Part 2 explained the architectural models behind Insulin on Board.
Part 3 moves from theory to real-world systems.
The table below distils the main IOB models, configuration options (Aggressive / Balanced / Protective), and their trade-offs.
Use this as a reference point before looking at specific apps, pumps, and hybrid closed-loop systems.
| Model – IOB deducted from corrections – Settings | Correction Behaviour | Insulin Stacking Protection | Exercise Hypo Awareness |
|---|---|---|---|
| Linear/Curvilinear decay – No IOB – Aggressive AIT 2–3h | Very Fast ↑↑ | None ✕ | Very Low ↓ |
| Linear/Curvilinear decay – No IOB – Balanced AIT 3.5–4.5h | Very Fast ↑↑ | None ✕ | Moderate ↔ |
| Linear/Curvilinear decay – No IOB – Protective AIT 5–6h | Very Fast ↑↑ | None ✕ | High ↑ |
| Linear/Curvilinear decay – Correction-Only – Aggressive AIT 2–3h | Very Fast ↑↑ | Partial | Low ↓ |
| Linear/Curvilinear decay – Correction-Only – Balanced AIT 3.5–4.5h | Fast ↑ | Moderate | Moderate ↔ |
| Linear/Curvilinear decay – Correction-Only – Protective AIT 5–6h | Constrained ↔ | Strong | High ↑ |
| Linear/Curvilinear decay – Meal + Correction – Aggressive AIT 2–3h | Constrained ↔ | Moderate | Low ↓ |
| Linear/Curvilinear decay – Meal + Correction – Balanced AIT 3.5–4.5h | Slow ↓ | Strong | Moderate–High ↑ |
| Linear/Curvilinear decay – Meal + Correction – Protective AIT 5–6h | Very Slow ↓↓ | Very Strong | High ↑ |
| Expected Glucose After-Meal Behaviour Based – Aggressive AIT 3h + Offset 45min + Meal Rise 1.7 mmol/L | Fast ↑ | Moderate ✓ | Limited (Units Hidden) |
| Expected Glucose After-Meal Behaviour Based – Balanced AIT 4h + Offset 60min + Meal Rise 3.0 mmol/L | Constrained ↔ | Strong ✓ | Limited (Units Hidden) |
| Expected Glucose After-Meal Behaviour Based – Protective AIT 5h + Offset 120min + Meal Rise 5.0 mmol/L | Slow ↓ | Very Strong | Limited (Units Hidden) |
| AID Systems: User and Algorithm Separate IOB – Aggressive User AIT 2–2.5h | Fast ↑ | Partial | Low ↓ |
| AID Systems: User and Algorithm Separate IOB – Balanced Default AIT 3–4h | Constrained ↔ | Moderate | Moderate ↔ |
| AID Systems: User and Algorithm Separate IOB – Protective User AIT 4.5–6h | Very Slow ↓↓ | Very Strong | Moderate ↑ |
Apps
In linear or curvilinear models, AIT determines both correction behaviour and how clearly the insulin tail is recognised. . Shorten it and corrections speed up — but exercise hypo risk is more likely to be underestimated. Lengthen it and stacking protection improves — but corrections can feel restricted.
In Expected Glucose Behaviour models, correction timing is governed by an expected glucose trajectory (using offset time and meal-rise settings).
However, true IOB is often hidden, limiting insulin-load awareness for exercise planning.
Refer back to the summary table above to see how each app can be configured into Aggressive (lots of corrections more hypo risk), Balanced or Protective (less corrections and less hypo risk) behaviour.
| App | IOB Architecture | What It Optimises For | Structural Limitation | Best Used When… |
|---|---|---|---|---|
| mySugr Accu-Chek Bolus Advisor | Expected glucose after-meal behaviour model (offset + meal rise + acting time). Correction IOB deducted for corrections. | Trajectory-aware correction timing and stacking governance | Does not display live insulin exposure (units on board hidden) | Perfect for personalised after-meal glucose control. |
| Diabetes:M (1 & 3) mylife App (3) InPen (3) RapidCalc (3) | Linear/curvilinear decay model. (1) No IOB recognised for corrections (2) Correction IOB deducted for corrections (3) Meal & Correction IOB deducted for corrections | Visible active insulin and straightforward stacking control | No correction IOB only for corrections; behaviour highly dependent on AIT setting and whether active insulin is enabled | You want to see total active insulin and are comfortable managing correction–exposure trade-offs. |
Common workarounds
- Shortening AIT to 2 hours to “unlock” corrections when the system feels too Protective. This allows more aggressive corrections but can increase correction stacking and give a false sense of security when being active or exercising in the 2-4 hours after a bolus.
- Repeated small corrections in the first 1–3 hours after eating when glucose is still within expected meal behaviour. This is a common pathway to “correction hypos”, or otherwise known as rage bolusing!
