Wellbore Genius

Downhole Simulator

Performance modeling

v1.0Solver: OK

Phase 6 inc. 3 · DFIT benchmarks

DFIT analytical benchmark suite ALL PASSED

Live view of the same suite that runs on every CI build. 3 benchmarks compared on a fixed log-spaced time grid (1 → 1024 s) with a 0.05 log-tolerance ceiling. Drift past tolerance = scoreboard flip required.

Per-benchmark results

Benchmark	Status	RMS log-residual	Max log-residual	Tolerance	N samples
kgd-width	PASS	7.02e-2	7.56e-2	1.15e-1	2
pkn-length	PASS	1.33e-1	1.80e-1	1.50e-1	2
carter-falloff	PASS	6.95e-8	9.82e-8	1.00e-5	2

Compare against a previous run

Upload a previously downloaded dfit-benchmarks-*.json to diff its PASS/FAIL + residuals against the current run above.

No baseline loaded yet.

Residuals over time

Shaded band = ±tolerance. Red markers = samples outside the band.

kgd-widthcoupledNewtonStep + MK moving-tip + reinit i=8 (Phase 7 inc 15) · live kernel

tol ±0.12 · max |log| = 7.56e-2ALL IN BAND

pkn-lengthcoupledNewtonStep (Phase 3.5 kernel, height-contained) · live kernel

tol ±0.15 · max |log| = 1.80e-11 OUT OF BAND

carter-falloffcoupledNewtonStep + Carter sink mass-balance recovery (Phase 7 inc 14) · live kernel

tol ±0.00 · max |log| = 9.82e-8ALL IN BAND

KGD fixed-tip vs moving-tip [m, log₁₀ residual]

Side-by-side widths on the canonical [1, 2] s window against the Spence–Sharp analytical reference. Moving-tip uses the MK-universal wrapper (Phase 7 inc 13); fixed-tip is the legacy strip Newton (Phase 6 inc 4). RMS is √(mean(log₁₀(w_run / w_ref)²)).

Show comparison

Toggle on to recompute both runners on the canonical inputs.

Window sweep (preview)

Compares the registry-locked window against a wider preview window per live-kernel entry. Doesn't touch the CI gate or the kernel.

Quick presets:

t_start [s]

t_end [s]

n_samples [—]

mesh adapt [—]

tip BC [—]

Mesh adapt & tip-BC are wired through to each runner: PKN uses adaptiveTipFocus + cohesiveTipBC, KGD uses bumped reinit iterations + cohesiveTipBC, Carter doubles cell count for tip-refined (no propagating tip — tip-BC echoed for traceability). Defaults preserve the registry rms byte-for-byte.

Registry: [1, 2] s·Preview: [1, 4] s, 3 pts

rms_log10 vs window

Sweeps t_end across the requested range with t_start fixed and plots the rms log10 residual returned by the active runner.

t_start [s]

t_end_min [s]

t_end_max [s]

n_steps [—]

Sensitivity sweep — Q · t-window · meshAdapt · tipBC

REGRESSIONS

Sweeps the registry-locked window plus a tighter / broader neighbour for each entry, across meshAdapt (uniform / tip-refined), tipBC (free / kic-clamped), and Carter Q sizing (constant / auto). Headroom (Δ = tolerance − rms) ≥ 0 ⇒ healthy. Default filter shows all 48 rows.

KGD width

4/12 pass

tol [log]: 1.150e-1

best Δ 4.477e-2 · worst Δ -1.282e-1

worst @ [1, 4]s · tip-refined · free

PKN length

2/12 pass

tol [log]: 1.500e-1

best Δ 1.657e-2 · worst Δ -3.846e-1

worst @ [0.25, 0.5]s · tip-refined · kic-clamped

Carter falloff (leak-off)

24/24 pass

tol [log]: 1.000e-5

best Δ 9.980e-6 · worst Δ 6.406e-6

worst @ [2, 10]s · tip-refined · kic-clamped · Q=auto

Filter:

meshAdapt:

tipBC:

Q (carter):

