Citation of the day
SvcHostSplitThresholdInKB tune
Legacy low-RAM experiment; it can change process overhead and fault isolation. Skip it on modern rigs unless pressure is measured.
Cited: SvcHost grouping behaviour · Windows 10 v1703 release notes
Diagnostics · then tune
Live DPC latency, thermals, driver health and PCIe link-state — read first, so you tune against numbers, not vibes. Every tweak cites a Microsoft Learn doc or a vendor whitepaper. One prompt applies the preset; snapshot revert at any time.
How it works
It launches in about 200 ms to a dashboard that already knows your hardware: live CPU, RAM and a tweaks-applied count, with one button to tune. The suite-wide auto-update banner sits up top, so you're always on the signed latest build.
This is the whole point. Browse ~100 audited tweaks, each with a risk rating and a real citation — a Microsoft Learn doc, an NVIDIA dev note, a vendor whitepaper. The other tuners don't print sources because they don't have any.
Your Tune lists every applied change with its live status. Snapshot-backed entries can restore their saved pre-state; script-based entries show their recovery path instead of pretending every inverse is universal. Undo is never a paid tier.
A one-shot Diagnostics snapshot reads your real specs, thermals, disk, driver health and live latency — so you tune against numbers, not vibes, and can prove the change actually helped on your rig.
The receipts
01 · The Audit
Microsoft Learn doc. Vendor whitepaper. NVIDIA dev note. The other tuners don't cite because they can't. We open the citation in your browser before you flip the switch.
02 · The Size
Tauri-on-Rust ships smaller than your discord avatar. No Electron. No 300 MB Chromium binary. Opens in 200 ms.
03 · The Undo
Whole batches apply behind a single elevation prompt. Snapshot-backed actions preserve their pre-state; script actions show the exact recovery path and refuse a guessed restore.
04 · The Presets
Esports · BR · Streamer · Frame Pacing · Network · Asta, plus community packs. Each bundle is inspectable before apply, and experimental entries are called out before you opt in.
05 · The Measure
DPC latency sparkline. Bufferbloat probe. PCIe link-state readout. We show the before, the after, and the delta — not a "trust me" toggle.
06 · The Hardware
Detects CPU vendor, RAM kit, OS build, laptop vs desktop. Hides every tweak that doesn't apply. Laptops never see PCIe ASPM off. AMD rigs never see Intel-only flags.
The audit
Hone has none. Paragon has none. We publish every citation, every category, every audit trail.
Citation of the day
Legacy low-RAM experiment; it can change process overhead and fault isolation. Skip it on modern rigs unless pressure is measured.
Cited: SvcHost grouping behaviour · Windows 10 v1703 release notes
Can help dual-monitor DWM stutter on some driver/display paths; test against a restore point.
NVIDIA driver-team forum responses + Hardware Unboxed multi-monitor coverage
Changes ACK/Nagle behavior per NIC; it matters only for some traffic patterns, so measure before keeping it.
Microsoft KB on Nagle's algorithm + competitive Apex/CS2 routing threads
Tags selected game packets; it helps only when the router/WAN honors DSCP and does not lower server ping by itself.
New-NetQosPolicy cmdlet docs + pro-circle Discord shared-WAN guides
Disables background capture; may reduce overhead when capture is active, so compare CPU and frame time before keeping it.
Xbox Game Bar docs · validated against Paragon + Hone presets
Changes the presentation path; compositor behavior and latency vary by game and display, so compare both modes.
DXGI swapchain presentation modes docs
Windows, driver, and GPU dependent; compare frame time and latency with HAGS on and off on the same game build.
NVIDIA driver release notes · frame-time analysis on Hardware Unboxed
Uses a driver profile only where supported; the in-game Reflex toggle is preferred, then compare the game indicator.
NVIDIA Reflex SDK developer docs
Changes prefetch and standby behavior; system-dependent, so use it only when memory pressure or stutter is measured.
Memory Manager (MMAgent) PowerShell reference
Flushes standby cache periodically; it may hide pressure but can hurt cache behavior, so compare long-session frame time.
Cache Manager + working set APIs docs · validated against RAMMap behaviour
Experimental BCD timer change; it can help or hurt by platform. Snapshot first and measure DPC and frame time.
BCDEdit timer settings reference
Experimental timer policy for supported CPUs; firmware and OS behavior vary, so keep the restore path and measure.
Intel SDM Vol 3B · CPUID feature flag 80000007h:EDX[8]
Changes foreground quantum bias; it may affect responsiveness but offers no universal FPS or input-latency guarantee.
Process Scheduling internals · Russinovich Windows Internals chapter 4
Runtime affinity policy; it may help a CPU-bound game on some rigs and hurt scheduling on others. Compare 1% lows.
CPU Sets Win32 API docs · Xbox dev-day session on per-process scheduling
NIC and driver dependent; it can alter packet CPU work and jitter, so compare under the same network traffic.
Intel + Realtek NIC tuning whitepapers
Experimental security trade-off; workload overhead varies, there is no fixed gain, and the exact revert path must remain available.
Set-ProcessMitigation cmdlet + Spectre/Meltdown mitigation docs
Keeps a more aggressive power policy; it may reduce idle savings but has no universal FPS gain. Compare clocks and frame time.
powercfg reference + Windows 10 IT Pro power plan docs
Sets a preferred display mode; it cannot override every driver/display setting or guarantee latency. Verify actual refresh.
Display Driver Model (WDDM) reference · monitor PNP key paths
Stops selected executables from Efficiency Mode demotion; Windows and game behavior vary, so compare background and active play.
Image File Execution Options (IFEO) + PowerThrottling docs
Changes name resolution only; it does not change in-match ping after connection. Compare resolver lookup time on your network.
Cloudflare 1.1.1.1 docs · DNS resolve-time benchmarks
Showing 20. The remaining 67 ship with the build — every one cited inline in the app.
100
tweaks · zero placebos
Tune what you measured. Revert what didn't help.
Three questions you were going to ask anyway
No. Every tweak cites a Microsoft Learn doc, NVIDIA dev note, or vendor whitepaper before you flip the switch. Twenty live on this page; the other 80 ship cited inside the build. Hone publishes none of theirs. Paragon publishes none of theirs.
No. Every action snapshots its pre-state to a local SQLite ledger before applying. One-click revert at any time, any tweak. Zero BIOS writes, zero kernel drivers — that line is a hard refusal in the engine.
Free. 3.6 MB. The catch is that a handful of the more aggressive tweaks (CPU set pinning, certain registry deep cuts) live behind a one-time $8 VIP unlock. Everything else — including full undo — is unlocked from the install.
Also in the apps
Communication
Stock Discord uses 1,041 MB on this machine. Ours uses 288. Same servers, same friends, same pings — we just stripped the extra work that wastes RAM and can add lag while you play.
Content
Done-for-you clip editing for streamers. We name the exact opponent, the exact weapon and the exact moment — cut it to vertical, caption it, and drop it in your drafts.
Fortnite
Two routes for every POI: cut, free-fall, and re-pull fast, or glide in safe. Built for the glider-cut meta. Current-season POIs.