Boost Plugin Tuning Guide

💉

Core Delivery Controls

Primary levers for how much insulin Boost delivers

Boost Scale ApsBoostScale

Direct multiplier on the calculated UAM Boost insulin. At the 3.0 maximum only the UAM Boost tiers (3 & 4) stand down (they require scale < 3) — the COB tiers (1/2), percent-scale tier 5 and acceleration tier 6 still run. Setting 3.0 does not disable Boost or fall back to plain oref1.

1.0

0.1 (min)3.0 (max — gates off UAM tiers 3/4)

Max IOB ApsBoostMaxIob

Boost's own IOB gate — distinct from AAPS's global Max IOB (OpenAPS SMB settings), and both apply. On V1, above this the boost tiers stand down and standard oref1 corrections continue. On V6 it is a hard headroom clamp: a dose can never push IOB past min(Boost Max IOB, AAPS Max IOB), with a graduated brake from 50% of it. If you raise one, check the other — the lower of the two wins.

1.0 U

0.1 U12.0 U

Bolus Cap ApsBoostBolus

Maximum insulin per individual Boost SMB delivery regardless of other calculations.

2.5 U

0.1 U10.0 U

Insulin Required % ApsBoostInsulinReqPct

What fraction of the algorithm's calculated insulin need to actually deliver as an SMB.

50%

30% (cautious)100% (full)

Percent Scale Factor ApsBoostPercentScale

Max scale factor when BG is 110–180 mg/dL, rising (delta > 3 mg/dL/5min) and accelerating. Only active if Percent Scaling is enabled. The lower bound was deliberately raised from 98 to 110 after data showed high hypo rates when it fired at BG 90–110.

200%

50%500%

Per-SMB size caps (AAPS OpenAPS SMB settings — not on the Boost screen)

Max minutes of basal per SMB (UAM) uamsmbmaxminutes

This is the one that matters for Boost meals — unannounced rises run the UAM path, so each SMB is capped at current basal × this ÷ 60. Default 15 (range 15–120). With SMBs every ~3 min this is a genuine delivery-rate cap, not just a per-shot nicety. Raise gradually (15 → 30 → 45), never straight to high values. Its sibling smbmaxminutes (also default 15) covers the with-carbs path. V6 note: V6's own doses are bounded by its state caps instead — but these caps shape the V1 base dose, which is V6's fallback and (since July 2026) the ceiling on V6's non-meal-state doses.

15 min

15 (default)120 (max)

📐

Dynamic ISF

How insulin sensitivity is calculated from TDD and BG

BG Impact on ISF ApsBoostDynIsfVelocity

How much current BG level shifts ISF. 0% = constant ISF from TDD only. 100% = ISF scales fully with BG level. V2: always 100%.

100%

0% (flat)100% (BG-reactive)

DynISF Normal Target ApsBoostDynIsfNormalTarget

Reference BG used in the ISF formula. Lower = more aggressive ISF at a given BG.

99 mg/dL

70 (aggressive)120 (conservative)

DynISF BG Cap ApsBoostDynIsfBgCap

BG above this is softened: effective BG = cap + (actual − cap) ÷ 3. Prevents extreme ISF at very high BG. A low cap softens more of the range = weaker ISF up high = less insulin (conservative); a high cap leaves the full BG signal in play (aggressive).

210 mg/dL

100 (most softening — conservative)300 (least softening — aggressive)

TDD Adjustment Factor ApsBoostDynIsfAdjustmentFactor

Scales the TDD used in ISF calculation. Below 100% = treats you as lower TDD = weaker ISF = less insulin.

100%

1% (very weak ISF)300% (very strong ISF)

BOOST v1 ISF Formula

ISF = 1800 ÷ (TDD × ln(target ÷ div + 1))

TDD is a blended 8h/7D/1D weighted average. Velocity scales BG sensitivity.

BOOST v2 ISF Formula (Chris Wilson)

ISF = 2300 ÷ (ln(target ÷ div + 1) × TDD² × 0.02)

Squares TDD — more sensitive to TDD changes. Velocity is always 1.0.

🏃

Activity Detection

Step-count thresholds and insulin adjustment when active

When ANY of the four windows below is exceeded, the Activity Profile % is applied and BG target is set to 150 mg/dL (if no temp target is active).

5-min window

420

steps triggers active

15-min window

800

steps triggers active

30-min window

1,200

steps triggers active

60-min window

1,800

steps triggers active

Activity Profile % ApsBoostActivityPct

Applied when active. Works like a profile switch: ISF is divided by (this ÷ 100) and basal multiplied by it — so <100% = higher ISF + lower basal = LESS insulin while you're moving. The default 80% cuts delivery roughly 20%. Values >100% would give more insulin (rarely wanted for activity).

80%

30% (least insulin)150% (more insulin)

Inactivity Profile % ApsBoostInactivityPct

Applied when steps/hr is below the inactivity threshold. Same mechanics: >100% = lower ISF + higher basal = MORE insulin during sedentary periods, countering sedentary insulin resistance. The default 130% delivers ~30% more. 100% = no change.