48 / 48 rows

Entry	Window [s]	meshAdapt	tipBC	Q	rms [log]	tol [log]	Δ [log]	Status
kgd-width	[1, 2] · n=2	uniform	free	—	7.023e-2	1.150e-1	+4.477e-2	PASS
kgd-width	[0.5, 1] · n=2	uniform	free	—	1.908e-1	1.150e-1	-7.576e-2	FAIL
kgd-width	[1, 4] · n=2	uniform	free	—	2.432e-1	1.150e-1	-1.282e-1	FAIL
kgd-width	[1, 2] · n=2	uniform	kic-clamped	—	1.070e-1	1.150e-1	+7.968e-3	PASS
kgd-width	[0.5, 1] · n=2	uniform	kic-clamped	—	1.999e-1	1.150e-1	-8.489e-2	FAIL
kgd-width	[1, 4] · n=2	uniform	kic-clamped	—	2.427e-1	1.150e-1	-1.277e-1	FAIL
kgd-width	[1, 2] · n=2	tip-refined	free	—	7.023e-2	1.150e-1	+4.477e-2	PASS
kgd-width	[0.5, 1] · n=2	tip-refined	free	—	1.908e-1	1.150e-1	-7.576e-2	FAIL
kgd-width	[1, 4] · n=2	tip-refined	free	—	2.432e-1	1.150e-1	-1.282e-1	FAIL
kgd-width	[1, 2] · n=2	tip-refined	kic-clamped	—	1.070e-1	1.150e-1	+7.968e-3	PASS
kgd-width	[0.5, 1] · n=2	tip-refined	kic-clamped	—	1.999e-1	1.150e-1	-8.489e-2	FAIL
kgd-width	[1, 4] · n=2	tip-refined	kic-clamped	—	2.427e-1	1.150e-1	-1.277e-1	FAIL
pkn-length	[0.5, 1] · n=2	uniform	free	—	1.334e-1	1.500e-1	+1.657e-2	PASS
pkn-length	[0.25, 0.5] · n=2	uniform	free	—	3.176e-1	1.500e-1	-1.676e-1	FAIL
pkn-length	[1, 2] · n=2	uniform	free	—	2.113e-1	1.500e-1	-6.130e-2	FAIL
pkn-length	[0.5, 1] · n=2	uniform	kic-clamped	—	3.368e-1	1.500e-1	-1.868e-1	FAIL
pkn-length	[0.25, 0.5] · n=2	uniform	kic-clamped	—	5.346e-1	1.500e-1	-3.846e-1	FAIL
pkn-length	[1, 2] · n=2	uniform	kic-clamped	—	2.081e-1	1.500e-1	-5.813e-2	FAIL
pkn-length	[0.5, 1] · n=2	tip-refined	free	—	1.334e-1	1.500e-1	+1.657e-2	PASS
pkn-length	[0.25, 0.5] · n=2	tip-refined	free	—	3.176e-1	1.500e-1	-1.676e-1	FAIL
pkn-length	[1, 2] · n=2	tip-refined	free	—	2.113e-1	1.500e-1	-6.130e-2	FAIL
pkn-length	[0.5, 1] · n=2	tip-refined	kic-clamped	—	3.368e-1	1.500e-1	-1.868e-1	FAIL
pkn-length	[0.25, 0.5] · n=2	tip-refined	kic-clamped	—	5.346e-1	1.500e-1	-3.846e-1	FAIL
pkn-length	[1, 2] · n=2	tip-refined	kic-clamped	—	2.081e-1	1.500e-1	-5.813e-2	FAIL
carter-falloff	[1, 2] · n=2	uniform	free	constant	6.947e-8	1.000e-5	+9.931e-6	PASS
carter-falloff	[0.5, 1] · n=2	uniform	free	constant	2.001e-8	1.000e-5	+9.980e-6	PASS
carter-falloff	[2, 10] · n=2	uniform	free	constant	9.980e-7	1.000e-5	+9.002e-6	PASS
carter-falloff	[1, 2] · n=2	uniform	free	auto	6.947e-8	1.000e-5	+9.931e-6	PASS
carter-falloff	[0.5, 1] · n=2	uniform	free	auto	2.001e-8	1.000e-5	+9.980e-6	PASS
carter-falloff	[2, 10] · n=2	uniform	free	auto	9.980e-7	1.000e-5	+9.002e-6	PASS
carter-falloff	[1, 2] · n=2	uniform	kic-clamped	constant	6.947e-8	1.000e-5	+9.931e-6	PASS
carter-falloff	[0.5, 1] · n=2	uniform	kic-clamped	constant	2.001e-8	1.000e-5	+9.980e-6	PASS
carter-falloff	[2, 10] · n=2	uniform	kic-clamped	constant	9.980e-7	1.000e-5	+9.002e-6	PASS
carter-falloff	[1, 2] · n=2	uniform	kic-clamped	auto	6.947e-8	1.000e-5	+9.931e-6	PASS
carter-falloff	[0.5, 1] · n=2	uniform	kic-clamped	auto	2.001e-8	1.000e-5	+9.980e-6	PASS
carter-falloff	[2, 10] · n=2	uniform	kic-clamped	auto	9.980e-7	1.000e-5	+9.002e-6	PASS
carter-falloff	[1, 2] · n=2	tip-refined	free	constant	2.482e-7	1.000e-5	+9.752e-6	PASS
carter-falloff	[0.5, 1] · n=2	tip-refined	free	constant	7.113e-8	1.000e-5	+9.929e-6	PASS
carter-falloff	[2, 10] · n=2	tip-refined	free	constant	3.594e-6	1.000e-5	+6.406e-6	PASS
carter-falloff	[1, 2] · n=2	tip-refined	free	auto	2.482e-7	1.000e-5	+9.752e-6	PASS
carter-falloff	[0.5, 1] · n=2	tip-refined	free	auto	7.113e-8	1.000e-5	+9.929e-6	PASS
carter-falloff	[2, 10] · n=2	tip-refined	free	auto	3.594e-6	1.000e-5	+6.406e-6	PASS
carter-falloff	[1, 2] · n=2	tip-refined	kic-clamped	constant	2.482e-7	1.000e-5	+9.752e-6	PASS
carter-falloff	[0.5, 1] · n=2	tip-refined	kic-clamped	constant	7.113e-8	1.000e-5	+9.929e-6	PASS
carter-falloff	[2, 10] · n=2	tip-refined	kic-clamped	constant	3.594e-6	1.000e-5	+6.406e-6	PASS
carter-falloff	[1, 2] · n=2	tip-refined	kic-clamped	auto	2.482e-7	1.000e-5	+9.752e-6	PASS
carter-falloff	[0.5, 1] · n=2	tip-refined	kic-clamped	auto	7.113e-8	1.000e-5	+9.929e-6	PASS
carter-falloff	[2, 10] · n=2	tip-refined	kic-clamped	auto	3.594e-6	1.000e-5	+6.406e-6	PASS

