AID system guide
The MiniMed 780G: how it works and what shapes its behaviour
Understanding Active Insulin Time, the five responsiveness levels, and what the GNL explorer output can help you explore about this fully closed-loop system.
What this system is
The Medtronic MiniMed 780G is a fully closed-loop automated insulin delivery system. It combines the MiniMed 780G pump with a Medtronic CGM sensor — the Guardian 4 — and uses the SmartGuard algorithm to make autonomous insulin delivery decisions in real time.
Unlike earlier hybrid closed-loop systems, the 780G is designed to manage glucose without requiring the user to input anything between meals. It adjusts basal rates continuously, suspends delivery to prevent hypoglycaemia, and delivers automatic correction boluses every five minutes when glucose is trending upward.
What makes the 780G distinctive among AID systems is its configuration around a single dominant variable: Active Insulin Time (AIT). Almost everything about how the system behaves — how aggressively it delivers corrections, how low it tries to keep glucose, how much tolerance it has for rising trends — flows from this one setting.
The key characteristics of the 780G
- Fully closed loop. The system does not need meal pre-announcements to function, though consistent meal bolusing tends to improve outcomes.
- Automatic correction boluses every 5 minutes. When glucose trends upward and insulin-on-board is low, the system delivers small corrections automatically without user input.
- SmartGuard hypoglycaemia prevention. The algorithm predicts and acts on falling glucose trends before hypoglycaemia occurs.
- AIT as the primary configuration lever. Shorter AIT makes the system more aggressive; longer AIT makes it more conservative. The target glucose also changes with AIT level.
- Variable glucose target. At the two most aggressive AIT settings (levels 4 and 5), the system targets 5.5 mmol/L (100 mg/dL). At more conservative settings, the target rises to 6.1 or 6.7 mmol/L.
This is a system that rewards consistency — consistent meal bolusing, consistent timing, and a willingness to let the algorithm work with less manual override. Many people find its behaviour changes noticeably as AIT is adjusted.
The primary lever: Active Insulin Time (AIT)
AIT is the single most important configuration variable in the 780G. Understanding what it does mechanistically helps make sense of everything else.
AIT tells the algorithm how long it should consider insulin to remain active after a bolus is delivered. This directly controls how the system calculates insulin-on-board (IOB) — the estimated amount of insulin still working in the body at any given moment.
The mechanism in plain terms: if the system believes insulin clears quickly (short AIT), it calculates a lower IOB. Lower IOB means fewer constraints on delivering new corrections. So a shorter AIT produces more frequent automatic corrections and more aggressive glucose management.
At AIT 2 hours (level 5), the system is at its most aggressive. It assumes insulin clears in two hours, so it recalculates IOB rapidly and is willing to deliver frequent corrections. The glucose target is at its lowest: 5.5 mmol/L / 100 mg/dL.
At AIT 3 hours (level 1), the system is most conservative. It assumes insulin lingers for longer, carries a higher IOB estimate, and holds back from delivering corrections until it is more certain they are needed. The glucose target rises to 6.7 mmol/L / 120 mg/dL.
This is not simply a “faster or slower” dial. AIT affects the target glucose, the frequency of automatic corrections, basal modulation activity, and what level of meal bolusing accuracy the system needs to function well. Shorter AIT settings demand more consistent meal management; they do not forgive missed or inconsistent boluses easily.
Many people and clinicians find that moving AIT to shorter settings produces a visible change in 24-hour glucose traces — tighter overnight control, more active suppression of post-meal rises, and a system that is harder to override without consequence. Whether that suits a particular individual’s lifestyle and management patterns is worth exploring carefully.
The five AIT levels
The 780G operates across five levels of responsiveness, each anchored to a specific AIT. The levels are not arbitrary — they represent a coordinated shift in how the system calculates corrections, what glucose target it pursues, and what it expects from the user.
| Level | Label | AIT | Target (mmol/L) | Target (mg/dL) | Auto-corrections | Basal modulation | Meal accuracy required | CF rule (mmol) | CF rule (mg/dL) |
|---|---|---|---|---|---|---|---|---|---|
| 5 | Very high | 2h | 5.5 | 100 | Very high | Very active | Very high — highly consistent bolusing required | 90 | 1600 |
| 4 | High | 2h 15 | 5.5 | 100 | High | Active | High — consistent bolusing expected | 95 | 1700 |
| 3 | Medium | 2h 30 | 5.5 | 100 | Moderate | Balanced | Moderate — standard meal bolusing | 100 | 1800 |
| 2 | Low | 2h 45 | 6.1 | 110 | Low | Moderate | Low — inconsistent bolusing tolerated | 105 | 1900 |
| 1 | Very low | 3h | 6.7 | 120 | Minimal | Conservative | Very low — minimal meal bolusing expected | 110 | 2000 |
Level 5 — Very high responsiveness (AIT 2h)
At level 5, the system assumes insulin is fully cleared within two hours. This produces the lowest possible IOB estimate after a bolus, allowing the algorithm to deliver frequent automatic corrections with minimal restraint. The glucose target is 5.5 mmol/L (100 mg/dL) — the tightest available.
