1. What this system is

CamAPS FX is a fully closed-loop AID system developed by CamDiabetes, a spin-out from the University of Cambridge. It runs on an Android phone and connects via Bluetooth to either a Ypsomed YpsoPump or a Tandem t:slim X2 pump. Unlike hybrid closed-loop systems that still require significant user input for meals, CamAPS FX is designed to function as a fully automated loop — managing glucose continuously, with the user providing context through settings and mode selection rather than moment-to-moment dosing decisions.

The algorithm at its core descends directly from the Cambridge Adaptive Algorithm — the same engine used in the landmark OPEN trial and in a large body of randomised controlled trial evidence in young children, pregnancy, and adults. It is among the most extensively studied AID algorithms in paediatric and obstetric populations.

What makes CamAPS FX distinctive

• Distributed algorithm design. Rather than a single parameter controlling responsiveness, multiple algorithm components shift together. There is no equivalent of a single “aggressiveness dial.”
• Personalised learning over time. The algorithm builds a model of the individual’s patterns — including meal responses, activity patterns, and circadian variation — and adjusts its insulin delivery accordingly. This adaptation happens continuously in the background.
• More physiological IOB tracking. CamAPS FX tracks insulin on board more physiologically than a simple linear decay model. This means pushing algorithm-delivered insulin higher allows the system to adapt more dynamically to real-time glucose changes — the same principle that underpins the Control-IQ optimisation strategy.
• Target glucose as the primary lever. The most direct way to shift the system’s overall behaviour is to change the target glucose. A lower target drives more aggressive delivery; a higher target is more permissive. This is the lever the GNL framework maps onto its five-level structure.
• Algorithm-delivered insulin as the optimisation strategy. CamAPS FX delivers up to 65% of total insulin algorithmically at the most responsive level. By anchoring basal at up to 65% of TDD, only 35% remains for meal boluses — carb ratios are set weaker to reflect this split. The remaining bolus insulin is divided by actual carb intake, giving the algorithm the most dynamic range to adapt.
• Carb estimation, not precise counting. CamAPS FX uses carb estimates to adjust meal-time delivery. The system is designed to tolerate a degree of carb estimation error — though accuracy still matters, particularly at the more aggressive end of the target range.

2. The optimisation strategy: increase algorithm-delivered insulin

CamAPS FX and Control-IQ share the same fundamental optimisation strategy. Both track insulin on board more physiologically than systems that rely on a short active insulin time to clear IOB from the display. The approach is to push algorithm-delivered insulin higher — anchoring basal at up to 65% of TDD — so that only 35% remains for meal boluses. Carb ratios are set weaker to reflect this split.

The primary lever in CamAPS FX is target glucose. A lower target means the system treats a wider glucose range as “above goal” and delivers more insulin, more frequently. A higher target allows glucose to run higher before the algorithm increases delivery. This is different from Medtronic 780G and Omnipod 5, which primarily use a shorter AIT to lower device-reported IOB and enable more frequent auto-corrections.

Why this matters for carb ratios

When up to 65% of insulin is delivered algorithmically through basal modulation, as little as 35% remains for bolus. The remaining bolus insulin is divided by actual carb intake — which is why Output 1 (carb-informed) in the explorer is particularly valuable for CamAPS FX. It calculates the carb ratio from the bolus fraction that remains after algorithm-delivered insulin is accounted for, giving the algorithm the best chance to adapt dynamically.

How this compares to the other strategy

Medtronic 780G and Omnipod 5 take a different approach — they lower IOB visibility by using a shorter AIT, which makes the algorithm assume insulin clears faster and enables earlier auto-corrections. The trade-off is that device-reported IOB is lower than physiological IOB, which creates a risk during and after exercise. CamAPS FX avoids this trade-off by tracking IOB more physiologically, though its IOB model is not published — making IOB somewhat more opaque than Control-IQ’s visible (but linear) 5h AIT model.

3. The five levels

The GNL explorer maps CamAPS FX settings onto five levels. Each level describes a distinct combination of target glucose, basal/bolus split, and correction factor rule. The optimisation strategy is the same at every level: increase algorithm-delivered insulin and set carb ratios weaker to reflect the split. What changes across levels is how aggressively this is applied.

Level 5 — Very high responsiveness (target 4.4 mmol/L / 80 mg/dL)

At level 5, the target glucose is set at 4.4 mmol/L — towards the lower end of the normal fasting range. The algorithm treats most glucose readings as “above target” and delivers insulin frequently and in larger increments. Basal is anchored at 65% of TDD, leaving only 35% for meal boluses — carb ratios are at their weakest, reflecting the maximum algorithm-delivered insulin split.

Because the algorithm is doing the most work at this level, it has the greatest dynamic range to adapt to real-time glucose changes. The correction factor rule (80 mmol/L / 1400 mg/dL) is the most aggressive. This level tends to be most appropriate when the priority is minimising time above range, where hypoglycaemia risk has been carefully assessed, and where the user has experience managing the system at this level. It is not typically a starting point.

Level 4 — High responsiveness (target 5.0 mmol/L / 90 mg/dL)

At level 4, the target is 5.0 mmol/L — a tighter target than most closed-loop defaults but with a small degree of tolerance above level 5. Basal is anchored at 60% of TDD, leaving 40% for bolus. The correction factor rule (85 mmol/L / 1500 mg/dL) is still aggressive but with slightly more margin than level 5.

This level is often appropriate for people who have experience with tighter targets and who are comfortable with the algorithm delivering the majority of insulin. It tends to produce a good balance between low time above range and manageable hypoglycaemia risk for many experienced users.