A/B comparison (preview)

Runs two solver settings on the same window range and plots rms_log10 / max_log10 deltas (B − A). Doesn't touch the CI gate.

mesh adapt [—]

tip BC [—]

carter sizing [—]

mesh adapt [—]

tip BC [—]

carter sizing [—]

t_start [s]

t_end_min [s]

t_end_max [s]

n_steps [—]

Registry-window comparison

Scores every live-kernel entry on both registry windows ([1, 2] s and [0.5, 1] s) under the same solver settings. Use it to evaluate the meshAdapt / tipBC tightening work before flipping the scoreboard. Doesn't touch the CI gate or the kernel.

Mesh adapt

Tip BC

Carter sizing

Entry	Registry	rms [1, 2] s	rms [0.5, 1] s	Δ rms (tight − wide)	max [1, 2] s	max [0.5, 1] s
KGD width	[1, 2] s	7.0234e-2	1.9076e-1	1.2052e-1	7.5640e-2	2.6220e-1
PKN length	[0.5, 1] s	2.1130e-1	1.3343e-1	-7.7875e-2	2.9365e-1	1.8038e-1
Carter falloff	[1, 2] s	6.9466e-8	2.0013e-8	-4.9453e-8	9.8172e-8	2.8212e-8

Compare saved runs

Pin the current density / tolerance configuration as a snapshot and view PASS/FAIL + residual charts side-by-side.

No saved runs yet. Adjust the controls above and hit Save current run.

CI log preview

DFIT benchmark suite @ 2026-05-24T14:33:06.036Z — PASS (3 benchmarks)
  [PASS] kgd-width        rms=7.02e-2 max=7.56e-2 tol=1.15e-1 n=2
  [PASS] pkn-length       rms=1.33e-1 max=1.80e-1 tol=1.50e-1 n=2
  [PASS] carter-falloff   rms=6.95e-8 max=9.82e-8 tol=1.00e-5 n=2

Truncation diagnostic — width & half-length over time

legacy vs mkTipFeedback vs reinit, per KGD benchmark

Benchmark curves (1/1)

site-kgd-canonical

Apply preset:

Import JSON

legacymkTipFeedbackreinit

Width w(t) [m]

Half-length a(t) [m]

Wellbore pressure p(t) [Pa]

peak |Δw| mk−legacy: 6.51e-4 m · peak |Δw| reinit−legacy: 6.51e-4 m · peak |Δp| mk−legacy: 3.54e+7 Pa · peak |Δp| reinit−legacy: 3.54e+7 Pa · rows: 2

Temperature T(t): — KGD kernel is isothermal; row schema reserves T columns for future thermal-coupled kernels.