This level is behaviourally demanding. The system expects boluses to be delivered consistently before meals and does not have much tolerance for omitted or significantly late doses. People who use this level well tend to describe it as a partnership — the algorithm does its part, but it needs accurate meal input to avoid stacking corrections on top of unannounced carbohydrate.
Many experienced 780G users find this level produces the best time-in-range results when meal management is consistent. It is worth exploring with CGM data and care team support rather than jumping to immediately.
Level 4 — High responsiveness (AIT 2h 15)
A marginal step back from level 5, with a slightly longer AIT and a fractionally more conservative correction factor. The glucose target remains at 5.5 mmol/L. The difference in day-to-day behaviour is subtle for most people — basal modulation is still active, and automatic corrections remain frequent. This is often where people transitioning toward more responsive behaviour start, before assessing whether level 5 suits their routine.
Level 3 — Medium responsiveness (AIT 2h 30)
The midpoint of the system. Basal modulation is described as balanced — the algorithm is active but not at its most assertive. Automatic corrections are delivered at a moderate rate. The glucose target remains 5.5 mmol/L, which means the system is still pursuing the same endpoint — it is just using somewhat more conservative IOB accounting to get there. This tends to be the starting configuration for many new 780G users, and a stable operating point for those with consistent but not highly structured meal management.
Level 2 — Low responsiveness (AIT 2h 45)
At level 2, the glucose target rises to 6.1 mmol/L (110 mg/dL). The algorithm is more tolerant of higher glucose before initiating corrections, and its IOB calculations are more conservative. This level tolerates irregular or inconsistent meal bolusing better than levels 3–5 — partly because it is not working as hard to maintain a tight target. People transitioning to AID from manual injection therapy, or those managing high variability in meal timing, may find this a more comfortable starting point.
Level 1 — Very low responsiveness (AIT 3h)
The most conservative level. AIT of three hours means the system carries a higher IOB estimate for longer after every dose, making it markedly reluctant to deliver automatic corrections. The glucose target rises to 6.7 mmol/L (120 mg/dL). Basal modulation is conservative and automatic corrections are minimal. This level is not designed for optimised glucose management — it is more relevant during periods of instability, illness, unusual activity, or when the user wants to minimise algorithm activity while they establish CGM data patterns. It may also suit people in early AID onboarding who need time to build confidence in the system.
On target glucose and what it means behaviourally: the 780G does not simply aim for a target — it manages around it. At levels 3–5, where the target is 5.5 mmol/L, many people find the system keeps overnight glucose noticeably lower than on manual or hybrid closed-loop therapy. At level 1, the system is essentially optimising for 6.7 mmol/L and will not act aggressively until glucose is clearly above that. This has practical implications for time-in-range, and is worth exploring against individual CGM data rather than assuming one level suits everyone.
How the GNL shared explorer engine works
The GNL Multi-System Explorer uses a shared physiological engine across all four AID systems it covers — including the 780G. Understanding what the engine is doing helps put the output in context.
Sensitivity class: TDD divided by weight
The engine first classifies insulin sensitivity using total daily dose (TDD) divided by body weight in kilograms. This produces a units-per-kilogram-per-day figure, which is then mapped to one of five sensitivity bands: very high, high, moderate, low, and very low. This classification determines which correction factor and carbohydrate ratio ranges are applied.
Higher sensitivity (lower TDD per kg) tends to produce stronger correction factors and higher carbohydrate ratios — meaning a given unit of insulin is expected to move glucose further and cover more carbohydrate. Lower sensitivity (higher TDD per kg) produces weaker correction factors and lower carbohydrate ratios, reflecting that more insulin is needed to produce the same effect.
Correction factor: the CF rule divided by TDD
Correction factors are derived by dividing a rule constant by TDD. The rule constant varies by AIT level — at level 5 (most aggressive), the 780G uses the 90 rule in mmol/L terms (1600 rule in mg/dL); at level 1 (most conservative), it uses the 110 rule (2000 rule). This means the explorer produces a different correction factor estimate for each level, with the most responsive level giving the strongest correction and the most conservative giving the weakest. These are directional — they describe the tendency, not the precise individual value.
