Validation dossier

Perf-Hub's calculation engines compared against public industry benchmarks — the numbers in the “Perf-Hub” column are computed NOW, in this browser, by the same code the app runs. Nothing is transcribed.

Anti-collision — ISCWSA error model (MWD Rev. 4)

PASS(max deviation 0.68%)

Reproduction of the official ISCWSA “Error Model Example File 1” (test well 1: vertical → 2°/30 m build to 60° → tangent → build to 90° → horizontal, TD 8,000 m, lat 60° N). The full 28+ term engine with the official weighting functions computes NEV position covariances and compares them against the public spreadsheet's “Totals” rows, component by component, at 4 checkpoints.

Official source: ISCWSA validation spreadsheet (iscwsa.net), via the public dataset of the open-source welleng project.

MD (m)	Component	Official (m², 1σ)	Perf-Hub (m², 1σ)	Deviation
1200	nn	4.5111	4.5111	< 0.001%
1200	ee	4.5111	4.5111	< 0.001%
1200	vv	0.7037	0.7037	-0.00%
1200	ne	0.0000	0.0000	≈ 0 ✓
1200	nv	0.0000	0.0000	≈ 0 ✓
1200	ev	0.0000	0.0000	≈ 0 ✓
2100	nn	43.0442	42.8961	-0.34%
2100	ee	14.4041	14.5017	+0.68%
2100	vv	3.4530	3.4530	-0.00%
2100	ne	-7.9409	-7.7358	-2.58%
2100	nv	-0.8506	-0.8029	-5.61%
2100	ev	-2.4206	-2.4827	+2.57%
5100	nn	2102.8867	2101.5497	-0.06%
5100	ee	217.9769	217.9642	-0.01%
5100	vv	116.9514	116.9514	< 0.001%
5100	ne	-538.1731	-537.1587	-0.19%
5100	nv	-23.5877	-23.3479	-1.02%
5100	ev	-60.6667	-60.9793	+0.52%
8000	nn	8547.7769	8545.7732	-0.02%
8000	ee	712.6932	712.7925	+0.01%
8000	vv	468.1270	468.1270	< 0.001%
8000	ne	-2252.3087	-2251.3987	-0.04%
8000	nv	-52.5280	-52.2775	-0.48%
8000	ev	-123.1390	-123.4655	+0.27%

Tolerance: ±1.5% on variances (nn/ee/vv); ±7% on cross-covariances (ne/nv/ev — small values, more sensitive to fixture geometry). In the vertical section the cross terms are zero by symmetry.

Torque & Drag — Johancsik (1984), SPE 11380-PA

PASS(max deviation < 0.001%)

Build-and-hold reference well (1,000 m vertical → 3°/30 m build to 60° → tangent to TD 3,000 m; 5″ 19.5 lb/ft drill pipe, μ = 0.30, 10 ppg fluid). In the straight sections the soft-string model has an exact analytic solution — the engine is compared to the closed form, with no room for tuning.

Anchor	Closed form	Analytic	Perf-Hub	Deviation
Tension at top of tangent — trip-out	W·(cos θ + μ·sin θ)	57.664 kips	57.664 kips	< 0.001%
Tension at top of tangent — trip-in	W·(cos θ − μ·sin θ)	18.229 kips	18.229 kips	< 0.001%
Tension at top of tangent — rotating	W·cos θ	37.947 kips	37.947 kips	< 0.001%
Torque at top of tangent — rotating	μ·W·sin θ·r	4.108 klbf·ft	4.108 klbf·ft	< 0.001%
Vertical section — Δtension (friction-free)	ΔF = W_vert (μ não contribui)	54.210 kips	54.210 kips	< 0.001%

Tolerance: ±0.1% (measured agreement is on the order of 1e-6). The curved section is verified in CI against an independent re-implementation of the paper's recursion, station by station.

Hole cleaning — Larsen (SPE-25872-PA)

PASS(13/13)

The production engine (larsenCTFV/larsenBed) runs here in the browser with the paper's Appendix A inputs and is compared against the 13 published quantities (CTV, μa, correction factors, CTFV, Qcrit and bed concentration).

Source: Larsen, Pilehvari & Azar, SPE Drilling & Completion, June 1997 — Appendix A (mud 1, 110 gpm, 65°, 5" × 2.375").

Quantity	Official (m², 1σ)	Perf-Hub (m², 1σ)	Deviation
Cconc	1.470 %	1.465 %	\|Δ\| 0.0049 (tol 0.01)
Vcut	1.320 ft/s	1.322 ft/s	\|Δ\| 0.0021 (tol 0.01)
μa	27.60 cP	27.57 cP	\|Δ\| 0.032 (tol 0.2)
V̄slip	3.150 ft/s	3.148 ft/s	\|Δ\| 0.0017 (tol 0.02)
Cang	1.026 —	1.026 —	\|Δ\| 0.00043 (tol 0.002)
Csize	1.100 —	1.104 —	\|Δ\| 0.0040 (tol 0.005)
Cmwt	1.000 —	1.000 —	\|Δ\| 0.0 (tol 0)
Vslip	3.560 ft/s	3.565 ft/s	\|Δ\| 0.0046 (tol 0.02)
Vcrit	4.880 ft/s	4.887 ft/s	\|Δ\| 0.0067 (tol 0.03)
Aann	0.1056 ft²	0.1056 ft²	\|Δ\| 9.9e-7 (tol 0.0001)
Qcrit	231.0 gpm	231.6 gpm	\|Δ\| 0.59 (tol 2)
C̄bconc	33.50 %	33.60 %	\|Δ\| 0.10 (tol 0.3)
Cbconc	30.40 %	30.45 %	\|Δ\| 0.053 (tol 0.3)

Absolute per-quantity tolerances (paper rounding); the Computed column comes from the live engine, never transcribed.

Method and reproducibility

These same comparisons run as regression tests in the project's 700+ test suite, in CI, on every code change — the page cannot show “pass” with a regressed engine.
Beyond the benchmarks above, the solvers are anchored to independent closed forms: catenary well (zero contact via T·|κ| = w·sinθ), Dawson-Paslay buckling, Miner fatigue and minimum curvature against hand-worked cases.
Documented limitations: the ISCWSA residual comes from fixture geometry (our minimum curvature × the spreadsheet's stations), not from the weighting functions; the production path uses Rev. 5.11 (Rev.4→Rev.5 differences documented in code).