The Glucose Never Lies®
GNL Grace
The knowledge educator and six Explorers, one place
Grace is the evidence-grounded T1D knowledge educator. The six Explorers are the interactive tools that sit alongside her, live inside the GNL app. 1.5 million patient-days. 77 safety tests. Zero genuine failures. Trademarked, insured, compliance-documented.
- Use Grace
- The six Explorers
- What is Grace
- How she works
- Evidence base
- Safety
- IP and compliance
- Research
- Licensing
- Contact
Use Grace
Three ways to use Grace
Start anonymously on any page, register in the GNL app for more depth, or upgrade for research-grade access. Every tier runs on Opus 4.7, the top-spec Anthropic model. No hidden tiers, no shallow fallbacks. Login and registration happen in the GNL app.
Grace Basic
Free, anonymous
Try Grace on any page. 3 short evidence-backed answers per visit. Opus 4.7 quality, no signup needed.
- 3 questions per visit
- Short answers, GNL links
- No account needed
Available on every GNL page
Start here
Grace Pro
Free with your GNL account
30 consolidated answers per month. 2 to 3 paragraphs. Evidence grades A to D on every claim. Links to key papers. Login and register happen in the app.
- 30 questions per month
- Consolidated topic summaries
- ISPAD / ADA comparative view
- Free, no payment needed
Grace Max
£50/month + tax, personal use
Research depth. Manuscripts, algorithms, peer review, evidence synthesis. Up to 8,000 token answers, all 10 modes, extended thinking, file uploads.
- 35 questions per month
- All 10 research modes
- Extended thinking, file uploads
- Cancel any time
Organisational and commercial use
Manufacturers, charities, NGOs, public health bodies, insurance, reimbursement teams, clinical platforms. Grace deployed at your scope, with your branding, under a bespoke agreement. Robin Hood pricing: commercial contracts fund the evidence base that keeps personal access affordable.
The Explorers
Six Explorers, one GNL app
Every Explorer is a deterministic educational tool built from 1.5 million patient-days of real-world data. Tap a pill below to read what each Explorer does, who it’s for, the physiology it applies, and the inputs and outputs. AID Algorithm Explorer and Exercise Planning each open nested pills for system-specific or phase-specific detail. Open the tool itself in the GNL app.
What it does
In the presence of insulin, exercise can lower glucose up to eight times faster than a correction dose. This Explorer calculates the expected drop across three short activity windows so you can see how sensitive you are to active insulin before you move.
Who it’s for
Anyone on MDI, pump or an AID system who wants to understand activity impact when insulin is still active. Useful for planning post-meal walks, activity breaks, or pre-bed movement.
Evidence basis
Insulin sensitivity modulates the glucose-lowering effect of activity. Active insulin is tracked across recent dose history. Trend is applied as a final modifier. Scaling derived from population-level U/kg exposure data: 0.025 to 0.20 mmol/L per minute depending on load.
Inputs
- Body weight
- Glucose units (mmol/L or mg/dL)
- Bolus doses (last 8 hours)
- Current glucose and trend
- Activity type, walking or moderate aerobic
Outputs
- Active insulin (units, U/kg)
- Effective lowering rate
- Estimated drop at 10 minutes
- Estimated drop at 20 minutes
- Estimated drop at 30 minutes
Activity bands
What it does
Insulin on board is the most important, and most invisible, factor in exercise hypoglycaemia risk. This Explorer estimates how much carbohydrate you may need for a thirty-minute session based on your real IOB, body weight, and the type of activity planned.
Who it’s for
People on MDI, pump or AID planning structured exercise sessions who want to pre-empt hypos in the thirty-minute window.
Evidence basis
IOB drives hypo risk during activity. Carbohydrate need scales with U/kg exposure, body weight, exercise intensity (aerobic, mixed, anaerobic), and current glucose trend. Population-average requirements, not individual prescriptions.
Inputs
- Body weight
- Current glucose and trend
- Exercise type (aerobic, mixed, anaerobic)
- Bolus doses (last 8 hours)
- Carb source preference (optional)
Outputs
- Total active insulin
- IOB per kg body weight
- Carbohydrate estimate for 30 minutes
- Food equivalents (tabs, juice, sweets)
What it does
Four major AID algorithms make decisions differently. This Explorer models all four across responsiveness levels 1 to 5, showing algorithm-calculated basal rates, correction factors, insulin-to-carb ratios, and target glucose by time block. Two methods: carb-informed when daily carbs are entered, rule-based when only TDD is available.
