Technical note
Saying "we have anti-collision" tells you nothing. This page says which model, which terms, how the scan works, how the covariances combine, and - just as plainly - what the program will not do. Check it against whatever you run now.
ISCWSA Rev 5.13 propagation for stationary survey tools on a fixed platform. Our own implementation, written from the published definition.
The 27 weighting functions implemented: DREF, DSF, DST, ABXY_TI1, ABXY_TI2, ABZ, ASXY_TI1, ASXY_TI2, ASXY_TI3, ASZ, MBXY_TI1, MBXY_TI2, MBZ, MSXY_TI1, MSXY_TI2, MSXY_TI3, MSZ, AZ, DBH, AMIL, SAGE, XYM1, XYM2, XYM3E, XYM4E, XCLA, XCLH.
The built-in term set is the public ISCWSA A001Mc MWD + SRGM Rev 5 data (Set A, September 2022), carried as published. Its four declination terms root-sum-square to 0.360° and its four drillstring-interference terms to 5,000 deg·nT, as they should. The lumped SRGM terms are split into their un-modelled and observatory parts so they can be correlated between two wells that share them.
Default course tortuosity is 1°/100 ft. The ISCWSA XCL error supplement r16 is followed for the legacy COMPASS tortuosity table, including its 30-foot misalignment floor.
Not implemented: Rev 4 and earlier term sets. Floating-platform depth terms - the riser rows are skipped, so a floating-rig model is not accepted as one. Gyro error models (see section 6).
Four choices, set per well - and per path, so the plan, the as-drilled well, the geosteering copy and every offset can each carry their own:
There is no default and no silent fallback. With nothing chosen the program raises a blocker naming what is missing. You also have to tick a box confirming the model applies to this well and its survey history before the setup will save.
Two steps.
Closest approach. The exact closest point on each finite minimum-curvature arc, solved analytically, endpoints included. Offsets given as coordinates only, with no inclination and azimuth, use straight segments between the imported points.
Minimum separation factor. From that point the program searches along the offset for the lowest separation factor - a window of three times the centre distance, held between 30 m and 500 m, sampled 48 times and then bisected down to 5 cm. When the minimum sits away from the closest centre point, it says so.
Offset trajectories are interpolated by minimum curvature; the uncertainty is interpolated between the two bracketing stations and never extrapolated past the ends of the profile. A comparison point more than 5 cm off the configured path is refused outright.
What this is not: a travelling-cylinder plot, a ladder plot, or an all-stations-by-all-offsets minimum-separation table. The live scan is one reference station at a time - the survey you have selected, or the point you are following in 3D.
In the north / east / vertical-down frame, in square metres, at one sigma. Weighting functions are evaluated along-hole, highside and lateral and then projected onto that frame. Each station takes half the course either side, in the standard Rev 5 form. Lateral weighting is azimuth weighting times sin(INC), with its finite vertical limit handled.
Four propagation modes, treated properly: random terms add in quadrature station to station; systematic, well-by-well and global terms accumulate as vectors and are squared at the end, so they stay correlated along the hole. Following Rev 5 §4.7.1, systematic terms are independent between survey-tool runs while well and global terms stay correlated across every run in the well. Survey-tool runs must be contiguous and cover the whole history from surface; a gap or an overlap is refused.
Combining two wells: the relative covariance is the reference plus the offset, minus the correlated cross terms for sources the two genuinely share:
Crel = Cref + Coff − ρ( a b⃗ᵀ + b a⃗ᵀ )
Only sources you have explicitly approved for that pair are correlated - never every global term on the sheet. Today that means the geomagnetic terms on the built-in model, at ρ = 1. Starting uncertainty at the tie-in is either a horizontal and vertical one-sigma or a full six-component matrix. A first survey below surface is a blocker unless you confirm a ground-level start.
Ellipsoid axes come from a Jacobi decomposition that refuses a matrix which is not positive semidefinite - a wrong covariance import cannot become a plausible-looking ellipse.
Read from the IPM file itself: short name, revision number and current status. A model marked withdrawn or provisional is refused, with a message telling you to obtain an approved one.
Tools are assigned to depth intervals by a plain From-MD / To-MD / tool-name program, which can be read straight out of an anti-collision report PDF's survey-tool program block, or copied from the reference plan onto the as-drilled well. If two library versions match one name, that is an error, not a coin toss.
An interval with no tool assigned does not get a default. The whole calculation is withheld and the gap named.