Registry

kgd-width · tol 0.115 (log10) · runner: coupledNewtonStep + MK moving-tip + reinit i=8 (Phase 7 inc 15) (live kernel)
pkn-length · tol 0.15 (log10) · runner: coupledNewtonStep (Phase 3.5 kernel, height-contained) (live kernel)
carter-falloff · tol 0.00001 (log10) · runner: coupledNewtonStep + Carter sink mass-balance recovery (Phase 7 inc 14) (live kernel)

Phase 6 inc 4 wired the kgd-width entry into the live coupledNewtonStep kernel (1D strip, 41 cells, [1, 2] s window). PKN + Carter still self-reference until their kernels land.

Timestamp: 2026-05-24T14:33:06.036Z

μ × dt solver diagnostics

Convergence status, residual trend, and failure-mode label for every (μ [cp], dt [d]) cell in the viscosity sweep grid.

MetricApply dt cap (μ ≥ 2 ⇒ dt ≤ 20 d)

Pass22Solver tol.0Transient0Both0Skipped (dt cap)2

Tuning presetsCaptures dt cap, dt levels, per-μ caps & metric — rerun with one click.

Load|

μ [cp] \ dt [d]	0.2	0.5	1	5	20	100
0.5	Pass	Pass	Pass	Pass	Pass	Pass
1	Pass	Pass	Pass	Pass	Pass	Pass
2	Pass	Pass	Pass	Pass	Pass	Skipped (dt cap) —
5	Pass	Pass	Pass	Pass	Pass	Skipped (dt cap) —

Sparklines show per-step CG residual on a log scale. Failure-mode labels come from classifyFailureMode with default thresholds (residualTol = 1e-6, settleRatioTol = 0.05).

Drift-guard tolerance-margin heatmap [μ × dt]

Color = 1 − |Δ| / envelope. Green = envelope sits well above observed deviation; amber = tight; red = envelope violated.

MetricApply dt cap (μ ≥ 2 ⇒ dt ≤ 20 d)

Safe22Comfortable0Tight0Failing0Skipped2

μ [cp] \ dt [d]	0.2	0.5	1	5	20	100
0.5	1.001.00e+3 psi	1.001.00e+3 psi	1.001.00e+3 psi	1.001.00e+3 psi	1.009.99e+2 psi	1.009.99e+2 psi
1	1.002.00e+3 psi	1.002.00e+3 psi	1.002.00e+3 psi	1.002.00e+3 psi	1.002.00e+3 psi	1.002.00e+3 psi
2	1.004.00e+3 psi	1.004.00e+3 psi	1.004.00e+3 psi	1.004.00e+3 psi	1.004.00e+3 psi	—skip
5	1.001.00e+4 psi	1.001.00e+4 psi	1.001.00e+4 psi	1.001.00e+4 psi	1.001.00e+4 psi	—skip

Envelope per μ comes from tolerancesFor({ viscosityCp: μ }).pInitEnvelopePsi. Observed |Δ| uses the synthetic probe'smaxAbsStepDelta from the diagnostics recorder — same quantity the failure classifier inspects.

Pass/fail matrices — capped vs uncapped (with headroom)

Tightens the analytical envelope to E / (1 + headroomFraction) and recolors both matrices live. Use it to ask: "what still fails if I require an extra X% buffer?"

Metric

Headroom fraction [%]0%

Effective envelope per μ = tolerancesFor(μ).pInitEnvelopePsi / 1.00. Capped fails: 0 · Uncapped fails: 0.

Downloads

Wide / long / fail-frequency CSVs for both capped & uncapped grids at the current metric and headroom.