Who it’s for
People on AID systems wanting to understand algorithm behaviour, setting trade-offs, and the difference between a more and less aggressive responsiveness level.
Evidence basis
Control-IQ and CamAPS FX maximise algorithm-delivered insulin with high basal-percentage strategies. Omnipod 5 and MiniMed 780G use shorter active insulin times for earlier corrections at the cost of less IOB visibility. Outputs derived from published correction factor, basal percentage, and carb ratio ranges per manufacturer documentation.
Inputs
- Age
- Body weight
- Total daily dose (units)
- Responsiveness level (1 to 5)
- Daily carbohydrate intake (optional)
Outputs
- Target glucose per time block
- Basal rate per time block
- Carb ratio per time block
- Correction factor per time block
The four algorithms, how they differ
What it does
The rule of 15 is a starting point, not an answer. This Explorer has two tabs: weight-scaled carbohydrate for hypo treatment or prevention, and correction protocols for high glucose with elevated ketones, based on your actual TDD and body weight, not population averages.
Who it’s for
People on MDI, pump, or any AID system managing acute lows and highs, plus pre-exercise hypo prevention planning. Tab 2 is tuned for sick-day guidance.
Evidence basis
Tab 1: rule-of-15 baseline scaled by body weight and CGM trend. Tab 2: correction protocols for high glucose with ketones follow sick-day guidance, 10% of TDD for mild ketones, 20% of TDD for significant elevation.
Inputs
- Body weight
- Total daily dose
- Glucose units
- Current glucose
- CGM trend
- Food preference (Tab 1) / ketone level + duration (Tab 2)
Outputs
- Carbohydrate grams + food equivalents (Tab 1)
- Correction insulin dose (Tab 2)
- System-specific action steps
Two tabs
What it does
Builds a complete glucose management plan for any session. Insulin adjustment, carbohydrate strategy, recovery carbs, and overnight considerations for evening sessions. Adapts to MDI, pump, and every major AID system. Adaptation buttons for “went low during”, “went high during”, and similar real-world scenarios.
Who it’s for
People on MDI, pump and all AID systems planning structured exercise sessions and needing phase-specific guidance.
Evidence basis
Therapy-adapted guidance. IOB determines carb need. Insulin adjustment (basal percentage, bolus reduction, AID mode) varies by therapy type and responsiveness level. ISPAD 2022 overnight-hypo prevention tree applied for sessions after 4pm.
Inputs
- Therapy type (MDI, Pump, CamAPS FX, 780G, Control-IQ, Omnipod 5)
- Exercise type and intensity
- Duration (10 to 360 minutes)
- Time of day
- Body weight, hypo risk profile
- Bolus doses (last 9 hours)
Outputs
- IOB estimate at start
- Carb amounts by glucose band
- Mode activation and return guidance
- Overnight snack or basal adjustment
- Adaptation suggestions mid-session
Four phases
What it does
Alcohol suppresses hepatic glucose output, meaning your liver cannot rescue you from a hypo the way it normally would. This Explorer maps the average glucose risk window across the drinking period and the overnight hours that follow, with system-specific harm-reduction strategies.
Who it’s for
People on MDI, pump, or any AID system who want to understand hypo risk during and after drinking, and see how their therapy type changes the best insulin strategy.
Evidence basis
Alcohol suppresses hepatic glucose output roughly one hour per unit consumed. Carb-containing drinks cause an initial glucose rise then a delayed fall. AID systems benefit from ninety-minute pre-activation. Insulin reductions range from 10 to 50% of basal, depending on duration and therapy.
Inputs
- Glucose units
- Therapy type
- Drinking duration (30 min to 24 hours)
- Drinks consumed (type, quantity, ABV, carbs)
Outputs
- Drink summary
- Visual risk timeline (safety and warning zones)
- System-specific insulin guidance
- Before-bed and morning-after checklist
- Copy-able safety plan
Genesis
What is Grace, and why does she exist?
Grace was named on 6 April 2026. The name belongs to Grace Pemberton, John Pemberton’s daughter. It is not a coincidence. Everything GNL builds is ultimately built for the next generation, for young people growing up alongside this condition, and for everyone who lives with it or supports someone who does.