130%

100% (no change)200% (much more insulin)

Inactivity Threshold ApsBoostInactivitySteps

Steps per hour below which inactivity mode triggers. Higher = easier to trigger the sedentary (more-insulin) adjustment.

500/hr

0 steps/hr1,000 steps/hr

🌙

Night Mode & Sleep-in

SMB suppression near target — and, since July 2026, the schedule that decides when V6 doses at all

Night Mode

Two jobs: (1) suppresses SMBs in the night window while BG is near target (within BG Offset — a genuine high still gets corrected); (2) defines when V6 dosing is active — V6's amplified doses run only OUTSIDE the night/sleep period, falling back to the V1 base dose overnight. With night mode OFF, V6 runs 24/7 with no scheduled off-switch — recommended ON for V6 users.

Default: OFF

Night mode auto by sleep (HR)

Lets the HR/step sleep detector extend night mode past the clock (still asleep) or start it early (pre-sleep) — and, on V6, flex the dosing window with your actual sleep. Needs a worn HR source.

Default: OFF

Night Mode BG Offset ApsBoostNightModeBgOffset

SMBs suppressed when BG < (profile target + offset). Higher offset = wider suppression zone = more conservative overnight.

27 mg/dL

0 (only at target)90 mg/dL (wide zone)

Disable Night Mode if COB > 0

Allow SMBs during night if carbs on board

Default: OFF

Disable Night Mode with Low TT

Allow SMBs if a low temp target is active during night window

Default: OFF

Sleep-in Detection

Sleep-in Hours ApsBoostSleepInHours

Window after night-mode end where a lie-in is detected: if your 60-min steps stay below the threshold, Boost/V6 stays down and night-mode SMB rules keep applying until you're actually up. The false-AWAKE backstop for mornings when HR-based sleep detection misfires. Longer window = conservative for longer.

2 hrs

0 hrs (disabled)18 hrs

Sleep-in Step Threshold ApsBoostSleepInSteps

60-min step count below which the lie-in holds within the window. Higher = easier to trigger (stays conservative longer into the morning).

250/hr

0 steps/hr1,000 steps/hr

⚙️

Feature Toggles

Enable/disable major Boost features

TDD-based ISF

Calculates ISF from Total Daily Dose. Required for Boost's dynamic ISF to function. Mandatory for V2.

Default: OFF

TDD Sensitivity Adjustment

Adjusts sensitivity using 24h vs 7D TDD ratio (like Autosens). Start with OFF.

Default: OFF

Boost Percentage Scaling

Scales SMBs when BG is 110–180 mg/dL, rising (delta > 3) and accelerating. Uses Percent Scale Factor.

Default: OFF

Circadian ISF

Applies time-of-day multiplier to ISF based on circadian polynomial model. Low overnight, peaks afternoon.

Default: OFF

High TT Raises Sensitivity

When exercise temp target is active, increase ISF (reduce insulin delivery).

Default: OFF

Low TT Lowers Sensitivity

When correction temp target is active, decrease ISF (increase insulin delivery).

Default: OFF

Allow Boost with High Temp Target

By default, Boost is disabled when a high temp target is set. Enable to override this.

Default: OFF

⏰

Time Window & Safety

When Boost is active and data source safety checks

Changed July 2026: the old Boost start/end times are retired. Boost/V6 is now active whenever you are not in the night/sleep period — i.e. the schedule is your Night Mode window (22:00–07:00 by default), optionally flexed by HR/step sleep detection (auto by sleep) and held down through a morning lie-in by the steps check. Overnight, V6's amplified dosing stands down and the night-mode-aware V1 base dose runs instead. There is nothing to set here beyond the Night Mode card — if a rise isn't being chased at 3am, that is by design.

Cumulative SMB cap (60 min) ApsBoostCumulativeSmbCap60Min

Rolling one-hour ceiling on total SMB volume — the anti-stacking backstop that bounds dose frequency where the per-shot caps bound magnitude. Auto-config sets it ≈ one commit-shot + two holds; the default (10 U) is deliberately non-binding until auto-config personalises it. Applies to V1 and to the V6 override alike, and since July 2026 it fails closed (SMBs pause for a cycle) if the phone can't read recent bolus history.

10 U

0 = cap disabled (not conservative)10 U/h

Safety Settings

Allow All BG Sources / Bypass Version Expiry

No longer user settings. In this fork both are baked in (ON) as a deliberate maintainer decision for the beta cohort — the toggles were removed from the UI in June 2026. If you build from source and want stock-AAPS behaviour, revert the forced flags in code.

Baked in: ON

⚠️ The remaining live safety levers are both Max IOBs, the dose caps, the cumulative SMB cap and Night Mode. Tune them one at a time, in small steps, with several days between changes.