Capped (μ ≥ 2 ⇒ dt ≤ 20 d)

0 fail · 0 tight · 22 safe · 2 skip

μ \ dt	0.2	0.5	1	5	20	100
0.5	✓	✓	✓	✓	✓	✓
1	✓	✓	✓	✓	✓	✓
2	✓	✓	✓	✓	✓	—
5	✓	✓	✓	✓	✓	—

Uncapped (full dt grid)

0 fail · 0 tight · 24 safe · 0 skip

μ \ dt	0.2	0.5	1	5	20	100
0.5	✓	✓	✓	✓	✓	✓
1	✓	✓	✓	✓	✓	✓
2	✓	✓	✓	✓	✓	✓
5	✓	✓	✓	✓	✓	✓

✓ = pass / safe · ! = tight · ✗ = envelope violated · — = skipped by dt cap. Cell color follows the same diverging palette as the tolerance-margin heatmap above.

Failure-prone heatmap [μ × dt] — worst of all metrics

Color = failed metrics ÷ ran metrics at this (μ, dt) acrossp_wf·p_cell·p_probe. Top-N most failure-prone cells get a ranked ring.

Pass66Fail0Skip6

Apply dt cap (μ ≥ 2 ⇒ dt ≤ 20 d)Headroom [%]0%Highlight top-N3

μ [cp] \ dt [d]	0.2	0.5	1	5	20	100
0.5
1
2
5

Sweep results summary — what changed

Diffs current settings against the baseline (capped grid, headroom = 0). Lists re-enabled cells, newly failing cells, recoveries, and outstanding regressions, plus the drift-guard tests they gate.

MetricCurrent grid uses dt capCurrent headroom [%]10%

Re-enabled2Newly failing0Recovered0Still failing0

Change	μ [cp]	dt [d]	Baseline	Current	Δ headroom [psi]
Re-enabled	2	100	skipped	safe	—
Re-enabled	5	100	skipped	safe	—

Drift-guard tests affected by the visco sweep

Re-run these vitest files before merging any envelope or sweep change. Each one snapshots a piece of the (μ, dt) plumbing surfaced on this page.

src/lib/reservoir/__tests__/viscoSweepMatrixCsv.test.ts— WIDE / LONG / fail-frequency CSV builders.
src/lib/reservoir/__tests__/viscoToleranceMargin.test.ts— Envelope vs deviation status bucketing.
src/lib/reservoir/__tests__/viscoDiagCsv.test.ts— Per-(μ, dt) diagnostics grid CSV.
src/lib/reservoir/__tests__/viscoDiagFailingCsv.test.ts— Failing-only CSV (rank by worst headroom).
src/lib/reservoir/__tests__/viscoSolverDiagnosticsDemo.test.ts— Synthetic μ × dt grid that drives the page.
src/lib/__tests__/adaptiveEnvelopeCalibration.test.ts— Per-μ envelope re-calibration from observed deviations.
src/lib/__tests__/perPointMultiplierSearch.test.ts— Per-(μ, dt) multiplier lift search.

Quick command: bunx vitest run src/lib/reservoir/__tests__/viscoSweepMatrixCsv.test.ts src/lib/reservoir/__tests__/viscoToleranceMargin.test.ts src/lib/reservoir/__tests__/viscoDiagCsv.test.ts src/lib/reservoir/__tests__/viscoDiagFailingCsv.test.ts src/lib/reservoir/__tests__/viscoSolverDiagnosticsDemo.test.ts src/lib/__tests__/adaptiveEnvelopeCalibration.test.ts src/lib/__tests__/perPointMultiplierSearch.test.ts

Per-(μ, dt) adaptive multiplier search

Lifts envelope ONLY at failing cells until |Δ| · (1 + headroomFraction) ≤ E_μ · m. Increase headroom to push failing cells under the cap; capped cells remain genuine envelope blow-outs.

Headroom fraction [%] — 5 %Pass condition: |Δ| · (1 + 0.05) ≤ E_μ · mMax multiplier cap — 8×Required m above this cap ⇒ status = fail (capped).