Basal and bolus distribution across time blocks
The engine distributes basal insulin across time blocks using population-level circadian patterns. These patterns describe how basal requirements tend to shift across the 24-hour day on average — typically with higher requirements around dawn and lower requirements through the middle of the day. Bolus insulin is then the complement: total daily dose minus total basal.
For the 780G, AIT level influences the overall balance. More responsive levels tend to be associated with higher effective basal modulation — the algorithm is more active in suppressing and augmenting basal across the day. Less responsive levels produce a flatter, less dynamic basal pattern from the algorithm’s perspective.
Carbohydrate ratio logic
Carbohydrate ratios are estimated using the sensitivity band alongside total bolus insulin. The explorer uses the carbohydrate-to-insulin ratio implied by the bolus dose and entered daily carbohydrate (Output 1) or a rule-based estimate from sensitivity class alone (Output 2). At higher AIT levels, the expectation of consistent meal bolusing is built into the directional guidance — the system is working harder to correct post-meal rises, so it is more sensitive to inaccurate carbohydrate inputs.
AIT as a system context label in the 780G output
For the 780G specifically, AIT is displayed in the explorer output as a system context label alongside the standard results table. It is not a separate calculation input — the user selects a responsiveness level (1–5) and the AIT is a property of that level. The explorer shows users which AIT the selected level corresponds to, and what that implies for correction factor, target glucose, and expected meal management requirements.
What the engine is not doing: it is not simulating the 780G algorithm. It is using population-level physiological rules to describe the directional tendency at each level of responsiveness. The output is a starting-point framework for discussion, not a prediction of individual performance.
How to use the explorer output
The explorer generates a table of exploratory values — correction factors, carbohydrate ratios, and basal estimates — for the selected 780G responsiveness level. Here is how to read that output accurately.
Direction, not destination
The numbers in the output describe the direction of change, not a destination to programme directly into a pump. If the output shows a correction factor of 2.8 mmol/L and the current setting is 3.5 mmol/L, the signal is that more responsive management at this level tends to use a stronger correction — not that 2.8 is correct for this individual.
Many people find this framing useful: the output tells them whether their current settings are aligned with, above, or below what tends to be associated with the selected level on average. The gap between current settings and the output range is the conversation starter.
Ranges reflect genuine uncertainty
The explorer outputs are presented as ranges rather than single numbers. This is intentional. Population-level patterns cluster around a mean, but real-world variation is wide — commonly 30 to 40% of people sit close to the central estimate, and the rest are distributed across a broad range. A single-number output would imply a precision that the engine does not have.
Using the explorer well means holding the range in mind: which part of the range does current CGM data suggest this individual sits in? That question is far more productive than asking whether the specific midpoint is right.
CGM data is the real-world check
The GNL explorer is a pre-clinical educational tool. The only way to know whether a setting is working is to observe glucose behaviour over time on a CGM. Post-meal traces, overnight patterns, and response to corrections all provide feedback that no calculator or explorer can replicate. The output is a hypothesis; CGM data is the test.
This is especially true for the 780G, where AIT changes affect both the algorithm’s behaviour and the user’s meal management demands simultaneously. Moving to a shorter AIT without adjusting meal bolusing consistency tends to show up clearly in post-meal glucose data within a few days.
What the output does not account for
- Missed, delayed, or underestimated meal boluses
- Exercise effects — particularly exercise that precedes or follows meals
- Growth, puberty, or significant weight change
- Illness, infection, or steroid use
- Hormonal variation across the menstrual cycle
- Individual variation in insulin absorption from infusion sites
The right use of this output: bring it to your diabetes care team as a starting point for discussion. It is most useful as a way of structuring the conversation about whether current 780G settings are likely to be over- or under-tuned for the level being used — not as a prescription to implement without clinical review.
Limitations and educational frame
The GNL Multi-System Explorer and this explainer are educational resources for people living with type 1 diabetes and the clinicians who support them. They are not clinical decision support tools. They do not generate personalised recommendations and have not been validated as a medical device.
What these outputs are
- Directional estimates based on population-level physiological rules
- A framework for structured conversation with a diabetes care team
- An educational resource for understanding how AIT shapes 780G behaviour
- A starting point for exploring whether current settings are broadly aligned with a chosen level
What these outputs are not
- A prescription or clinical recommendation
- A validated prediction of individual glucose response
- A replacement for CGM data review or clinical assessment
- An endorsement of any specific AIT level or configuration
No proprietary claims are made about Medtronic algorithms. The SmartGuard algorithm operates with complexity and real-time adaptivity that this educational framework does not and cannot replicate. The five-level AIT framework is a simplified representation for educational purposes.
This content is for educational exploration only. It describes average responses and general principles. It is not medical advice and cannot replace individual clinical guidance from your diabetes care team.