Three library layers: the bundled library, your own imported files, and an Office-approved company catalogue that is signed, hashed file by file, and re-checked after download so a model withdrawn mid-sync cannot become current. Office approval beats a local file of the same name. Each job pins the full text of the models it used, by SHA-256 - updating the library never silently changes a finished well, and if Office later withdraws a model a job is pinned to, the job says so.
IPM formulas are read by a bounded arithmetic parser - no scripting engine, no reflection, nothing executable. Every tool file we ship has been run through the engine on a real well: the ones that load, work. A file that names a survey input we do not supply is refused while you are still choosing, not halfway through a job.
There is no geomagnetic model in the program. No BGGM, no HDGM, no IGRF, no WMM. Nothing is computed from position and date.
Magnetic dip and total field strength are entered from your well plan, per well, in nT, µT or gauss. If they are missing, anti-collision is blocked until they are entered - they are never assumed.
Declination and grid convergence are stored as report information and are never applied to anything. Bearings are used exactly as entered. This is deliberate: silently rotating somebody's surveys is a good way to move a well on paper.
Gravity is the one modelled quantity - WGS84 normal gravity from latitude and ground elevation, marked as a regional approximation, with a measured or Office-supplied value taking precedence.
A gyrocompass does not use the magnetic field at all. It finds north from the horizontal part of the earth's turn, which is the full rotation rate times the cosine of your latitude: strongest at the equator, nothing at the pole - which is why a gyrocompass cannot find north at the pole.
Stationary gyrocompass tools are supported. The approved tool file carries its own weighting formulas, as it does for an MWD; what the engine supplies is the survey inputs those formulas ask for - latitude, the horizontal earth rate worked out from it, the azimuth, the gyro toolface. Latitude comes from the pad position already entered for the satellite picture, so nothing extra is typed.
The test we hold it to: the same gyro tool, down the same well, at rising latitude. Its uncertainty has to climb, because the earth rate it works with is falling. On one of the published tools that is 2.30 m sideways at the equator, 2.53 m at 56°, and 8.40 m at 85°. A magnetic model put through the same test does not move at all, which is also correct. If either of those ever stopped being true the build would fail.
What is still not supported: continuous gyro - the running, integrating kind with a live toolface and mode switching. Those files are refused at import, by name, with the reason. Some gyro files also fold their latitude dependence into their own magnitudes rather than writing it out; those are used exactly as written. For anything this engine does not model, import a full one-sigma covariance export from the survey company - that path is fully supported and gives the same anti-collision answer.
One, on every job - there is nothing to pick:
The surface margin (0.3 m) and projection (0.5 m) are yours to set. Each well's hole or casing size counts; none entered uses the standard 311 mm (12 1/4 in) hole. The reference well's radius can come from the actual hole size at that depth in your BHA sections.
Evaluation sigma and display sigma are kept separate - one drives the separation factor, the other draws the ellipsoids and the reported gap. Both can be read automatically from an attached anti-collision report; conflicting values are treated as unusable rather than averaged into something plausible.
Warning limits are yours: a separation-factor floor and a centre-distance floor, either one alone or both required, default 1.5 and 20 m.
Not implemented: a formal minimum-allowable-separation-distance rule set, or a graded alarm ladder. Warnings are on or off against the limits you set, with the reason written out. Where a figure has been scaled from an imported report rather than calculated here, it is marked (est.), banner-flagged, and excluded from alarms.
Minimum curvature, for every well, every plan, every import and every interpolation. There is no alternative method and no legacy mode - old jobs and new jobs are computed the same way. Dogleg uses the haversine form; a 180° dogleg is refused rather than approximated.
No map projection is applied to survey data at all. The job runs in the coordinate frame of its plan. There is no UTM or latitude-longitude conversion of survey coordinates, and no grid scale factor.
North reference - magnetic, true or grid - is carried per well. An offset well whose north reference does not match the job's is hidden from anti-collision rather than rotated into place, because a rotation nobody checked is worse than a well you can see is missing.
The one datum conversion in the product is the NAD27-Canada grid shift used to place the satellite picture. It moves the picture. It never touches a survey or an anti-collision coordinate.
Written down so you do not have to find out later:
An anti-collision answer can be wrong in two directions. Too cautious costs a day. Too confident costs a well. Everywhere the program cannot stand behind a number, it withholds it and names what is missing - it never picks a default and carries on. If you find a place where it guesses, tell us and we will treat it as a defect.
SharpShooter Gold is not affiliated with or endorsed by the ISCWSA. The published error-model data is used as published; the implementation is our own. Trademarks named on this page belong to their owners.
Ask Pneuma
SharpShooter