Pass16Lifted (m > 1)2Capped (fail)0Min headroom for 0 fails0 %

μ [cp]	dt [d]	\|Δ\| [psi]	E_μ [psi]	m	E·m [psi]	headroom [psi]	status
0.5	0.2	0.000	1.250	1.00×	1.250	1.250	pass
0.5	1	0.100	1.250	1.00×	1.250	1.145	pass
0.5	5	0.200	1.250	1.00×	1.250	1.040	pass
0.5	100	0.600	1.250	1.00×	1.250	0.620	pass
1	0.2	0.000	2.500	1.00×	2.500	2.500	pass
1	1	0.100	2.500	1.00×	2.500	2.395	pass
1	5	0.300	2.500	1.00×	2.500	2.185	pass
1	100	1.200	2.500	1.00×	2.500	1.240	pass
2	0.2	0.000	5.000	1.00×	5.000	5.000	pass
2	1	0.100	5.000	1.00×	5.000	4.895	pass
2	5	0.400	5.000	1.00×	5.000	4.580	pass
2	100	12.000	5.000	2.53×	12.650	0.050	lifted
5	0.2	0.000	12.500	1.00×	12.500	12.500	pass
5	1	0.100	12.500	1.00×	12.500	12.395	pass
5	5	0.600	12.500	1.00×	12.500	11.870	pass
5	100	32.000	12.500	2.69×	33.625	0.025	lifted

Pilot grid: 4 viscosities × 4 dt levels (synthetic). Base envelope per μ comes from tolerancesFor({ viscosityCp: μ }).spreadAcrossSweepPsi with the tight preset.

μ × dt pass/fail matrix chart

Interactive view of the chart-ready LONG CSV (buildViscoSweepMatrixLongCsv). Paste your export or load the demo dataset; click a cell to pin its numbers.

Pass24Fail0Skipped0

LONG CSV (viscosity_cp, dt_days, metric, status, deviation_psi, envelope_psi, headroom_psi)

viscosity_cp,dt_days,metric,status,deviation_psi,envelope_psi,headroom_psi
0.50,0.20,p_wf,pass,0.23,1000.00,1049.77
0.50,0.50,p_wf,pass,0.23,1000.00,1049.77
0.50,1.00,p_wf,pass,0.23,1000.00,1049.77
0.50,5.00,p_wf,pass,0.28,1000.00,1049.72
0.50,20.00,p_wf,pass,0.89,1000.00,1049.11
0.50,100.00,p_wf,pass,1.40,1000.00,1048.60
1.00,0.20,p_wf,pass,0.45,2000.00,2099.55
1.00,0.50,p_wf,pass,0.45,2000.00,2099.55
1.00,1.00,p_wf,pass,0.45,2000.00,2099.55
1.00,5.00,p_wf,pass,0.45,2000.00,2099.55
1.00,20.00,p_wf,pass,1.09,2000.00,2098.91
1.00,100.00,p_wf,pass,2.91,2000.00,2097.09
2.00,0.20,p_wf,pass,0.90,4000.00,4199.10
2.00,0.50,p_wf,pass,0.90,4000.00,4199.10
2.00,1.00,p_wf,pass,0.90,4000.00,4199.10
2.00,5.00,p_wf,pass,0.90,4000.00,4199.10
2.00,20.00,p_wf,pass,1.18,4000.00,4198.82
2.00,100.00,p_wf,pass,5.10,4000.00,4194.90
5.00,0.20,p_wf,pass,2.26,10000.00,10497.74
5.00,0.50,p_wf,pass,2.26,10000.00,10497.74
5.00,1.00,p_wf,pass,2.26,10000.00,10497.74
5.00,5.00,p_wf,pass,2.26,10000.00,10497.74
5.00,20.00,p_wf,pass,2.26,10000.00,10497.74
5.00,100.00,p_wf,pass,6.26,10000.00,10493.74

μ [cp] \ dt [d]	0.2	0.5	1	5	20	100
0.5
1
2
5

Color intensity scales with |headroom| (deeper green = more headroom; deeper red = larger overshoot vs envelope). Multi-metric inputs collapse to the worst status per (μ, dt) across the selected metric chips.

Downhole Simulator

DFIT analytical benchmark suite ALL PASSED

Tip-kernel knobs

Local controls

Per-benchmark results

Compare against a previous run

Residuals over time

KGD fixed-tip vs moving-tip [m, log₁₀ residual]

Window sweep (preview)

rms_log10 vs window

A/B comparison (preview)

Registry-window comparison

Compare saved runs

CI log preview

Truncation diagnostic — width & half-length over time

Registry

μ × dt solver diagnostics

Drift-guard tolerance-margin heatmap [μ × dt]

Pass/fail matrices — capped vs uncapped (with headroom)

Downloads

Capped (μ ≥ 2 ⇒ dt ≤ 20 d)

Uncapped (full dt grid)

Failure-prone heatmap [μ × dt] — worst of all metrics

Sweep results summary — what changed

Drift-guard tests affected by the visco sweep

Per-(μ, dt) adaptive multiplier search

μ × dt pass/fail matrix chart