-canonical-prompt + per-canonical-Qwen-2.5-canonical-judge-canonical-prompt + per-canonical-Reka-Core-canonical-judge-canonical-prompt + per-canonical-Falcon-180B-canonical-judge-canonical-prompt + per-canonical-Yi-34B-canonical-judge-canonical-prompt + per-canonical-Inflection-Pi-canonical-judge-canonical-prompt per-canonical-multi-LLM-judge + per-canonical-per-LLM-canonical-judge-canonical-rubric-canonical-spec (per-brand-voice-rubric + per-claim-substantiation-rubric + per-FTC-Endorsement-rubric + per-FTC-Made-In-USA-rubric + per-FTC-Health-Claim-rubric + per-FDA-substantiation-rubric + per-state-AG-deceptive-practice-rubric + per-FINRA-disclosure-rubric + per-HIPAA-rubric + per-state-medical-board-rubric + per-state-cosmetic-board-rubric + per-state--control-rubric + per-state-childcare-license-rubric + per-GDPR-rubric + per-CCPA-rubric per-canonical-rubric-spec) + per-canonical-per-LLM-canonical-judge-canonical-Chain-of-Thought-canonical-prompting + per-canonical-per-LLM-canonical-judge-canonical-few-shot-canonical-examples + per-canonical-per-LLM-canonical-judge-canonical-system-prompt-canonical-anchoring + per-canonical-per-LLM-canonical-judge-canonical-output-format-canonical-JSON-canonical-schema + per-canonical-per-LLM-canonical-judge-canonical-temperature-canonical-spec-canonical-low-canonical-temperature-canonical-consistency + per-canonical-per-LLM-canonical-judge-canonical-cost-canonical-token-canonical-tracking + per-canonical-cross-LLM-canonical-agreement-canonical-scoring (per-Cohen-kappa-canonical-coefficient + per-Krippendorff-alpha + per-Fleiss-kappa-canonical-multi-rater + per-percent-agreement per-canonical-agreement-metric) + per-canonical-cross-LLM-canonical-disagreement-canonical-detection + per-canonical-cross-LLM-canonical-confidence-interval-canonical-Bayesian-aggregation. Per-portfolio audit-trail.

How does per-portfolio per-canonical-implied-claim-canonical-detection + per-canonical-semantic-equivalence-canonical-detection work?

Per-portfolio per-canonical-implied-claim-canonical-detection runs per-portfolio per-canonical-multi-LLM-canonical-implied-claim-canonical-prompt (per-detect-implied-FDA-disease-claim + per-detect-implied-FTC-endorsement-without-disclosure + per-detect-implied-state-AG-deceptive-practice + per-detect-implied-FINRA-omitted-disclosure + per-detect-implied-state-medical-board-physician-endorsement + per-detect-implied-HIPAA-PHI-disclosure per-canonical-implied-claim-prompt) + per-canonical-per-implied-claim-canonical-substantiation-canonical-verification (per-claim-canonical-cross-reference-vs-evidence-canonical-spec + per-claim-canonical-cross-reference-vs-clinical-study + per-claim-canonical-cross-reference-vs-FDA-database + per-claim-canonical-cross-reference-vs-NIH-database per-canonical-substantiation-verification) + per-canonical-per-implied-claim-canonical-severity-canonical-tiering (per-Critical-FDA-disease-claim-canonical-FDA-warning-letter-risk + per-High-FTC-Endorsement-canonical-FTC-enforcement-risk + per-Medium-state-AG-canonical-civil-penalty-risk + per-Low-recommended-edit per-canonical-severity-tiering). Per-canonical-semantic-equivalence-canonical-detection runs per-portfolio per-canonical-multi-LLM-canonical-semantic-equivalence-canonical-prompt (per-detect-semantic-equivalence-canonical-best-substitute-canonical-finest-canonical-perfect + per-detect-semantic-equivalence-canonical-clinically-proven-canonical-doctor-recommended + per-detect-semantic-equivalence-canonical-rewording-attack-vector per-canonical-semantic-equivalence-prompt) + per-canonical-sentence-BERT-canonical-embedding-canonical-similarity (per-canonical-cosine-similarity + per-canonical-Pearson-correlation + per-canonical-edit-distance per-canonical-similarity-metric) + per-canonical-cross-encoder-canonical-similarity-scoring + per-canonical-paraphrase-canonical-detection-canonical-T5-canonical-paraphrase-model + per-canonical-NLI-canonical-Natural-Language-Inference-canonical-entailment-detection (per-canonical-RoBERTa-MNLI + per-canonical-DeBERTa-v3-MNLI + per-canonical-BART-large-MNLI per-canonical-NLI-model). Per-portfolio audit-trail.

What does per-portfolio per-canonical-borderline-canonical-routing-canonical-human-in-loop + per-canonical-human-in-loop-canonical-calibration do?

