Connects to tosu or Etterna, reads your current chart, and estimates its Dan tier live. Powered by the ISOR 3.2 engine — #1 exact-tier rate on VSRG Dan benchmarks — featuring native Etterna simfile support (.sm/.ssc), Wife% accuracy tracking, and 12 customizable overlay skins.
Ten Reform tiers · α through κ · logos by icyilc
Real-time difficulty evaluation, pattern classification, and telemetry for competitive VSRGs.
Calculates Reform and Celestial Dan ratings instantly upon song select. Evaluates sustained physical strain, local note density, and difficulty progression using calibrated isotonic regression curves.
Directly parses .sm and .ssc simfiles with real-time rate scaling (0.5×–2.0×).
Injects theme telemetry hooks for Til Death, Rebirth, and Simply Love while evaluating live Wife% accuracy
metrics.
Breaks down charts into granular archetypes including Chordjacks, Jumpstream, Handstream, Tech, and Stamina. Computes continuous NPS distributions and dominant MSD difficulty components.
Includes nine customizable HUD presets with configurable window positioning, opacity, and blur settings. Press G to render high-resolution PNG density charts directly to your Desktop.
Introduced in DanOverlay v3.0 and refined through v3.2, ISOR is a high-precision Dan estimation engine specifically engineered for osu!mania 4K rice. Designed to eliminate heuristic step cliffs and rate discontinuities, ISOR calculates difficulty through four continuous physiological and mathematical pillars:
Pool Adjacent Violators monotonic calibration on difficulty frontier anchors, guaranteeing strictly monotonic tier progression across rate changes.
Models physical finger stamina and muscular fatigue across dual-decay biomechanical streams combined with Shannon column transition entropy.
Continuous convex triangulation synthesizing modulated Star Rating, burst choke density, and MinaCalc skillset profiles.
An L2-regularized Ridge meta-layer correcting non-linear residuals over a 98-dimensional feature vector for smooth C⁰ continuous accuracy.
ISOR was measured locally against every public Dan estimation engine using the methodology and map pool of Leo_Black's VSRG DanEstimation Benchmark (osu!mania 4K rice). It holds the highest exact-tier rate of any estimator and answers every map on the scale; Koishi remains ahead on mean absolute error, which the tables below show alongside the exact rate.
The core arena — 1103 benchmark maps across tiers 11.0 to 17.0, every engine measured on the maps it answers. Lower MAE is better; the exact-rate column is the one ISOR leads.
| Rank | Engine | Coverage | MAE | RMSE | Exact ≤0.20 | Close ≤0.50 |
|---|---|---|---|---|---|---|
| #1 | Koishi Partial scope | 1096 / 1103 (99.4%) | 0.1510 | 0.1996 | 71.53% | 97.99% |
| #2 | Mixed | 1103 / 1103 (100.0%) | 0.1699 | 0.2275 | 69.72% | 96.01% |
| #3 | XDC1 Partial scope | 1102 / 1103 (99.9%) | 0.1717 | 0.2314 | 67.97% | 95.74% |
| #4 | ISOR (DanOverlay v3.2) | 1103 / 1103 (100.0%) | 0.1722 | 0.2499 | 71.89% | 95.38% |
| #5 | ROXY (Leo_Black) Partial scope | 1096 / 1103 (99.4%) | 0.1980 | 0.2740 | 63.78% | 94.98% |
| #6 | Azusa Partial scope | 1101 / 1103 (99.8%) | 0.2183 | 0.3085 | 59.13% | 93.28% |
| #7 | Companella | 1103 / 1103 (100.0%) | 0.3666 | 0.5061 | 38.98% | 76.97% |
| #8 | Daniel Partial scope | 1099 / 1103 (99.6%) | 0.4088 | 0.4809 | 23.38% | 67.52% |
| #9 | Sunny (Native) | 1103 / 1103 (100.0%) | 0.5156 | 0.6797 | 36.17% | 57.21% |
Probability density histogram of delta errors on the High Rice scale (Tiers 11–17). ISOR lands 793 of the 1103 maps inside the exact window (71.9%, MAE 0.1722, σ = 0.2499); its residuals cluster tightly around zero with a thinner tail than the field.
Direct comparison of Mean Absolute Error across mechanical archetypes on the core benchmark pool. Lower MAE is better.