Level 3 — Medium responsiveness (target 5.5 mmol/L / 100 mg/dL)

Level 3 is the neutral reference point in the GNL framework. The target of 5.5 mmol/L is close to the default setting in many CamAPS FX clinical protocols. Basal is at 55% of TDD, leaving 45% for bolus. The correction factor rule (90 mmol/L / 1600 mg/dL) represents the standard reference point.

Level 3 is often the starting point recommended in clinical guidance and is appropriate for people who are new to the system, those stepping back from a more aggressive level during an unsettled period, or those who want a stable baseline from which to explore.

Level 2 — Low responsiveness (target 6.0 mmol/L / 110 mg/dL)

At level 2, the target of 6.0 mmol/L gives the algorithm more tolerance above the normal range before it increases delivery. Basal is at 50% of TDD, leaving 50% for bolus — the user is taking more responsibility for insulin delivery through meal boluses rather than relying on algorithm-delivered insulin.

The correction factor rule (100 mmol/L / 1800 mg/dL) is more conservative. Level 2 may be appropriate during periods of high activity, illness recovery, or when establishing confidence with a new algorithm before tightening targets.

Level 1 — Very low responsiveness (target 7.0 mmol/L / 130 mg/dL)

Level 1 sets the target at 7.0 mmol/L — a deliberately permissive posture. Basal is at 45% of TDD, leaving 55% for bolus — the algorithm is doing the least work, and the user carries the most responsibility through meal boluses. The correction factor rule (110 mmol/L / 2000 mg/dL) is the most conservative.

This level tends to be appropriate when hypoglycaemia risk is the primary concern, when activity levels are unpredictable and high, during illness or surgical recovery, or as a transitional setting during initial system use. Many people find that this level significantly reduces hypoglycaemia frequency at the cost of higher average glucose and more time above range.

4. How the shared engine works

The GNL Multi-System Explorer uses a shared physiological engine to produce exploratory output ranges across all four AID systems it covers. Understanding how this engine works helps interpret the numbers it generates for CamAPS FX.

Sensitivity class

The engine first classifies insulin sensitivity using total daily dose (TDD) divided by body weight in kg, expressed as units per kilogram per day (U/kg/day). This ratio is a well-established proxy for overall insulin sensitivity — a lower ratio tends to indicate higher sensitivity, and a higher ratio tends to indicate lower sensitivity. It is used as an anchor for the correction factor and carb ratio estimates that follow.

As with all population-level rules, this is a starting estimate. Individual sensitivity varies considerably — across time, with illness, activity, hormonal cycles, and other factors. The sensitivity class is not a fixed label; it is a physiological reference point for generating an initial range.

Correction factor rule

The engine derives a base correction factor (CF) using a CF rule divided by TDD. This rule — a form of the widely used “1500 rule” or “1800 rule” depending on units — gives an estimate of how many mmol/L (or mg/dL) one unit of insulin is likely to lower glucose on average. The five levels in the CamAPS FX framework each have a different CF rule value (80 to 110 in mmol/L terms, 1400 to 2000 in mg/dL terms), reflecting the algorithm’s effective correction behaviour at each level of engagement.

These CF values are used in the engine’s calculations to generate exploratory ranges. They are not settings entered into CamAPS FX — the system calculates its own effective corrections internally.

Basal and bolus distribution across time blocks

The engine distributes estimated insulin delivery across time blocks representing morning, afternoon, evening, and overnight periods. It uses population-level anchors for basal fraction (the proportion of TDD delivered as background insulin) and bolus fraction (the proportion associated with meals and corrections).

Carb ratio logic

The engine uses two carb ratio approaches depending on the data available. Where the user provides carb intake information, it applies a carb-informed ratio — an estimate based on how much insulin is typically needed per gram of carbohydrate, derived from TDD and estimated carb consumption. Where carb data is not available, it falls back to a rule-based estimate anchored to TDD and sensitivity class.

For CamAPS FX, the carb ratio in the engine’s output should be understood as an exploratory reference range. CamAPS FX uses carb announcements as inputs, not as fixed bolus calculations — the algorithm integrates the carb estimate with real-time glucose data and its learned model to determine actual delivery. The engine’s carb ratio estimate reflects average expected need across the population distribution for a given sensitivity class, not a setting to programme into the device.

5. How to use the explorer output

The AID system explorer generates a set of exploratory output ranges for CamAPS FX based on the values you enter. Here is how to interpret those numbers with appropriate context.

What the output is

The numbers are directional estimates — a starting point for conversation, not a prescription. They describe where the physiological engine calculates average responses to sit, given a particular combination of TDD, weight, level, and entered settings. They are generated from population-level rules, not from your individual glucose data.

The output is most useful as a frame of reference: does the correction factor the engine suggests sit in a plausible range given your experience? Does the carb ratio estimate align with what you observe on your CGM after meals? When the output diverges significantly from what you observe in practice, that divergence is itself informative — it may point to individual variation worth exploring with your care team.

What the output is not

• It is not a recommendation to change your settings.
• It is not a calculation of your personal correction factor or carb ratio.
• It is not derived from your CGM data or your glucose history.
• It does not account for the CamAPS FX algorithm’s learned model of your individual patterns.
• It does not reflect real-time changes in sensitivity due to illness, hormones, activity, or other factors.

How to use it well

Many people find the explorer most useful when they bring the output to a review with their diabetes care team. The ranges can prompt useful questions: “The engine suggests my correction factor at level 3 would sit around X — does that match what my CGM data shows after corrections?” That kind of question — anchored in both the explorer output and real glucose data — tends to generate more useful conversations than either source alone.

It is also worth exploring how the output changes as you move between levels. The difference in correction factor between level 3 and level 5, for example, illustrates the mechanism behind why a lower target tends to produce more aggressive delivery — even if the exact numbers in your individual case will differ from the engine’s estimates.