- When AIT is set to longer durations (typically 4–6 hours) and meal and correction insulin are pooled together as IOB, the bolus calculator can heavily restrict further corrections. If carbohydrate intake is underestimated, the system may block additional insulin even though glucose levels continue to rise. As a result, some users adopt workarounds such as entering “ghost carbohydrates”, giving pen injections outside the pump, or setting the correction factor unrealistically strong to force larger corrections. These strategies may restore the ability to correct highs, but they also undermine the accuracy of IOB tracking and can introduce unintended risks, particularly delayed insulin stacking and hypoglycaemia if the underlying system behaviour is not well understood.
- Using a behaviour-based app for exercise planning without a visible IOB number. If insulin exposure is hidden, the user may underestimate risk and over-trust a “safe-looking” screen.
The practical goal is not to eliminate trade-offs. It is to recognise which trade-off you are choosing, and then manage the predictable risks deliberately rather than reactively.
Standard Pumps (Without AID)
Standard pumps typically use a linear or curvilinear decay model. Meal and correction insulin are usually pooled, so the pump cannot know why insulin was given — only that it exists.
- Shorter AIT: faster corrections
- Longer AIT: slower corrections
This means underestimated meals can feel like “blocked corrections” in Protective configurations, while Aggressive configurations increase the risk of stacking. Exercise safety is heavily dependent on how realistically AIT reflects insulin duration.
Refer back to the summary table above to position your pump configuration in Aggressive, Balanced or Protective territory.
| Manufacturer | Systems | Bolus Calculator IOB Model | User Settings |
|---|---|---|---|
| Medtronic – Bolus Wizard | 530G → 780G (bolus calculator) | Curvilinear decay (meal + correction pooled and deducted from corrections) | AIT/DIA user adjustable (2–8 hours in 15 minute increments) |
| Insulet – Omnipod | DASH Omnipod 5 (Manual Mode) | Linear decay (meal + correction pooled and deducted from corrections) | AIT user adjustable (2–6 hours in 30 minute increments) |
| Tandem – t:slim X2 | Manual Mode | Linear decay (meal + correction pooled and deducted from corrections) | AIT user adjustable (2–8 hours in 15 minute increments) |
| Ypsomed – mylife YpsoPump | Without AID | Linear decay (meal + correction pooled and deducted from corrections) | AIT/DIA user adjustable (2–8 hours in 15 minute increments) |
Common workarounds (and why they can backfire)
- Shortening AIT to fix “blocked corrections” when meals are underestimated. This often converts a meal dosing problem into a stacking problem and increases late hypoglycaemia risk.
- Chasing highs with multiple corrections in the 1–3 hours after a meal. With pooled IOB, this can compound insulin exposure even if the pump appears to permit it.
- Using pump IOB as an “exercise green light” without considering meal boluses and recent corrections together. Total exposure matters, not just whether a correction was given.
AID / Hybrid Closed Loop Systems
AID systems add a second layer: a user-facing bolus calculator and an internal control algorithm. These layers may not use the same insulin accounting. In practice, displayed IOB may not reflect total insulin exposure generated by automation.
Refer back to the summary table above for Aggressive, Balanced, and Protective behaviour bands, but recognise that automation can moderate (or add to) insulin exposure in ways the user does not directly see.
| System | User-Facing Bolus IOB Model | User Settings Exposed | What The Algorithm Adds | Structural Limitation |
|---|---|---|---|---|
| Medtronic 670G / 780G | Bolus Wizard and SmartGuard Algorithm: Curvilinear decay (meal + correction IOB) | AIT/DIA adjustable for bolus calculator and algorithm correction behaviour (AIT 2–8h in 15 minute increments) | Automated basal adjustments and system-specific micro-corrections take into account total IOB. | When AIT is set very short (2 hours), the bolus calculator permits very aggressive correction dosing, which can improve post-meal glucose control. There is some protection against hypoglycaemia because the system can suspend or reduce basal insulin if glucose begins to fall. However, this protection is limited when bolus doses are large. Basal reduction cannot remove insulin that has already been delivered, so substantial insulin exposure may still occur. When AIT is at 2 hours, exercise-related hypoglycaemia risk can be high, and the displayed IOB will significantly underestimate the true physiological insulin exposure. If AIT is set longer (4 hours or more), both meal and correction insulin remain counted as IOB for longer. This slows the ability of the bolus calculator to issue further corrections when carbohydrates have been underestimated. In practice, this may require users to enter additional carbohydrates (“ghost carbs”) or use other workarounds to allow further insulin delivery when a meal has been under-bolused. As a result, many users choose the shortest AIT that does not produce excessive hypoglycaemia, often around 2–3 hours (15-minute segments). However, users must recognise that the displayed IOB will still underestimate total insulin exposure when considering exercise or activity-related hypoglycaemia risk. There is no single correct setting. The key is understanding how the model behaves and adjusting expectations and strategies accordingly. It is also important to recognise that displayed IOB does not fully represent total insulin exposure when automated insulin delivery features are active. |