Full coverage run — 1262 maps across the entire rice scale, tiers 1.0 to 18.0.
| Rank | Engine | Coverage | MAE | RMSE | Exact ≤0.20 | Close ≤0.50 |
|---|---|---|---|---|---|---|
| #1 | Koishi Partial scope | 1106 / 1262 (87.6%) | 0.1540 | 0.2046 | 70.89% | 97.47% |
| #2 | XDC1 Partial scope | 1144 / 1262 (90.6%) | 0.1803 | 0.2498 | 66.87% | 94.67% |
| #3 | ISOR (DanOverlay v3.2) | 1262 / 1262 (100.0%) | 0.1962 | 0.2875 | 67.83% | 92.63% |
| #4 | Mixed | 1262 / 1262 (100.0%) | 0.2075 | 0.3020 | 65.06% | 91.20% |
| #5 | ROXY (Leo_Black) Partial scope | 1119 / 1262 (88.7%) | 0.2093 | 0.2966 | 62.65% | 93.66% |
| #6 | Azusa Partial scope | 1260 / 1262 (99.8%) | 0.2555 | 0.3713 | 55.32% | 88.33% |
| #7 | Daniel Partial scope | 1140 / 1262 (90.3%) | 0.4216 | 0.5012 | 22.72% | 66.40% |
| #8 | Companella | 1262 / 1262 (100.0%) | 0.4276 | 0.6408 | 37.08% | 73.06% |
| #9 | Sunny (Native) | 1262 / 1262 (100.0%) | 0.5563 | 0.7452 | 35.10% | 55.07% |
Daniel and ROXY run on a partial scope of the full-scale pool (81.1% and 78.0% coverage respectively), so their numbers are not directly comparable to the 100%-coverage engines. ISOR estimates every map on the scale.
All 1262 evaluated rice beatmaps plotted against the ideal identity line (y = x). Shaded corridors highlight exact fit (≤0.20 Dan) and close fit (≤0.50 Dan) tolerance bands. Hover any point to inspect full beatmap details.
The 60 verified canon maps of the official Dan Reform course packs (DDMythical, Emik03, Thaumiel, Cloverwisp) — held out from calibration.
ISOR does not lead on this scope, and the cards are ordered to show it.
On the 60 canon maps ISOR scores 0.3213 MAE; ROXY reaches 0.2435 on the 46 it answers, and Koishi
0.2253 on the 48 it answers — on those same 48 ISOR scores 0.2844. ROXY (46/60) and Daniel (51/60)
answer only part of the canon, so their figures rest on a subset rather than the full pool, and
Koishi files no result for this 60-map selection at all, which is why its card reports the wider
131-map official pack set.
The reason is structural: these are official dan courses sitting mostly in the high tiers, largely
outside the rice model's gate, so the physics projection answers alone here. It is the engine's
weakest scope and the page shows it as such rather than dropping the tab.
Evaluation transparency & reproducibility: these benchmark metrics originate from independent local testing conducted to evaluate estimation accuracy, strictly adhering to the methodology and map pool of Leo_Black's VSRG DanEstimation Benchmark.
For 100% reproducible transparency, every single raw prediction row across all 1,364 benchmark beatmaps is publicly auditable and verifiable:
Zero configuration required. Select your game below for quick setup instructions:
Open osu! stable or osu! lazer and navigate to song select.
A lightweight background reader for live osu! game memory.
Run DanOverlay.exe — auto-connects to tosu instantly.
Open DanOverlay.exe — auto-detects active Etterna themes.
Automatically bridges Til Death, Rebirth, Simply Love, and fallback.
Estimates Dan rank, MSD profiles, and Wife% grade badges in real time.
The overlay ships with six distinct visual designs — three originals and three new in v2.0. Pick the one that fits your stream layout.
Original




New in v2.0
.png)
.png)
.png)
Click the overlay window and press G to generate a full-resolution density graph — complete NPS timeline with per-skillset MSD bars. Saved to your Desktop automatically.
Straight answer: it works well most of the time, and then occasionally it'll confidently call your Zeta-tier speedjack map "Delta High" and act completely normal about it.
Anything with unusual density, hybrid patterns, or charts between two skillsets — the classifier can misfire. Sometimes badly. I hope the upcoming rescues and corrections will help, but honestly some maps are just hard to classify.
If it reads your Eta stamina marathon as "Alpha LOW" — that's not a glitch, it's just the algorithm having a terrible moment. No crash, no warning.
