Spectroscopy evidence flows into ICH-classified action items, dossier-section drafts, and ALCOA+ ledger entries — automatically. The regulatory work is the output of the science, not a separate document written in parallel.
Why this exists
Even with a well-organised internal data lake, the bridge from a final spectrum interpretation to a regulatory-ready argument is, in most laboratories, invisible by default. Reviewers spend weeks reconciling a structure-elucidation narrative against the raw evidence.
The FDA's January 2025 AI framework + EMA's reflection paper + ICH Q2(R2) all raised the bar — simultaneously asking R&D groups to adopt more AI, prove reproducibility, document every parameter, and keep the human chain visible to inspectors. The spreadsheet-and-email toolchain doesn't satisfy any of those four.
Regulatory Intelligence Hub closes that gap programmatically. The dossier is a by-product of the science — not a parallel document that gets reconciled in week 11.
The regulatory pipeline
Every stage emits a typed Pydantic record. Every output keys back to the input. The same recipe-hash replay that works for spectroscopy works here — re-derive any ledger entry from any prior date with bit-identical output.
Incoming evidence from SpectraCheck — peaks, categories, impurity matches, solvent hits — maps onto regulatory categories: residual solvent, elemental impurity, genotoxic alert, nitrosamine risk.
Emits
evidence_id · regulatory_category · ich_class
Each mapped finding lands in the right framework slot: ICH Q3C (residual solvent), Q3D (elemental impurity), M7 (mutagenic), Q3A/Q3B (drug substance / product impurities). Pharmacopoeia priors layered in (USP-NF, EP, JP).
Emits
framework_ref · pharmacopoeia · acceptance_window
Risk-based credibility framework runs end-to-end. FDA Stage 1-4 oversight gates evaluated automatically — items requiring human review surface in the queue with the exact gate that triggered them.
Emits
risk_tier · fda_stage · human_review_required
Every numerical claim hyperlinks back to its source: the spectrum file, the picked peak, the literature window, the reviewer who approved it. No confidence number without a trail — ever.
Emits
claim · source_uri · citation_chain
Dossier-section drafts in CTD format: Module 3.2.S.3.2 (Impurities), 3.2.S.4 (Control of Drug Substance), 3.2.P.5 (Control of Drug Product), AI/ML model documentation per FDA Jan 2025 framework.
Emits
ctd_section · draft · open_items
No release without an explicit human signoff. Reviewer attribution recorded with timestamp, role, and the exact artefact version they approved. Liability stays where regulators expect it.
Emits
reviewer · role · signed_at · artefact_hash
ALCOA+ audit-event entry written: attributable, legible, contemporaneous, original, accurate, complete, consistent, enduring, available. Inspection-ready by default.
Emits
audit_event · alcoa_fields · inspector_view_url
Frameworks covered
Each framework is mapped to concrete code paths — audit ledger, dossier composer, action-item routing — not abstract policy alignment.
Global pharmaceutical guidelines
AI / regulatory submissions (Jan 2025)
AI in medicinal-product lifecycle
Cross-framework foundation
Use cases shipped
Each row in this grid is a typed pipeline — not a workflow we'd build for you on request. Inputs come from SpectraCheck or your existing data lake; outputs land in the audit ledger and the dossier composer.
Auto-classifies detected solvent peaks against ICH Q3C Class 1/2/3 limits. Action items raised only when observed concentration crosses framework thresholds.
Inputs
Spectroscopy peaks · solvent_hit table · API
Outputs
Q3C verdict · acceptance status · dossier section draft
Maps detected elemental signals onto the Q3D 24-element catalog. PDE limits per Class 1/2A/2B/3. Risk assessment by route-of-administration.
Inputs
ICP-MS / element analysis · API · drug product
Outputs
Q3D risk matrix · PDE comparison · open items
Structural alert detection (ICH M7) + literature-prior matching. Flags candidate nitrosamines + suggests confirmatory MS/MS acquisition.
Inputs
Structure SMILES · synthesis route · API
Outputs
M7 verdict · recommended MS/MS · risk tier
Tracks the full analytical lifecycle: specificity, linearity, accuracy, precision, range, robustness. Every parameter recipe-hash-linked to the validation campaign.
Inputs
Method runs · acceptance criteria · campaigns
Outputs
Q2(R2) report · parameter coverage · gaps
FDA Jan 2025 model-card registry: training data lineage, performance bounds, validation corpus, intended use, known limitations.
Inputs
Model artefacts · validation runs · changelogs
Outputs
FDA Stage 1-4 documentation · audit-ready
Watches regulatory sources (ICH, FDA, EMA, pharmacopoeias) for version changes. Impacts auto-routed to affected dossiers, rules, and action items.