🎯

Quick Reference: V1 Tuning Direction

Starting points when you need more or less aggressiveness from the V1/V2 tier engine

To make Boost V1 more conservative

Lower Boost Scale (e.g. 0.7–0.8)
Lower Max IOB to limit how long Boost stays active
Lower Bolus Cap to reduce per-SMB size
Lower Insulin Required % (e.g. 30–40%)
Lower DynISF Velocity to reduce BG reactivity
Raise DynISF Normal Target (less aggressive ISF reference)
Enable Night Mode with a larger BG Offset
Lower Inactivity Profile % toward 100% (no extra insulin when sedentary)
Enable High TT Raises Sensitivity

To make Boost V1 more aggressive

Raise Boost Scale (e.g. 1.2–2.0 — at 3.0 the UAM tiers 3/4 gate off)
Raise Max IOB to allow more sustained Boost delivery
Raise Bolus Cap to allow larger SMBs
Raise Insulin Required % toward 100%
Raise DynISF Velocity to 100% (full BG reactivity)
Lower DynISF Normal Target (stronger ISF reference)
Enable Percent Scaling for rising-BG amplification
Raise Inactivity Profile % (more insulin when sedentary)
Enable Low TT Lowers Sensitivity for correction targets
Raise TDD Adjustment Factor above 100%

Boost V2 differences to keep in mind: V2 uses TDD² in its ISF formula — it is more sensitive to TDD changes than V1 and reacts more strongly to short-term variations. BG Velocity is always 100% in V2 and cannot be reduced. Start with conservative delivery settings (lower Scale, lower Max IOB) when switching to V2.

Common Scenario Guides — Boost V1

Specific settings responses to real-world control problems — start with the highest-priority fixes first. Running V6? Its scenario guides are on the Boost V6 tab.

What is likely happening

Boost's COB tiers (Tier 1 <25 min, Tier 2 <40 min) deliver insulin proportional to insulinReq, which is driven by current ISF and the gap between BG and target. If ISF is tuned aggressively or insulinReqPct is high, these tiers stack significant insulin on top of the meal bolus.

If the profile ISF is too low (number too small), the algorithm overestimates how much insulin is needed per mg/dL correction. Combined with a large Max IOB headroom, this allows continued stacking well into absorption.

Boost's Fast Carb Protection reduces aggressiveness when a recent low + fast rise is detected, but it cannot fire if the hypo itself is triggered by the overcorrection — the protection is for pre-emptive fast carb scenarios, not post-delivery correction.

Settings to change

Insulin Required % first
Reduce to 35–40%. This directly scales how much of the calculated need is acted on during COB tiers.
ApsBoostInsulinReqPct → 35–40%
Boost Max IOB first
Lower to 0.5–0.8 U. This is Boost's own IOB gate (not AAPS's global Max IOB): V1's boost tiers stand down above it, and on V6 it hard-clamps dose headroom. Lowering it is the cleanest stacking limiter.
ApsBoostMaxIob → 0.5–0.8 U
Bolus Cap
Reduce to 1.0–1.5 U as a hard ceiling on any single SMB delivery.
ApsBoostBolus → 1.0–1.5 U
Boost Scale
If hypos persist after the above, reduce scale to 0.7–0.8. This reduces the Tier 3+ UAM bolus magnitude.
ApsBoostScale → 0.7–0.8
Profile ISF (outside Boost settings)
Verify ISF is not too low (too aggressive). A too-low ISF means insulinReq is overestimated at every step. Fasting basal testing first, then ISF.
Profile CR (outside Boost settings)
A too-low CR (too aggressive) means the pre-meal bolus already overdoses before Boost fires. Check CR separately from Boost.
Running V6? The V6-side levers for post-meal hypos (Hypo Caution, Aggression, the dose caps) are in Scenario 1 on the Boost V6 tab.

What not to do

Don't push Scale to 3.0 expecting to disable Boost — it only gates off the UAM tiers (3/4); the COB, percent-scale and acceleration tiers keep running. If you need Boost fully off, switch the APS plugin instead. And don't reach for max scale as a hypo fix at all — you lose UAM meal coverage while keeping the tiers that were likely stacking.
Don't raise the DynISF Normal Target as a first fix — this makes ISF less aggressive at a given BG but doesn't address the stacking behaviour during COB tiers.
Don't extend Night Mode hours into the day to suppress post-meal SMBs — this is a blunt instrument that will block needed corrections at other times.
Don't crank the SMB-minutes caps down as a first fix — uamsmbmaxminutes / smbmaxminutes already default to the minimum (15). They ARE a real rate cap (SMB interval is fixed at ~3 min), but if ISF/CR are wrong the correction pressure just persists for longer; fix the profile signal first.
Don't ignore profile calibration — Boost amplifies what the profile tells it. A miscalibrated ISF or CR will always produce bad meal outcomes regardless of Boost settings.

What is likely happening

The algorithm is calculating an insulinReq but delivering only a fraction of it — either because insulinReqPct is low, or because Max IOB is hit quickly and Boost falls back to oref1 which delivers standard corrections.

During a post-meal plateau where BG is stable but high (flat delta), Boost uses Tier 7 (mild correction) or standard oref1 logic rather than the more aggressive UAM tiers, because acceleration signals are gone. The correction speed is then entirely governed by ISF and Max IOB headroom.

If Percent Scaling is disabled, the rising 110–180 mg/dL window never gets the amplified SMB it's designed to handle. Enabling this is specifically designed for this pattern.

