Probabilistic decisioning for regulated pharma
Synlogica Terminus reads your evidence — logger files, freight tenders, supplier offers — and turns it into a clear, governed recommendation. Every step is recorded for the auditor. Your people make the final call.
BUILT IN EU · GDPR · GAMP 5 CAT 4 · 21 CFR PART 11 · EU GMP ANNEX 11 · EU GDP
Follow batch PL-2026-0417-C through the chain of custody ↓
Decision package
Three decisions, one engine
Three decision domains, each a Terminus module sharing one governed decision framework — so rules, versioning and the audit trail behave identically whether the question is a temperature excursion, a freight tender or a supplier price.
Temperature excursions assessed with Arrhenius + Monte-Carlo against the product's stability model — a recommendation with stated uncertainty, not a reviewer's guess.
See Quality →Tender leakage measured across SKU × lane × month, lanes scored by tail-risk — so spot-freight inflation is caught before invoices reconcile.
Read the tender-leakage paper →MESO playbooks for API procurement — simultaneous-equivalent offers framed against a qualified BATNA when supplier qualification is the binding constraint.
Read the MESO playbook →Terminus M4 is being validated under Synlogica's Integrated Management System (IMS) — ISO 9001 (quality), ISO 27001 (information security) and ISO 42001 (AI management) — on top of its GAMP 5 Category 4 lifecycle.
IMS certification target Q4 2026GRIP · the engine under the hood
Every recommendation carries the exact rule and model versions that produced it — reproducible bit-for-bit, and packaged so a reviewer or inspector can reconstruct what was decided, on what evidence, under which ruleset.
Deterministic, versioned thresholds and policies — approved limits, storage classes, quality-agreement clauses. Tenant-configurable, never silently changed.
Arrhenius kinetics, Bayesian elasticity, Monte-Carlo coverage — each with a stated confidence interval and the data-quality flags behind it.
Maps the analysis to an advisory recommendation — and hands it, with its evidence, to the authorized human who makes the binding call.
Inside Terminus · illustrative UI
An illustrative preview on synthetic data — showing how a sealed decision package and the Decision Hub actually look. Not a customer screenshot.
Illustrative interface on synthetic data — representative of Synlogica Terminus, not a specific customer deployment.
M4 · Quality Analysis
Step 1 · Evidence in
Drop the logger export and the stability documents. Terminus parses and screens every input — gaps, outliers, out-of-range readings — before a single number is computed (ALCOA+).
Step 2 · Kinetics
The temperature profile becomes MKT and Arrhenius k(T) — how much of the product's stability budget the excursion actually consumed, not just “how long it was warm”.
Step 3 · Probability
10,000 Monte-Carlo runs turn the kinetics into a probability with a stated confidence interval — “96% within the stability budget”, not a point estimate or a reviewer's gut feel.
Step 4 · Decision
Terminus states an advisory verdict with the full sealed trail — inputs, model versions, ruleset. The authorized human (QA / QP) makes the binding release call, recorded in your QMS.
RELEASE · 2026-06-18 Illustrative interface on synthetic data — the same worked batch (PL-2026-0417-C) at every step.
The same chain of custody runs for freight (Terminus M1) and supplier negotiation (Terminus M5) — three modules, one governed platform.
Illustrative figures for this workflow. For the sourced, measured comparison and method, see “Measured, not estimated” below. ↓
Read how the engine works →Measured, not estimated
The numbers below come from comparing governed decisioning with typical manual QA excursion workflows. The full method and anonymized pilot data are in the white paper — so you can audit the claim, not just take our word for it.
100% of evidence traced, input-by-input
Read the method + anonymized pilot data →Compare
| Typical manual workflow | Terminus | |
|---|---|---|
| Excursion assessment | Spreadsheets, reviewer-dependent | Versioned model + stated uncertainty |
| Decision lead-time | ~2 days | under 60 seconds |
| Audit defensibility | Reconstructed by hand | Decision package, reproducible |
| Method consistency | Varies by reviewer | Same engine, tenant-configured rules |
| Validation posture | Ad-hoc | GAMP 5 Cat 4 — qualified once |
Compliance & trust
Every decision is built to answer the questions an inspector will really ask. Hosted in the EU. Backed by evidence. Mapped to the standards you already work with.
Talk to us
Working session
We start with a working review of a sample decision package and its validation assumptions against your intended use — to agree the configuration baseline and the shared-responsibility split before anything goes live.
Bring a real workflow — a recent excursion, an upcoming tender, or a supplier price claim — and we walk it through Terminus together.
FAQ
No — it is advisory by design. Synlogica Terminus produces a decision-support recommendation (RELEASE / CONDITIONAL / HOLD) with full traceability; the formal batch disposition is made and signed by the Authorized Quality Reviewer (QA / QP), recorded in your QMS.
Terminus is positioned as configurable software (GAMP 5 Category 4), validated through a risk-based lifecycle — intended use, configuration, requirements traceability, IQ/OQ/PQ, data-integrity and audit-trail verification. Synlogica provides product and supplier evidence; the regulated user owns the validated state for their intended use.
Records are versioned; controlled edits require a reason and an e-signature gate (password re-authentication); the audit trail is immutable, attributable and exportable as CSV. Sensor data is screened for outliers, out-of-range readings and gaps before use (ALCOA+).
Three domains on one engine: Quality (temperature excursions and shelf-life impact), Transport (tender leakage and freight risk), and Negotiation (MESO framing for API procurement).
Cloud-native — Kubernetes on AWS EC2 with AWS S3 for durable, WORM-capable storage, under the AWS shared-responsibility model. EU data residency or an alternative cloud / on-prem topology is configurable per your hosting policy.
No. LLMs assist only with extraction and structuring of unstructured inputs (e.g. parsing a stability PDF or an unknown logger layout), always confirmed before use. The recommendation itself comes from deterministic, version-pinned rules and the kinetic / probabilistic models.
A full trail from logger file to client report — for quality, transport and negotiation.