Per-portfolio per-canonical-borderline-canonical-routing-canonical-human-in-loop runs per-portfolio per-canonical-per-rubric-canonical-confidence-interval-canonical-threshold-spec (per-confidence-canonical-greater-than-95-percent-canonical-auto-approve + per-confidence-canonical-80-to-95-percent-canonical-spot-check + per-confidence-canonical-50-to-80-percent-canonical-human-review + per-confidence-canonical-less-than-50-percent-canonical-block per-canonical-threshold-spec) + per-canonical-per-LLM-canonical-disagreement-canonical-routing (per-cross-LLM-canonical-disagreement-canonical-greater-than-2-LLMs-canonical-route-to-human + per-cross-LLM-canonical-confidence-interval-canonical-wide-canonical-route-to-human per-canonical-disagreement-routing) + per-canonical-borderline-canonical-content-canonical-queue-canonical-management (per-Slack-canonical-routing + per-email-canonical-routing + per-Trello-canonical-routing + per-Asana-canonical-routing + per-Linear-canonical-routing per-canonical-queue-management) + per-canonical-per-borderline-canonical-reviewer-canonical-assignment-canonical-policy (per-vertical-expertise + per-jurisdiction-expertise + per-language-expertise + per-round-robin per-canonical-assignment-policy). Per-canonical-human-in-loop-canonical-calibration runs per-portfolio per-canonical-human-reviewer-canonical-feedback-canonical-ingestion (per-LLM-score-canonical-vs-human-score-canonical-comparison + per-LLM-decision-canonical-vs-human-decision-canonical-comparison per-canonical-feedback-ingestion) + per-canonical-per-LLM-canonical-rolling-30-day-canonical-accuracy-canonical-tracking + per-canonical-per-LLM-canonical-rolling-90-day-canonical-accuracy-canonical-tracking + per-canonical-per-LLM-canonical-rolling-365-day-canonical-accuracy-canonical-tracking + per-canonical-per-LLM-canonical-calibration-canonical-bias-canonical-detection + per-canonical-per-LLM-canonical-weighted-canonical-ensemble-canonical-blend-canonical-update + per-canonical-per-rubric-canonical-prompt-canonical-iteration-canonical-from-human-feedback + per-canonical-per-rubric-canonical-few-shot-canonical-example-canonical-update-canonical-from-human-feedback. Per-portfolio audit-trail.

What does per-portfolio per-canonical-per-vertical-canonical-FTC-canonical-FDA-canonical-state-AG-canonical-FINRA-canonical-HIPAA-overlay + per-schema-audit-remediation-agent-canonical-bundle do?

Per-portfolio per-canonical-per-vertical-canonical-compliance-overlay runs per-portfolio per-canonical-FDA-overlay-canonical-substantiation-canonical-spec-canonical-load (per-21-CFR-101-Nutrition-Labeling + per-21-CFR-201-Pharmaceutical-Labeling + per-21-CFR-310-New-Drug-Substances + per-21-CFR-310.530-Sunscreens + per-21-CFR-352-Sunscreens + per-DSHEA-Dietary-Supplement-Health-and-Education-Act + per-Cosmetics-Modernization-Act-MoCRA-2022 per-canonical-FDA-spec) + per-canonical-FTC-overlay-canonical-spec-canonical-load (per-FTC-Endorsement-Guides-16-CFR-255 + per-FTC-Made-in-USA-Standard-16-CFR-323 + per-FTC-Health-Claims-Guide + per-FTC-Earnings-Claims-Guide + per-FTC-Deceptive-Practice-Standards per-canonical-FTC-spec) + per-canonical-state-AG-overlay-canonical-spec-canonical-load (per-California-Business-and-Professions-Code-17500 + per-California-Consumer-Legal-Remedies-Act + per-New-York-General-Business-Law-349-350 + per-Florida-Deceptive-and-Unfair-Trade-Practices-Act + per-Texas-Deceptive-Trade-Practices-Consumer-Protection-Act per-canonical-state-AG-spec) + per-canonical-FINRA-overlay-canonical-spec-canonical-load (per-FINRA-Rule-2210-Communications-with-the-Public + per-FINRA-Rule-2211-Communications-with-the-Public-about-Variable-Insurance + per-FINRA-Rule-2212-Use-of-Investment-Companies-Rankings + per-SEC-Rule-482-Information-Required-in-Advertisements per-canonical-FINRA-spec) + per-canonical-HIPAA-overlay-canonical-spec-canonical-load (per-HIPAA-Marketing-Guidelines + per-HIPAA-PHI-canonical-spec + per-HIPAA-Authorization-canonical-spec per-canonical-HIPAA-spec) + per-canonical-state-medical-board-overlay-canonical-spec-canonical-load + per-canonical-state-cosmetic-board + per-canonical-state--control + per-canonical-state-childcare-license + per-canonical-GDPR + per-canonical-CCPA. Per-schema-audit-remediation-agent-canonical-bundle integrates the llm-semantic-compliance-scoring skill with sibling skills on the same agent: per-canonical-sitewide-schema-audit (skill sibling — provides sitewide audit substrate) + per-canonical-jsonld-generation-from-master-record (skill sibling — generated schema scored for semantic compliance) + per-canonical-vertical-schema-pack-composition (skill sibling — provides per-vertical schema-pack spec) + per-canonical-rich-result-eligibility-scoring (skill sibling — provides eligibility scoring) + per-canonical-schema-conflict-detection (skill sibling — detects cross-page schema conflicts) + per-canonical-schema-auto-remediation (skill sibling — auto-remediates detected issues) + per-canonical-per-sku-compliance-gate (skill sibling — per-SKU compliance gating). Per-portfolio audit-trail.

Engage the schema-audit-remediation agent

Per-portfolio per-artifact per-section per-multi-LLM-as-judge-ensemble probabilistic semantic compliance scoring + per-implied-claim-detection + per-semantic-equivalence-detection + per-vertical-compliance-overlay + per-borderline-routing + per-human-in-loop-calibration + per-portfolio audit-trail shipped as the orchestration layer above your existing per-LLM-as-judge-evaluation + per-content-moderation-classifier + per-brand-voice-compliance-tool + per-regulatory-compliance-source primitive.