Boost tools: Simulator · Tuning Guide · Analyser

Boost Analyser — V1 vs V6, on your own data

Every Boost build since the multi-user shadow beta logs both engines' decisions every cycle: what your running engine delivered, and what the other engine decided on identical inputs. This page reads that telemetry straight from your Nightscout and shows you the real, paired comparison. Works with AAPS Boost builds and the Trio (iOS) Boost port — the site type is auto-detected from the data.

Privacy: this page has no server. Your URL and token are used only by your browser to call your Nightscout directly; nothing is sent anywhere else. Use a read-only token (Nightscout → Admin tools → Subjects, role readable).

Nightscout URL

Read token

Look back

remember URL & token in this browser (localStorage) — untick on shared machines

Your actual glycaemia (on the engine you were running)

Time in Ranges

Dose decisions: V1 vs V6

Per-cycle dose difference (V6 − V1), all paired cycles. Left of centre = V6 decided less insulin than V1; right = more. The big bar at zero is both engines agreeing.

Night = 22:00–07:00 in your browser's timezone. “V6 extra at night, unannounced” is the pattern the V6 night gate (July 2026) now suppresses on current builds.

How the engines dose a meal, on average

All auto-detected rises (≥50 mg/dL within 90 min) aligned at onset. Lines are the mean cumulative insulin each engine decided from 30 min before onset — the shape difference IS the story: V6 holds, fires its commit-shot, then tapers; V1 ramps steadily. The gap is shaded (cyan where V6 has delivered more, amber where V1 has). Modal V6 state underneath.

V1 cumulative (mean) V6 cumulative (mean) ribbon: modal V6 state

Show the individual meals…

Where the engines differ across your day (counterfactual AGP)

Your actual BG by time of day (green median + interquartile band) against the projected other-engine band (cyan, dashed median). Lower medians are not free: wherever the other engine's 25th percentile dips below 70 mg/dL the low-side tail is filled amber — that is the hypo risk a lower median line hides. Same absorbed-Δdose × logged-ISF projection as the panels below, with an IOB-feedback ledger (the other engine's later intents shrink as its counterfactual insulin accumulates) and an LGS emulation (its further insulin suspends once projected BG reaches 90). Validated limitation: cross-checked against a dual-era user's real V1 history, even this bounded projection overstates low-side tails roughly 3× — rescue carbs and the zero-temp basal cut can't be modelled from decision data. So read the amber areas as where the hypo pressure concentrates, never as projected time-below-range.

actual median + band other engine, projected median + band other engine's projected low tail (<70)

Likely-avoided lows (first-order estimate)

How this is estimated — read before trusting it. For every BG nadir ≤95 mg/dL, the net dose difference between the engines over the preceding 3 hours is absorption-weighted (triangular activity, DIA 5h) and multiplied by your logged per-cycle ISF to project how much deeper (or shallower) the nadir would have been under the other engine. The range shown applies a 50% discount for engine feedback — the counterfactual engine would have reacted to falling BG (zero-temps, LGS), and carbs you might have eaten aren't modelled. Read these as pressure toward hypo, not predicted numbers.

Show the occasions…

Counterfactual highs (same method, other tail)

Show the occasions…

Levers to consider

V6 state machine & safety gates

Notes & caveats

Decision-level comparison only — see the banner at the top. For the validation methodology behind Boost changes (replay before ship), see backtesting/README.md in the repo.
Cycles are deduplicated to 5-minute buckets (AAPS re-invokes more often during SMB delivery).
On V6-active builds, V1's dose is parsed from the reason text (base would=…); cycles where that isn't present are counted as unpaired and excluded from dose deltas.
If your build predates the shadow telemetry (~May 2026), no boostV5_* fields exist and this page can't compare — update Boost first.
Trio sites are auto-detected per dataset: structured boostV5_* fields → AAPS; the boostV5[…] tag in suggested.reason (or the device: "Trio" signature) → Trio, where stock trio-oref is the acting engine and Boost V6 runs in shadow. The acting dose is trio-oref's suggested SMB (suggested.units; absent = 0), Boost V6 is parsed from the tag. AAPS-only telemetry (Phase-3 gate fields, cap values, the base would= counterfactual) has no Trio equivalent and is shown as n/a; Boost-ACTIVE Trio cycles don't log the oref counterfactual and are counted unpaired. Mixed windows (uploader switched) are analysed per cycle in each record's own mode.
TING = Time In Normo-glycaemia, 3.5–7.8 mmol/L (63–140 mg/dL), reported alongside standard TIR 70–180.