Every bad call gets investigated. The goal is always a structural fix — not a special case for one map, but a change that handles that class of chart correctly.
It's like asking a friend who plays a ton of osu!mania to size up any map on the spot. Knows the Reform pack well, has strong opinions, solid gut reads — and occasionally says something completely unhinged. You still ask anyway.
osu!mania
Etterna
v3.2 · Standalone .exe (71.37 MB) · Zero Installation Required
ISOR 3.2 — The Rice Model
config/rice_model.json and evaluated in pure Python with zero ML runtime dependencies.
[12.0, 15.0) DP where out-of-fold validation holds, leaving calibrated physics outside the
gate. Reaches 71.89% exact rate (±0.20) on Tiers 11–17 (1,103 maps, #1 of
9 estimators) and 67.83% exact / 0.1962 MAE out-of-fold across all 1,262 rice maps with
100% coverage.
src/08_isor_engine/ISOR.csv alongside full methodology & caveats in
src/08_isor_engine/README.md, plus a standalone matplotlib chart generator.
Three New Skins & Unified Window Sizing
config/skin_dimensions.json with exact client-area Win32 sizing
(AdjustWindowRectEx). Every skin now opens in its native collapsed/default layout with zero
overflow or leftover space.Half the Size & Lighter Runtime Footprint
pandas, tzdata, scipy, PyQt6, and
dev packages, and compiling bytecode with --optimize 1.tools/measure_overlay_memory.ps1 to audit real runtime footprint.
New Game Support — Native Etterna Integration
Save/ directory..dan_backup) — zero manual setup required under standard configurations.
.sm and .ssc files with full support for multiple BPM shifts,
STOPS, DELAYS, split timings, and hold/roll mechanics in under 15 ms.
ScreenEvaluation and renders official Etterna grade
badges (AAAAA, AAAA, AAA, AA, A,
B, C, D, F) with authentic theme color accents.
Core Engine Performance & Vectorization
extract_choke_and_local_strains from O(W · log N) binary
searches to an O(W + N) two-pointer monotonic sliding window.
New Engine — ISOR (Isotonic Strain Organic Residual)
Estimation Modes & Celestial Recalibration
None for out-of-scope maps — never a fake low rating.
Bug Fixes & Robustness
sys._MEIPASS/config/ in PyInstaller builds.
rhythm_profile and strain modules
are bundled correctly.dp=None instead
of a fake 1.00 / low-tier result.New Streamer Skin — "Josh type shit" (UI-9)
--cfg-blur) and brightness (--cfg-brightness) controls in Settings.#00ff00 background mode with
automatic suppression of text blur and halos for clean OBS keying.Window Sizing & Core Fixes
SetWindowPos execution
in the host bridge to guarantee instantaneous, 100% reliable window resizing on Windows.Custom-Rate Interpolation & HT Fixes (osu!lazer)
Skin 7 (Sunny Rebirth) & Skin 8 (Hero-Dan Lazer) Visual Polish
rgba(10, 12, 18, 0.84) + blur(12px)), crisp borders, and
drop shadow for maximum readability over background animation.left / total).New Skins — Sunny Rebirth & Hero-Dan Lazer
Sunny Strain Components Engine
Pre-Open-Source Calculation Fixes
analyze_map now interpolates
results across nearest native clock rates (0.75× / 1.0× / 1.5×), ensuring strictly monotonic Dan
estimates across any custom speed multiplier.UI & System Bug Fixes
ui.panel was null on overlays without
#danPanel.
vh units with
percentage height chains.Calculation Pipeline & Detection Fixes
float() conversion, returning star rating 0.0. This caused maps to display as "1st Dan Low"
regardless of actual difficulty. The fix skips empty lines during parsing, restoring correct detection
on all maps.
Bug Fixes
Performance Optimisation
(well, on my shitty pc I didn't notice a difference in calculation speed but in theory it should work lol)
osu!mania 7K — Official Support
Three New Skins
Celestial Mode — DP Scale
Signicial Mode Improvements
General
New Features
Calibration
Feedback
Calibration & Fixes
New Features
Bug Fixes
Modern skin — the new design
Settings Persistence
%APPDATA%\DanOverlay\settings.json — no need to reconfigure after
updates.Visual Customization
Startup & Onboarding
Bug Fixes
Scoring Modes
LN Course — Automatic Detection
Beyond Ceiling
Resize & Controls
Intro & UI
Compatibility