The problem Grace solves
T1D education is broken in a specific way. The evidence exists. The clinical guidelines exist. The research is out there, in journals, in conference proceedings, in data that has never been translated into plain language. What is missing is a knowledge partner who holds all of it, reports it honestly, and is built to the safety and compliance standards that make it trustworthy in an institutional context.
Grace fills that gap. She holds the GNL evidence wiki, every validated algorithm assumption, every Via Negativa finding, and all 33 population analyses, and she explains them in language that is clear, grounded, and honest about uncertainty.
What Grace is not
- A diagnostic tool
- A prescriptive clinical advisor, she never says “if your glucose is X, do Y”
- A medical device (Class I, II, or III)
- A replacement for a diabetes care team
- Trained on generic internet data, her knowledge base is curated, evidence-graded, and fully auditable
Grace is a knowledge educator. She helps people with T1D, their families, healthcare professionals, researchers, and organisations understand how T1D management works, what the real-world data shows, what the evidence supports, what remains uncertain, and where the science is still developing. Every output is framed as population-average evidence. Every response is built on the full GNL evidence base. No fabrication. No hallucination of clinical values. No individual prescription.
Architecture
How Grace works, four layers
Grace is not a general-purpose chatbot with a few diabetes facts added. She is built on four distinct layers of structured knowledge, each evidence-graded, each auditable.
Clinical guidelines, the floor
ISPAD 2024 (all 25 chapters) and ADA 2026 are loaded in full. These are the international standards that every recommendation is built from. When Grace cites a clinical position, it is traceable to these guidelines. She does not paraphrase or reconstruct from memory, the source documents are in her context.
GNL evidence wiki, 78 concept pages
Every core T1D topic has a dedicated concept page: sleep, bolus insulin, exercise, AID systems, CGM, hypoglycaemia, carbohydrate counting, alcohol, menstrual cycle, and more. Each page has an explicit evidence grade (A, B, or D). Grade B pages are grounded in real-world population data from the 1.5 million patient-day dataset. Grade D pages are clearly labelled as GNL-constructed logic pending upgrade. No topic is presented with more confidence than the evidence supports.
Real-world validation outputs, 33 analyses
Grace holds all 33 real-world population analyses, covering 1.5 million+ patient-days across a longitudinal real-world dataset (Cockpit 1.0/daily dataset, Syno by Syntactiq Dynamics FlexCo, syntactiq.ai). These are not clinical trial results. They are population-level patterns from everyday T1D management: who achieves Time in Range, what predicts hypoglycaemia, how AID systems perform in the real world, what sleep does to glucose control. Via Negativa findings, where the data contradicts received wisdom, are included, not suppressed.
Safety framework, hard-coded, cannot be overridden
Five safety rails are built into Grace at the system level. They cannot be disabled by a user, a partner, a clever prompt, or any roleplay framing. They were tested across 77 adversarial cases. They passed every time. See the Safety and Testing section for the full specification.
Knowledge base in numbers
| Component | Count | Notes |
|---|---|---|
| Core concept pages | 78 | All major T1D education topics |
| Source summaries | 475 | One per primary paper, verified citations |
| Evidence grade maps | 25 | Per topic, overall grade, per-study grades, gaps |
| Clinical guideline pages | 3 | ISPAD 2024 (25 chapters) + ADA 2026 |
| Real-world source pages | 24 | All completable Syno analyses loaded |
| Total evidence items referenced | 960+ | Citations across all source summaries, 233 in the citation database |
| Safety files | 4 | Contradictions, retracted papers, confidence map, audit trail |
Evidence base
Where the data comes from
Grace’s evidence base is built on a real-world longitudinal dataset covering more than 10,000 people with T1D over more than 10 years; more than 1.5 million patient-days assessed. 33 population analyses have been completed. This is not survey data. This is not a single clinical trial. It is everyday T1D management, captured, analysed, and reported honestly.
What Via Negativa means
Via Negativa, from the Latin for “the negative way”, is GNL’s core evidence discipline. Where the data shows something different from received wisdom, GNL reports it. Counterintuitive findings are not suppressed. They are the most valuable ones.
- −Step count: the 10,000-step target has no real-world basis in T1D glucose outcomes. The benefit plateau is at 4,000-5,000 steps.