A too-high ISF (conservative number) underestimates insulinReq — each SMB chips away slowly. Verify ISF is accurate for corrections.

Settings to change

Check Boost is actually active at meal times first
Since July 2026 the schedule is the Night Mode window (Boost runs when you're NOT in the night/sleep period) — the old start/end times are retired. Late dinners after night-mode start get the conservative night-mode dosing by design; if evening meals plateau, review your night-mode start time rather than hunting for a window setting.
Night Mode start/end · check the NS reason line
Both Max IOBs first
Raise Boost Max IOB (default is a deliberately-low 1.0 U — the most common reason correction stalls) — and check AAPS Max IOB too: on V6 the LOWER of the two is the hard ceiling. Small steps: 1.0 → 1.5 → 2.0, watching the loop graph for IOB pinned at the limit.
ApsBoostMaxIob → 1.5–2.5 U (and verify AAPS Max IOB ≥ it)
Enable Percent Scaling
Specifically designed for BG in the 110–180 rising-and-accelerating band. Set Percent Scale Factor to 200–300% to amplify SMBs as BG rises through this band.
ApsBoostEnablePercentScale → ON

ApsBoostPercentScale → 200–300%
Insulin Required %
Raise to 70–80%. Delivers more of the calculated need per loop cycle rather than a conservative fraction.
ApsBoostInsulinReqPct → 70–80%
UAM SMB minutes (then SMB minutes)
Both default to the minimum (15 min ⇒ each SMB ≤ basal×0.25). For unannounced meals the UAM one is the cap that binds. Raise gradually — 15 → 30, live with it, then 45 if still throttled. Don't jump to 60+ in one go: it quadruples the per-SMB ceiling.
uamsmbmaxminutes → 30 (first step) · smbmaxminutes → 30
DynISF Normal Target
Lower from 99 to 90 mg/dL. This makes the ISF formula produce a stronger (lower) ISF value, increasing insulinReq for a given BG gap.
ApsBoostDynIsfNormalTarget → 90 mg/dL
Running V6? V6-side plateau levers (Confirmed/Committed caps, Aggression, the night gate) are in Scenario 2 on the Boost V6 tab.

What not to do

Don't raise Boost Scale above 2.0 before verifying Max IOB and Insulin Required % are correct — scale amplifies Tier 3+ UAM boluses but has little effect on plateau corrections where acceleration is flat.
Don't rely on Scale alone for plateaus — scale primarily affects the fast-rise UAM tiers. Once BG is flat and high, the plateau is corrected by ISF and IOB headroom, not scale.
Don't raise TDD Adjustment Factor aggressively (e.g. 200%+) to force stronger ISF — this distorts the Dynamic ISF calculation away from your actual insulin needs and can create instability.
Don't fight the night schedule — if the "plateau" is overnight, reduced dosing near target is deliberate (genuine highs beyond the BG offset still get corrected). Adjust the night window or basal, not aggression.
Don't tighten CR to compensate for post-meal plateaus — if Boost is active and adequately configured, a tight CR risks pre-meal overdose. Correct the loop behaviour first.

What is likely happening

Persistent above-target BG between meals usually signals a profile calibration problem rather than a Boost tuning problem. The algorithm can only correct what the profile's ISF and basal allow it to correct.

If TDD-based ISF is disabled, Boost uses the profile ISF directly. If that ISF is too conservative (too high a number), every correction is undersized. Dynamic ISF's purpose is to adapt — without it, the algorithm is no better than standard oref1.

Night Mode or a morning lie-in hold may be suppressing corrections into hours when you are actually up and high. Conversely, if the highs cluster in sedentary hours, the inactivity adjustment (>100% = more insulin when sedentary) may simply be set too low to counter your sedentary resistance.

A too-low Max IOB hits the ceiling quickly during sustained correction periods. Autosens will attempt to compensate but is bounded by its min/max ratio limits.

Settings to change

Enable TDD-based ISF first
If not already on, this is the most impactful change. Boost's Dynamic ISF is the core mechanism for adapting corrections to your actual insulin needs.
ApsBoostUseTdd → ON
Check when Boost is actually active first
The schedule is now the Night Mode window (+ sleep detection + morning lie-in hold). If your "stuck high" hours are early morning, check the lie-in settings (sleep-in hours/steps) aren't holding Boost down after you're up — the NS reason line shows "Sleep-in" when they are.
Night Mode window · Sleep-in hours/steps · NS reason line
Both Max IOBs
Raise Boost Max IOB in small steps (1.0 → 1.5 → 2.0) and make sure AAPS Max IOB isn't the lower, binding limit — on V6 the lower of the two wins. The loop graph shows IOB pinned at the ceiling when this is the constraint.
ApsBoostMaxIob → 1.5–2.5 U (verify AAPS Max IOB ≥ it)
DynISF Normal Target
Lower from 99 to 90–95 mg/dL to make the ISF formula produce stronger corrections at elevated BG.
ApsBoostDynIsfNormalTarget → 90–95 mg/dL
Review Night Mode & Inactivity settings
Confirm Night Mode hours do not overlap with when you are awake and high. If the highs cluster in sedentary hours (desk work, evenings on the sofa), raise Inactivity Profile % — above 100% it delivers MORE insulin when your steps/hr drop below the threshold, which is exactly the counter for sedentary resistance.
Night Mode window — check overlap