Inputs
Source feeds · pharmacopoeia versions · diffs
Outputs
Change-impact report · action items · ETA
Anatomy of a ledger entry
Below is the exact shape of an audit-event row when a reviewer signs a residual-solvent verdict. Every field is queryable by an inspector with no extra interpretation step.
audit_events · aev_018f3d9c · ICH Q3C residual-solvent verdict
| ALCOA+ field | Recorded value | What it means |
|---|---|---|
| Attributable | dr.chen@pharmaco.com · Senior Reviewer · 2026-05-28T14:32:18Z | Reviewer identity + role + signing timestamp |
| Legible | JSON · UTF-8 · stable Pydantic schema v3.4.0 | Machine + human-readable; schema versioned |
| Contemporaneous | Logged 14:32:18Z · 0.4s after detection event | Recorded at the moment of decision, not backfilled |
| Original | Raw FID SHA-256: 4f7a…b29e (immutable vault, never overwritten) | Source artefact preserved bit-identical |
| Accurate | Recipe hash: r9c2…1d3a (replay-verified bit-identical output) | Reproducible by recipe replay any time |
| Complete | All 12 evidence layers + 3 cross-modal checks documented | No partial-evidence claims permitted |
| Consistent | ICH Q3C Class 3 mapping confirmed against pharmacopoeia v2024.2 | Framework version pinned to decision |
| Enduring | Retention class: 25-year regulatory minimum | Retention policy attached to event |
| Available | Inspector view URL: /audit/event/aev_018f3d9c · API: GET /audit-events/{id} | Queryable on inspection |
The honest comparison
Most R&D groups today run regulatory workflows through spreadsheets + email + shared drives. Here's exactly what flips when the dossier becomes a product of the pipeline, not a parallel document.
| Dimension | Spreadsheet workflow | MolTrace workflow |
|---|---|---|
| Where evidence lives | today Spreadsheets, email attachments, shared drives, lab notebooks | with MolTrace Typed Pydantic records, hyperlinked to raw spectra by SHA-256 |
| How claims get cited | today Hand-written narratives, manual cross-reference, reviewer memory | with MolTrace Auto-generated citation chain per claim: spectrum → peak → reference window → reviewer |
| How signoff happens | today Email approval, PDF wet-signature, signed-and-scanned | with MolTrace In-app signoff queue · reviewer + role + timestamp + artefact hash recorded atomically |
| What an inspector sees | today Reconstructed audit trail from emails + version history + interviews | with MolTrace Single audit_events query · ALCOA+ fields per event · provenance URLs |
| Reproducibility (replay analysis 6 months later) | today Re-run from scratch · 'looks similar' is usually the verdict | with MolTrace Recipe-hash replay · bit-identical output guaranteed forever |
| Regulatory-change handling | today Manual scan of guidance documents · spot-check affected dossiers | with MolTrace Surveillance watches sources · diffs auto-routed to affected items with ETA |
Cross-module loop
The acetic-acid impurity from SpectraCheck's worked example — followed all the way through regulatory routing and back into ReactionIQ as a constraint on the next run. Audit ledger records every handoff.
Peak at 2.10 ppm classified as acetic acid impurity (confidence 0.93). Regulatory Hub receives the structured evidence object.
ICH Q3C: acetic acid is Class 3 (low toxic potential, no action below 5000 ppm). PDE limit retrieved from pharmacopoeia v2024.2.
Observed concentration below 5000 ppm threshold — informational only. Dossier section 3.2.S.3.2 updated; no human review queued.
Impurity limit auto-propagates as a Bayesian prior on the next reaction run. Audit ledger records the cross-module handoff.
Audit & data integrity
Every commitment below is implemented in production code, not aspirational. The audit_events table runs every claim through the same six pillars. Inspectors query directly; no reconstruction needed.
Immutable raw vault
Every FID is SHA-256 hashed. Vault path policy enforced. Never overwritten. Inspection-ready forever.
Recipe-hash provenance
Every processing run links a recipe hash to the unchanged raw archive. Replay any prior date and get bit-identical output.
Human signoff queue
FDA Stage 4 oversight gate. No regulatory document released without explicit qualified-human attribution.
21 CFR Part 11 alignment
Electronic-records + electronic-signatures requirements built into the audit_events table from day one.
Cross-modal contradiction warnings
HRMS exact mass disagreeing with NMR-implied formula raises a first-class warning before signoff — every time.
Tenant isolation by default
SOC 2 Type II controls, GDPR-compliant data residency, role-scoped audit-event ledger per tenant.
Pick a dossier section that's currently a 40-hour reconciliation. We'll walk through how the Regulatory Hub would handle the same evidence, end-to-end.