- −Carbohydrate compensation on exercise days: the primary mechanism is insulin reduction (9.2%, p<0.001), not carbohydrate addition.
- −Exercise for 18-30 year olds: exercise is a weak predictor of TIR in this age group (r=0.024). Bolus frequency (r=0.31) and sleep (r=0.14) far outweigh it.
- −Weekend glucose management: weekends show 0.5pp higher TIR than weekdays, driven by lower carbohydrate intake and more bolusing, not less.
- −Sleep regularity: the benefit is age-specific. In 18-30 year olds, the effect is absent. In 31-40 year olds, it is the strongest modifiable predictor (+13.3pp TIR).
These are Via Negativa findings. Grace teaches them. They are in her knowledge base and she cites the evidence behind each one.
Grade B topics, confirmed by real-world population data
| Topic | Dataset | Key finding |
|---|---|---|
| Sleep and T1D | 611-user cohort | Sleep regularity: +10.8pp TIR, #1 modifiable predictor (age 31-40) |
| Hypoglycaemia | 373,746 patient-days | Over-correction drives next-day TBR 6.0% vs 1.3% |
| Bolus insulin | 228-466,631 patient-days | Optimal zone: 4-6 boluses/day (pen users); 2.5% TBR |
| Exercise and T1D | 373,737 patient-days | Aerobic: 3.6% TBR vs resistance: 3.1% TBR |
| Activity and movement | 526-user profiles | Step plateau at 4,000-5,000, not 10,000 |
| AID Systems | 839 users, 409,056 days | 6.2pp real-world TIR advantage (p<0.001) |
| CGM use | 652 users, 345,114 days | 89% achieve ≥80% coverage; 9pp TIR benefit |
| Alcohol and T1D | 881 users, 272,837 days | 38% relative increase in next-day TBR >5% |
| Menstrual cycle | 62 users, 2,045 days | 11.5pp TIR swing across cycle phases |
Safety and testing
Validated, tested, insured
Grace went through a 77-case adversarial safety and compliance test suite before any public exposure. The results are documented in compliance dossier v8.0 and available to prospective partners and institutional clients.
Test suite results, 9 April 2026
| Category | Cases | Result |
|---|---|---|
| Escalation (DKA, ketones, severe hypo) | 4 | 4/4 PASS |
| Jailbreak attempts | 12 | 12/12 PASS |
| Evidence grading | 10 | 10/10 PASS |
| Safety rails | 15 | 15/15 PASS |
| Algorithm knowledge | 18 | 18/18 PASS |
| Language and framing | 11 | 11/11 PASS |
| Explorer signposting, adversarial | 18 | 18/18 PASS |
| Total genuine failures | 77 cases | 0 failures |
6 automated flags during testing = test package defects triggering on Grace’s own refusal language. None represent Grace failures. Explorer signposting adversarial suite added 9 April 2026.
Five safety rails, hard-coded, cannot be overridden
No individual prescription
Grace never writes “if your glucose is X, do Y”, “adjust your dose when…”, or any language that constitutes a clinical instruction to a specific person. All outputs are population-average educational statements.
Population-average framing
Grace never uses “your personalised plan”, “tailored to you”, or any language implying individual prediction. She presents what real-world populations show, not what any given person will experience.
Safety escalation, overrides all other rules
Ketones ≥1.5 mmol/L, glucose ≤2.2 mmol/L, DKA symptoms, severe paediatric hypo, Grace directs immediately to 999/A&E. No educational hedging. No waiting. Tested against four real escalation scenarios including a child in severe hypo with panic framing. All four passed with exemplary responses.
No wiki fabrication
If the knowledge base is not loaded, Grace acknowledges the absence and does not reconstruct algorithm values from memory. She never invents a clinical number. If she cannot confirm a figure from her loaded context, she says so.
Jailbreak resistance
Prompt injection, roleplay framing, false authority claims, urgency pressure, all tested, all refused. 12/12 jailbreak attempts resisted across the full adversarial test suite.
Insurance
All GNL educational platform activity, including Grace, is covered by professional indemnity and public liability insurance at £2 million per claim. Insurance documentation is available to institutional partners and prospective licensees on request.
Intellectual property and compliance
Protected at every layer
GNL’s algorithm suite, evidence base, and Grace are protected by registered trademarks, copyright, server-side code architecture, and a full compliance documentation framework. Compliance dossier v8.0 is available to institutional partners, insurers, and prospective licensees.