ApsBoostInactivityPct → raise (e.g. 130 → 150) for sedentary-hours highs
Enable Low TT Lowers Sensitivity
When you deliberately set a correction temp target, this makes the algorithm more aggressive. Useful for managing persistent highs intentionally.
ApsAutoIsfLowTtLowersSens → ON
Profile basal and ISF (outside Boost settings)
Persistent above-target BG is often a signal that the profile basal or ISF is fundamentally too conservative. Autosens will show a persistent ratio > 1.0 if it's trying to compensate. Address the root cause in the profile.
Running V6? The V6-side lever for chronic under-correction (Sensitivity) is in Scenario 3 on the Boost V6 tab.

What not to do

Don't raise TDD Adjustment Factor above 150% as a workaround for poor profile calibration — this forces the ISF formula to behave as if you use far more insulin than you do, creating instability and potential hypo risk when your real TDD is used.
Don't raise Boost Scale to near 3.0 — high scale inflates UAM bolus tiers but has minimal effect on persistent between-meal highs where BG is flat. It also increases hypo risk when you eventually come back on target (and at exactly 3.0 the UAM tiers 3/4 stop firing at all).
Don't ignore autosens ratio — if autosens is sitting persistently at its maximum (e.g. 1.2), it's hitting its ceiling trying to compensate for a miscalibrated profile. Widening autosens limits is not the fix; correcting the profile is.
Don't set a high BG target in the profile to "avoid lows" — if the profile target is 130+ mg/dL, the algorithm will never drive BG below that regardless of settings. Set a correct target and use night mode for overnight protection instead.
Don't increase Bolus Cap to try to force larger corrections — between-meal corrections are limited by insulinReq (ISF × BG gap), not bolus cap. Bolus cap only limits the ceiling; if insulinReq is already below the cap, raising the cap does nothing.

🤖

Boost V6 — the meal state-machine engine

A different model: same ISF as V1, but the 8-tier SMB ladder is replaced by a meal-state machine — and 24 knobs collapse to 3

V6 does not change how ISF is calculated — it reuses V1's insulinReq (DynISF, TDD, autosens all unchanged; those settings live on the Boost V1 tab and still matter). What changes is the SMB decision. Instead of picking one of eight tiers, V6 tracks where a meal is in its lifecycle and doses:

BOOST V6 SMB

SMB = aggression_budget × action_multiplier(meal_state) → velocity-scale → state-cap → safety gates

aggression_budget = max(30% × insulinReq, insulinReq × hypo-damper × post-exercise × sensitivity)

IDLE

×1.0

baseline — oref pass-through, no meal detected

OBSERVING

×0.3

small test dose while a rise is watched (score ≥ 0.44)

CONFIRMED

×1.8

meal confirmed — one catch-up shot per meal (capped by Confirmed cap)

COMMITTED

×1.0

sustained per-cycle holds (capped by Committed cap)

RECOVERING

×0.4

backing off as BG decelerates and IOB bites

The aggression knob scales only the CONFIRMED multiplier — concentrating user tuning on the single riskiest decision. The state is driven by a continuous meal-signal score (BG delta + acceleration + ML meal-likelihood + time-of-day + sustained rise, minus a recent-low penalty).

🎚️

V6 user knobs

The only three settings most V6 users ever touch — full detail in the README

Aggression ApsBoostV5Aggression

Scales the CONFIRMED catch-up dose only (1.8× base). Lower = gentler meal response; higher = firmer. Ceiling raised to 1.6 (July 2026) for users who run consistently high and want a firmer confirm; auto-config still seeds it at neutral or below and never auto-raises it.

1.0

0.7 (gentle)1.6 (firm)

Hypo Caution ApsBoostV5HypoCaution

How hard V6 trims the budget when the ML hypo-risk model is elevated. Higher = more backoff (and a lower floor: 0.50 → 0.25 across 1.0 → 2.0) = more conservative.

1.0

1.0 (least caution)2.0 (most caution)

Sensitivity ApsBoostV5Sensitivity

Per-user calibration on the whole budget. <1.0 trims V6 for users it runs hot on; >1.0 lets it run firmer for resistant users.

1.0

0.8 (runs hot)1.2 (resistant)

⚙️

Shared & advanced

Inherited safety limits + rarely-touched V6 internals — each one, and how auto-config chooses its value, is detailed in the Advanced settings reference

Shared with V1safety limits

Both Max IOBs (Boost's and AAPS's — the lower wins as V6's hard headroom clamp), Night Mode (+ BG offset — and it schedules V6; see the Night Mode card on the Boost V1 tab), the cumulative SMB cap (ApsBoostCumulativeSmbCap60Min — full setting on the V1 tab; auto-config sets it ≈ one commit-shot + two holds, and it fails closed if bolus history can't be read), and SMB / UAM enable all still gate V6. Overnight and during a lie-in V6 stands down to the V1 base dose.