IOB, responsiveness, and what it means before exercise
Active Insulin Time is not just a display setting — it is the primary mechanism through which the 780G controls how aggressively it behaves. Understanding what AIT is doing to the IOB calculation explains why level changes produce such visible effects on glucose management.
IOB is the brake. AIT controls how hard that brake is applied.
The SmartGuard algorithm checks insulin on board before deciding whether to deliver an automatic correction. If IOB is high, it holds back. If IOB is low, it acts. A shorter AIT produces a lower IOB estimate at any point after a bolus — which means fewer constraints on delivering corrections. Both meal and correction insulin are counted in the same pool. That is the mechanism behind responsiveness: every level change is an IOB accounting change, not a physiological one.
This creates a structural gap between what the device calculates and what is physiologically circulating. The table below illustrates this for a 0.1 u/kg bolus — a typical meal dose — comparing physiological insulin exposure against what AIT 2h and AIT 3h register as remaining IOB.
| Time post-bolus | Physiological estimate (% of dose still active) | AIT 2h device display (level 5) | AIT 3h device display (level 1) |
|---|---|---|---|
| 30 minutes | ~80% — insulin peaking, most still to act | 75% — broadly aligned | 83% — broadly aligned |
| 1 hour | ~65% — approaching peak action | 50% — already underestimating | 67% — closely aligned |
| 90 minutes | ~50% — at or near peak action | 25% — significantly underestimating | 50% — closely aligned |
| 2 hours | ~35% — still meaningfully active | 0% — device shows no IOB | 33% — closely aligned |
| 3 hours | ~15% — tapering but present | 0% | 0% — device shows no IOB |
| 4 hours | ~5% — trace amount | 0% | 0% |
Physiological estimate based on standard rapid-acting insulin analog action profile. Individual action duration varies with insulin type, infusion site, and temperature. The glucose-lowering potential at 2 hours for a 70 kg person taking a 7 unit bolus (0.1 u/kg) with a correction factor of 2.2 mmol/L (40 mg/dL) per unit is approximately 5.5 mmol/L (100 mg/dL) — still circulating at the point AIT 2h shows zero.
This is not a malfunction. The algorithm uses the shortened AIT to keep IOB low and stay willing to correct. But the displayed IOB no longer reflects what is physiologically present. This gap matters most in one specific situation: before exercise.
Negative IOB
The 780G also supports negative IOB. When the algorithm predicts glucose is heading below the active target of 5.5 mmol/L (100 mg/dL), it reduces or suspends delivery and registers this under-delivery as negative IOB — effectively subtracting from the IOB count. At higher responsiveness levels, the system is more active in both directions: adding corrections when glucose rises and generating negative IOB when glucose is predicted to fall. Negative IOB is an artefact of algorithm behaviour, not a physiological state.
The exercise problem — and why a heuristic helps
AID systems deliver insulin only. When exercise begins, glucose-lowering can accelerate rapidly — and the algorithm responds by reducing delivery. But if the device is already showing zero IOB at AIT 2h, it has no information about the physiological insulin that is still circulating and still working. The algorithm cannot protect against what it has stopped counting.
The four-hour heuristic: “Have I given a bolus for food in the last four hours?” If the answer is yes, there is physiologically circulating insulin — regardless of what the IOB display shows. Time elapsed is more reliable than the device display as an exercise safety check after a recent meal bolus on AIT 2h settings. The displayed IOB number should not be used as the sole guide to exercise safety in that window.
Understanding the gap between displayed and physiological IOB is one of the most important insights in managing AID systems at higher responsiveness levels. The GNL IOB guide explores this in depth, and the AID System Explorer lets you see what IOB looks like at each level compared to a physiological reference.
Activity to lower highs: the missing lever
AID systems are insulin-only and insulin is slow. Even at the most responsive settings, the algorithm cannot move glucose as quickly as activity can. Many people find that if they want time in range above 80% — or tight range above 50% — the most reliable additional lever is using activity to bring glucose down after meals, rather than relying entirely on the algorithm to manage post-meal rises. There is an explorer for this on GNL.
This content is for educational exploration only. It describes average responses and general principles. It is not medical advice and cannot replace individual clinical guidance from your diabetes care team.
Use the AID system explorer
The GNL Multi-System Explorer generates exploratory output ranges for the MiniMed 780G and three other systems. Use it as a starting point for discussion with your diabetes care team.
This content is for educational exploration only. It describes average responses and general principles. It is not medical advice and cannot replace individual clinical guidance from your diabetes care team.
Part of the GNL AID Systems guide
© 2024–2026 Glucose Never Lies®. All rights reserved. Registered trademark UK00004267795.