Registered trademarks
| Mark | Registration | Classes | Status |
|---|---|---|---|
| The Glucose Never Lies® | UK00004267795 | 41, 44 (Education, Health) | Registered |
| The Glucose Never Lies® (second mark) | UK00004360249 | 41, 42 (Education, Software) | Accepted, awaiting Journal |
| GNL Grace | UK00004370619 | 9, 41, 42 (Software, Education, SaaS) | Filed 9 April 2026 |
| GNL Grace | EU EUTM | 9, 41, 42 | Filed 9 April 2026 |
| GNL Grace | USPTO (US) | 9, 41, 42 | Due by 9 October 2026 (Paris Convention) |
Code and algorithm protection
- All algorithm source code is server-side. No algorithm logic is present in any browser-facing file. The engine cannot be extracted from the front end.
- US Copyright filed, GNL Explorer Suite, Case 1-15136552281 (7 April 2026).
- Algorithm documentation: Appendix A of compliance dossier v8.0. Internal record only, not published.
- Version-controlled audit trail: Git history + dossier change log + JS version numbers provide a complete record of every algorithm version ever deployed.
Compliance documentation
- Compliance dossier v8.0, company structure, insurance, legal basis, GDPR, explorer compliance profiles, Grace architecture, full test results, IP schedule, governance framework.
- ICO registered data controller (payment confirmed December 2025). UK GDPR and Data Protection Act 2018 compliant.
- VAT registered, GB 516 3272 08 (effective 01 April 2026).
- Incorporated, The Glucose Never Lies Ltd (Companies House).
Research collaboration
GNL as a research partner
Grace is not just an educator. She is a research engine. She holds the full validated evidence base and can assist with literature synthesis, manuscript development, methodology questions, and evidence-grading for T1D research projects. The GNL research pipeline has identified 30 high-value queries that remain open for future investigation.
Via Negativa, licensing
Your organisation’s version of Grace
Via Negativa is GNL’s IP licensing vehicle. Organisations, manufacturers, charities, coaching practices, academic institutions, clinical teams, can licence their own version of Grace. Not a generic AI. A properly built, safety-tested, compliance-documented T1D knowledge educator, configured for their specific context, framing, and purpose.
GNL already works with major players in diabetes coaching, device manufacturing, and the charity sector. Their names are not published here, confidentiality is part of the offer. What can be said: the framework is proven, the compliance documentation is in place, and the licensing model is operational.
White-label Grace
A fully branded version of Grace, deployed under your organisation’s name. All GNL compliance documentation and safety rails included. Configured with your specific clinical context, product focus, and communication guidelines.
Co-branded educator
Grace deployed alongside your organisation’s brand, jointly attributed to GNL and your team. Suitable for partnerships where the GNL evidence base adds independent credibility to your offer.
Research licence
Access to Grace as a research assistant for clinical teams and academic institutions. Literature synthesis, evidence grading, manuscript support, and Via Negativa auditing, all within the GNL compliance framework.
Integrated API licence
Grace’s knowledge engine integrated into your platform, app, or service via the GNL API. Developed with Phillip Hayes (Technical Director). Suitable for manufacturers and digital health platforms.
What every licence includes
- Access to the full GNL evidence base (78 concept pages, 475 source summaries, 960+ graded evidence items, 24 real-world analyses)
- All five safety rails, hard-coded, cannot be removed
- Compliance documentation for your organisation’s regulatory and governance needs
- Ongoing updates as the GNL evidence base grows
- Company-specific configuration: context, framing, language, and direction
- Via Negativa integrity: the evidence is reported honestly, including what doesn’t work
Examples
What a licensed Grace looks like
These are examples of the kinds of deployments that are possible through the Via Negativa licensing framework.
GNL platform
The full GNL ecosystem
Grace and the six Explorers sit inside a complete T1D education platform: validated tools in the app, a clinical podcast, foundational education pages, and a growing body of content built from real-world evidence.
Get in touch
Licensing enquiries, research collaboration, institutional partnership
All go directly to John Pemberton. GNL Grace is available to GNL app subscribers from April 2026. Wider partner and institutional access from Summer 2026.
john@theglucoseneverlies.comOr visit theglucoseneverlies.com
How Grace works →