Dose capsApsBoostV5ConfirmedCapU / ApsBoostV5CommittedCapU

Hard per-cycle ceilings: the CONFIRMED commit-shot (setting default 2.5 U) and the COMMITTED holds (default 0.5 U). Auto-config personalises them from your own history on first V6 activation — Confirmed ≈ your biggest typical single dose (1.5–7.5 U), Committed ≈ your routine SMB (0.25–2.5 U). The commit-shot must also be worth more than one hold (velocity-adjusted) before V6 will spend it — see the safety layer below.

Fast-carb confirmApsBoostV5FastCarbConfirm — default ON

A sharp (Δ ≥ 6 mg/dL/5min), accelerating (≥ 10%), score-corroborated (≥ 0.65) rise jumps straight to CONFIRMED in one cycle. Since July 2026 it is suppressed within an hour of a low (< 80 mg/dL) — rescue-carb rebounds no longer trigger a commit-shot onto carbs that are already spoken for. Auto-config turns it off for a low-prone history.

V6 acceleration primer (U)ApsBoostV5PrimerCapU — default 0 (off)

A small early dose on an accelerating rise while V6 is still in OBSERVING, reclaiming the roughly 15 minutes V1 used to gain by reacting to acceleration before the meal was confirmed. It fires once per meal session, only when acceleration exceeds 10 with a positive delta, and only with every floor clear: no recent low below 80 mg/dL, awake, not exercising, not inside the post-rescue window, and inside maxIOB headroom. The size scales with how sharp the rise is, capped at twice the base you set here.

The base is additive; anything the acceleration scaling adds beyond it is netted off the CONFIRMED commit-shot, so a fizzled session costs you the base only and a real meal moves insulin earlier rather than adding more of it. The accumulator decays with a 90-minute time constant and carries across sessions, so several fizzles in a row are credited against the next commit-shot rather than stacking. Auto-config-managed: derived per user as a fraction of your committed cap, clamped to 0.6 U. 0 = off, which is the setting default until auto-config or you set it.

Force bolus primerApsBoostV5PrimerBolusMode — default OFF

How the primer is delivered. Auto-config routes a hypo-prone history through a retractable temp basal over 30 minutes instead of a bolus, so the insulin can be withdrawn if the rise fizzles; a well-controlled history gets it as a bolus. This switch overrides that routing and forces the bolus. The floors and the commit-shot net-off are identical either way, so it changes the delivery route and nothing else. The routing flag itself (ApsBoostV5PrimerTbrFallback) is auto-config-managed and is not a dial you set.

Pre-meal targetApsBoostV6PreMealTarget

Anticipatory target-lowering ahead of a learned meal time. Off by default.

Confirm-sooner (A2) & Velocity-budget floor (C5)auto-config opt-ins

Two per-user levers for people who run high and confirm late: A2 brings the meal confirm forward a cycle when the score is already convincing; C5 puts a small floor under the velocity factor so a fast rise isn't throttled to nothing. Both are auto-config-managed and off unless your history is well-controlled — see the auto-config rule below. Aggression levers, not guards.

Phase-3 composed brake floorApsBoostV5ComposedFloorActive — default OFF

Enforces a 25% floor on the composed soft-brake multiplier during active meal sessions above 160 mg/dL with eventualBG above target — fixes the soft-brake stack compounding to sub-pump-step zero doses mid-meal (July 2026). All hard gates and dose caps still apply. Auto-config-managed, per-user: this is an aggression lever, not a guard — enabled only if your trailing 14-day time-below-range is within consensus targets (<70 mg/dL below 3.5%, <54 mg/dL below 0.8%).

Anticipation (routine learning)shadow — anticip=

A per-user model that learns when you tend to eat and exercise from your own history (refit periodically, never in the dosing loop) and predicts the next meal/walk ~45 min ahead. Read-only for now — it doses nothing; it is the substrate for future retractable, per-user anticipatory dosing (an anticipated action that unwinds if the event doesn't materialise).

The July 2026 safety layer (all automatic, all replay-validated against real cohort data before shipping):

Night gate — V6's override runs only outside the night/sleep period (see the Night Mode card).
Non-meal-state cap — in IDLE / OBSERVING / RECOVERING, V6 can never out-dose what V1 would have given on the same cycle; V6 only exceeds V1 when it actually holds a meal hypothesis.
Commit-shot worth check — the once-per-meal CONFIRMED shot only fires when the velocity-adjusted dose beats one routine COMMITTED hold, so a noisy pre-meal blip can't burn the meal's one catch-up shot on ~0.05 U.
Rescue-carb guard — the fast-carb path stands down for an hour after a < 80 mg/dL low.
Post-hypo meal-state cap — for 45 minutes after any BG < 75 mg/dL, even a CONFIRMED/COMMITTED meal state can't out-dose the hypo-restrained V1 base — a rescue-carb rebound inherits V1's post-rescue restraint instead of drawing a full V6 catch-up shot.
Composed rebound guard (late July 2026) — in that same 45-minute post-low window, whatever SMB the fallback tiers would deliver is scaled down by glucose (~30% below 6.7 mmol/L, ramping back to full by 9.4). This closes the confirm-crash: a modest rebound after a low used to be over-treated and then crash a second time. It only ever removes insulin.
Elevated-HR exercise stop (late July 2026, critical) — Boost stops adding insulin during hard exercise. The activity classifier no longer mis-reads a moving-but-low-step session as resting, and the resting branch no longer ignores an elevated heart rate — the loop could previously add insulin as BG fell during exercise.
Fail-closed anti-stacking — if recent-bolus history can't be read, SMBs pause rather than lose the cap.