| Omnipod 5 (Automated Mode) | Bolus calculator: linear decay (meal + correction IOB) | AIT adjustable for bolus calculator behaviour (AIT 2–6h in 30 minute increments) | Automation manages insulin delivery using a separate internal IOB pool. The exact method used to calculate this pool is not publicly defined. In some systems, negative IOB values can appear, which suggests basal reductions or suspensions are being subtracted from a pooled insulin estimate. What remains unclear is how this automated IOB interacts with user-delivered insulin. It is not always transparent whether meal boluses and user-entered correction doses are included in the same pool, or whether they are treated separately. It is also uncertain whether the automated system uses the same user-defined AIT for decay calculations or whether the algorithm applies its own internal insulin action curve that may be more physiologically modelled. As a result, the relationship between displayed user IOB and the algorithm’s internal insulin calculations can be difficult to interpret. | The Omnipod 5 system uses a relatively conservative control algorithm that primarily looks around 60 minutes ahead when making automated adjustments. Because of this short prediction horizon, insulin suspension often occurs earlier than in some other systems. In practice, this means a shorter user-defined AIT (typically around 2–2.5 hours) often allows more effective user-initiated corrections when required. When paired with a stronger correction factor, the user can address post-meal hyperglycaemia more effectively. Using a longer AIT often leads to prolonged hyperglycaemia because both meal and correction insulin remain counted as IOB for longer and the bolus calculator becomes slower to permit additional corrections. Many users therefore find that an AIT of around 2–2.5 hours, combined with a stronger correction factor and reverse correction disabled, allows more effective user corrections when required. |
| Tandem t:slim X2 with Control-IQ | Bolus calculator: linear decay (meal + correction IOB) | Fixed AIT 5 hours in Control-IQ | Control-IQ provides automated basal modulation and automated correction boluses based on the correction factor. User IOB from meal and correction boluses is incorporated using a fixed 5-hour insulin action curve. | Because IOB within the system can become both positive and negative relative to programmed basal, it is helpful for basal insulin to represent at least 50–60% of total daily insulin. This gives the algorithm enough range to increase or decrease insulin delivery effectively. Some users achieve this by using slightly weaker carbohydrate ratios so more insulin is delivered via automated basal adjustments rather than large bolus doses. A practical approach that often works well: Basal 50–60% of total daily insulin Weaker carbohydrate ratios Strong correction factors For exercise hypo risk prediction, the fixed 5-hour AIT is beneficial because displayed IOB is a good approximation of physiological insulin exposure. |
| CamAPS FX | Proprietary internal insulin action modelling (pharmacodynamic kinetics) | No meaningful user adjustment of insulin action model | The algorithm pairs carbohydrate entry and insulin delivery within its own internal model. | Highly specialised but opaque. The system uses a proprietary insulin action model and users cannot directly adjust insulin duration. Behaviour of the algorithm is therefore largely fixed and not transparent. The system includes a Boost mode, which strengthens the algorithm response without requiring additional carbohydrate entry. For many users this is preferable because it avoids manipulating carbohydrate entries or confusing data interpretation. |
Common workarounds (and why they can backfire)
- Trying to use AIT as the main control knob in AID. In many systems, AIT changes bolus calculator behaviour but does not fully control what the algorithm does in the background.
- Forcing more corrections by changing settings rather than changing the strategy. Ghost carbs, pen injections, or very strong correction factors.
- Assuming displayed IOB equals total insulin exposure. Automation can add insulin in the background, so exercise planning needs to consider recent automated delivery as well as manual boluses.
Summary
In Part 3 we moved from architecture to behaviour.
Across apps, standard pumps and AID systems, the pattern is consistent:
- Most systems pool meal and correction insulin together into a single decay model.
- AIT/DIA is the main behavioural control in non-AID systems.
- In AID systems, user-facing IOB (i.e what the user sees on the screen) does not always reflect total automated insulin exposure.
Short AIT increases correction responsiveness but weakens insulin tail recognition.
Long AIT improves stacking protection and exercise modelling, but can restrict corrections.
Blocked corrections, correction-induced hypos, and exercise lows are not random problems. They are predictable outcomes of where a system sits on the Aggressive–Balanced–Protective spectrum.
In reality,
The same setting is often used to control both correction behaviour and how realistically insulin exposure is represented.
Understanding these trade-offs reveals the central design limitation of current IOB systems which leads directly to the question:
“Could IOB be designed differently?“
The Future?
Part 4 explores practical options for solving this correction versus exercise dilemma.
Instead of simply shortening AIT to unlock corrections — and increasing late hypo risk — we will examine models that allow the best of both worlds.
The IOB Guide for T1D
- Hub: The Insulin On Board Guide for T1D
- Part 1 – The Insulin On Board–Physiology Mismatch
- Part 2 – Different Models For Calculating Insulin On Board
- Part 3 – Choosing a Device-Specific Insulin On Board Settings: What Are You Optimising For?
- Part 4 – The Future of calculating Insulin On Board: combining correction behaviour and exercise hypoglycaemia risk
- Part 5 – GNL Exercise Insulin on Board Calculator for T1D
- Part 6 – Reccomended Reading and Resources