V1 → V6 in one line: V1 exposes ~24 knobs and routes insulinReq through an 8-tier ladder; V6 exposes 3 and routes the same insulinReq through a meal state-machine. The headline trade-off: V6 deliberately holds back in OBSERVING (a 0.3× test dose) where V1 tiers may already fire — then catches up hard in CONFIRMED once the meal is real. See this side-by-side in the simulator (V1 tiers vs V6 panel, and the dual track on the BG Projection tab).

🎯

Quick Reference: V6 Tuning Direction

Starting points when you need more or less aggressiveness from the V6 state machine

To make Boost V6 more conservative

Raise Hypo Caution (1.0 → 1.3–1.5) — harder budget trim on elevated ML hypo-risk
Lower Aggression (1.0 → 0.85–0.9) — softens the CONFIRMED commit-shot
Lower Sensitivity (1.0 → 0.9–0.95) — trims the whole budget
Lower the Confirmed / Committed caps one step below their auto-config values
Lower Boost Max IOB — the hard headroom clamp (graduated brake from 50% of it)
Enable Night Mode — schedules V6's amplified dosing off overnight (V1 tab)

To make Boost V6 more aggressive

Raise Aggression (1.0 → 1.1–1.2, ceiling now 1.6) — firmer catch-up shot on confirmed meals
Raise Sensitivity (1.0 → 1.05–1.1) — firms the whole budget for resistant users
Raise the Confirmed cap modestly if commit-shots keep hitting it on meals that still peak high
Raise the Committed cap one step (0.5 → 0.75) if plateaus are held on capped holds
Raise both Max IOBs in small steps — the lower of the two is V6's hard ceiling

One knob at a time. Auto-config already personalised the caps and cumulative cap from your history on first activation (the cumulative budget = one confirm shot + two holds, sized from the caps that actually apply; and it never auto-raises a dose cap while your 14-day time-below-70 is over 4% — such raises are only suggested in the notification) — run the Analyser before touching anything: it detects cap-binding, gate and schedule patterns from your own Nightscout data. V1's DynISF/profile settings (Boost V1 tab) still shape V6's insulinReq input.

Common Scenario Guides — Boost V6

V6-side responses to the same real-world control problems. Profile and V1 base-dose fixes are on the Boost V1 tab — they still apply, because V6's budget starts from the same insulinReq.

What is likely happening

V6's riskiest dose is the CONFIRMED commit-shot (1.8× base × Aggression, capped by ApsBoostV5ConfirmedCapU). If lows land 2–3 h post-meal, either the commit-shot or the sustained COMMITTED holds are oversized for your absorption.

The ML hypo-risk damper already trims the budget when risk is elevated — Hypo Caution controls how hard (and how low its floor goes). Since July 2026, rescue-carb rebounds can no longer trigger a commit-shot (fast-carb confirm is suppressed for an hour after a < 80 mg/dL low), and non-meal states can never out-dose V1.

If lows also happen on V1 or without meals, the pressure is the shared insulinReq (profile ISF/CR) — a V6 knob won't fix that.

Settings to change

Hypo Caution up, Aggression down first
Raise Hypo Caution (1.0 → 1.3–1.5) so elevated ML hypo-risk trims the budget harder, and/or trim Aggression (1.0 → 0.85–0.9) to soften the commit-shot. Run the Analyser first — it will tell you whether V6's own dosing was actually the low-side pressure.
ApsBoostV5HypoCaution → 1.3–1.5 · ApsBoostV5Aggression → 0.85–0.9
Confirmed cap
If the Analyser shows commit-shots consistently preceding the lows, lower the cap one step below its auto-config value.
ApsBoostV5ConfirmedCapU → one step down
Boost Max IOB
The hard headroom clamp — a dose can never push IOB past min(Boost Max IOB, AAPS Max IOB), with a graduated brake from 50% of it. Lowering it limits total stacking across the whole meal.
ApsBoostMaxIob → 0.5–0.8 U
Profile ISF / CR and the V1 base dose
See Scenario 1 on the Boost V1 tab — a miscalibrated profile overstates insulinReq for V6 exactly as it does for V1.

What not to do

Don't zero the dose caps to "switch off" the meal engine — you'll strand V6 in states that can't dose while it still holds the meal hypothesis. Turn V6 off properly if you don't want it dosing.
Don't disable Fast-carb confirm as a hypo fix — the July 2026 rescue-carb guard already blocks the rebound-low case; disabling it just slows genuine fast-carb meals.
Don't drop Sensitivity to the floor (0.8) in one move — it trims every state including corrections. Small steps, several days apart.
Don't ignore the profile — if lows also happen without meals, no V6 knob will fix a too-low ISF or CR.

What is likely happening

V6 front-loads its insulin: OBSERVING test dose → CONFIRMED catch-up shot → COMMITTED holds. A plateau that holds 2+ hours usually means a cap or clamp bound: the commit-shot hit ApsBoostV5ConfirmedCapU, the holds are pinned at ApsBoostV5CommittedCapU, or IOB reached min(Boost Max IOB, AAPS Max IOB).

Once the rise flattens, V6 moves to RECOVERING/IDLE — where it is deliberately capped at the V1 base dose (July 2026 non-meal-state cap). From there the correction speed is the V1 story: ISF and IOB headroom.

If the meal fell inside the night/sleep period (late dinner after night-mode start, or a lie-in hold), V6's amplified dosing stood down by design.

Settings to change

Check V6 was actually active at meal time first
The schedule is the Night Mode window (plus sleep detection and the morning lie-in hold). Look for "V6 override skipped (Boost inactive)" in the NS reason line. If evening meals plateau, review your night-mode start time.
Night Mode start/end · NS reason line
Both Max IOBs first
The LOWER of Boost Max IOB and AAPS Max IOB is V6's hard ceiling. Raise in small steps (1.0 → 1.5 → 2.0), watching the loop graph for IOB pinned at the limit.
ApsBoostMaxIob → 1.5–2.5 U (and verify AAPS Max IOB ≥ it)
Caps + Aggression, guided by the Analyser
If commit-shots keep hitting the Confirmed cap on meals that still peak high, raise it modestly; if the plateau is held on capped COMMITTED holds, raise the Committed cap one step (0.5 → 0.75). A small Aggression raise (1.0 → 1.1) firms up the catch-up shot. The Analyser's lever panel detects both cap-binding patterns from your data.
ApsBoostV5ConfirmedCapU / ApsBoostV5CommittedCapU · ApsBoostV5Aggression → 1.1
The V1 base dose still matters
Once the plateau is flat, V6 is bounded by what V1 would give — so Percent Scaling, SMB minutes, Insulin Required % and DynISF Normal Target (Scenario 2 on the Boost V1 tab) shape the correction speed even on V6.

What not to do

Don't fight the night gate — if the "plateau" is overnight, V6 standing down is deliberate (the V1 base dose still corrects genuine highs). Adjust the night window or basal, not aggression.
Don't max Aggression (now up to 1.6) before checking the caps and Max IOB — Aggression only scales the once-per-meal commit-shot; if the cap or the IOB clamp binds, more Aggression changes nothing.
Don't raise Sensitivity for a meal-only problem — it firms the entire budget all day; use the meal-specific levers first.
Don't tighten CR to compensate for post-meal plateaus — a tight CR risks pre-meal overdose. Correct the loop behaviour first.

What is likely happening

Between meals V6 sits in IDLE — an oref pass-through that (since July 2026) can never out-dose the V1 base dose. Persistent between-meal highs are therefore almost always a profile / V1-base problem (ISF, basal, TDD-based ISF off), not a V6 state-machine problem.

The exception is the schedule: a night window that overlaps waking hours, or a morning lie-in hold (sleep-in steps) keeping V6 down after you're actually up.

Sensitivity is the one V6 knob that firms the whole budget, including corrections — but only after you've confirmed nothing else is binding.

Settings to change

Check when V6 is actually dosing first
If your "stuck high" hours are early morning, check the lie-in settings (sleep-in hours/steps) aren't holding Boost down after you're up — the NS reason line shows "Sleep-in" when they are.
Night Mode window · Sleep-in hours/steps · NS reason line
Sensitivity
If V6 chronically under-corrects everywhere (not just meals), a small Sensitivity raise (1.0 → 1.05–1.1) firms the whole budget. Confirm with the Analyser that no cap or gate is the real binding constraint first — "big peaks with nothing binding" points at the profile instead.
ApsBoostV5Sensitivity → 1.05–1.1
Both Max IOBs
Raise Boost Max IOB in small steps and make sure AAPS Max IOB isn't the lower, binding limit — the lower of the two wins as V6's hard ceiling.
ApsBoostMaxIob → 1.5–2.5 U (verify AAPS Max IOB ≥ it)
Fix the V1 base and the profile
TDD-based ISF, DynISF Normal Target, Inactivity Profile % (for sedentary-hours highs) and the profile basal/ISF checks are all in Scenario 3 on the Boost V1 tab — they set the correction dose V6's IDLE state is capped to.

What not to do

Don't raise Aggression for between-meal highs — it scales only the CONFIRMED commit-shot; it does nothing in IDLE, where these highs live.
Don't crank Sensitivity to 1.2 to mask a miscalibrated profile — the profile error will still show up everywhere else (meal detection, hypo-risk, V1 fallback).
Don't ignore autosens ratio — persistently pinned at its maximum means the profile is the problem; widening limits is not the fix.
Don't set a high BG target in the profile to "avoid lows" — the algorithm will never drive BG below the profile target regardless of V6 settings.