Model assurance & calibration
Compare models with operational evidence, identify material input assumptions, recalibrate responsibly and communicate limitations.
Technology
Build the decision layer around existing engineering systems.
The industry does not need another generic AI promise. It needs traceable engineering tools that solve identifiable operational problems.
Direction
Drilling Engineering Intelligence
A focused category combining physics, engineering models, operational data, engineering rules, specialist knowledge and carefully governed AI assistance.
01
Observe
Organise relevant signals, operating state and data quality.
02
Contextualise
Connect data to well conditions, engineering models and execution constraints.
03
Interpret
Apply engineering rules and specialist judgement while preserving uncertainty.
04
Act
Provide clear, traceable recommendations with measurable triggers.
Diagnostics & surveillance
Structure connection, fingerprinting, loss/influx, tripping and operating-window analysis into repeatable, reviewable workflows.
Engineering workflow QA
Support calculation traceability, document consistency, risk review and explicit separation of observation, interpretation and action.
Decision support
Encode measurable triggers, engineering logic and operational trade-offs without hiding the assumptions behind an opaque interface.
Technology principles
Engineering accountability stays visible.
01
Traceable calculations
Inputs, assumptions, methods and limitations remain reviewable.
02
Human technical authority
Tools support judgement; they do not conceal or replace engineering responsibility.
03
Fit-for-purpose scope
Every workflow begins with a defined decision, user and operating context.
Application portfolio
Software built for a decision that was being made without evidence.
Each application exists because an engineering judgement was being made on something other than the measurement that could have settled it.
How the tools fit together
Data shown
Screenshots use synthetic or published data. No client record, well, result or proprietary method appears in any image on this page.
01Model assurance
Anadromiki
Turns a real drilling record into a simulator batch schedule, then grades the simulation against what the well actually did.
A transient hydraulics model sets the pressure targets a narrow-margin section is drilled to. Whether the model deserved that authority is usually answered afterwards, by eye, from a simulated curve laid over a measured one — which is an impression rather than a result. The loop between what was predicted and what the well did is rarely closed, and almost never closed the same way twice.
01
Import the record
A time-indexed LAS, or a CSV export from a rig-data system. Channels are mapped to roles automatically, and a built-in profile takes over where a name match reaches the wrong answer.
02
Build the schedule
The record is segmented into rig states, cropped to the simulator’s starting bit depth, and written as a batch grid that goes straight into the simulator.
03
Grade the run
The measured PWD is scored against the simulation — bottomhole pressure, ECD, standpipe pressure, downhole and return temperature — split by pump state. Several runs are held at once, so a calibration is a series of trials rather than a single answer.
Where it refuses to help
It will not let a bad sample become a trend
A PWD frame reporting 32,562 psi is held forward across roughly 115 rows by ordinary gap-filling, and six such readings moved a graded bias by more than half. A sample is dropped only if it is statistically absurd, physically material for its unit, and returns to where it came from — so connections and pump shutdowns, which are the point of the analysis, survive untouched.
It will not assume a unit
A torque channel declared in 1000 ft·lbf written into a column headed lbf-ft is a thousand-fold error that nothing downstream would notice. Units are converted at the door and an unrecognised one is reported, never guessed.
It will not average away the event
After segmenting, each run is asked what it still conceals, and the ones hiding a real change are split at their steepest transition. An MPD connection sweeping the choke over five minutes becomes a staircase rather than one averaged row.
It will not quietly correct anything
Where the simulation and the PWD tool report at different depths, the difference is measured from the loaded files and stated — in psi and in ppg — and no number is adjusted. The engineer decides what to do about it.
Hydraulic model calibrationPWD interpretationRig-state segmentationPost-well review
02Circulating hydraulics
Piesis
A steady-state circulating hydraulics engine: annular pressure and ECD from surface to TD, and standpipe pressure leg by leg, for conventional and managed-pressure wells.
Most hydraulics questions on a narrow-margin section are steady-state questions — the ECD a rate puts on the shoe, the surface back-pressure that holds the bottom of the hole on a connection, how wide the rate window is between hole cleaning and the fracture gradient. They tend to be answered either in a full transient simulator, which is specialist-run and slow to set up for a quick question, or in a spreadsheet that shows its answer but not its assumptions. Neither suits an engineer who needs the number now and has to defend it later.
01
Describe the well
Survey, casing, riser and open hole, string and bit, and the mud as it was measured — viscometer readings, and density at temperature and pressure. A case can also be read from a simulator case file.
02
Solve the circulating state
The annulus is marched from surface to TD on published friction, rheology and density correlations, each cited. Standpipe pressure is assembled leg by leg — surface lines, string bore, motor and MWD, bit nozzles, annulus and back-pressure — rather than from one lumped formula.
03
Ask the operating question
The solved case is read against the pore and fracture curves: the rate window, the back-pressure needed to hold a target at a depth, the bit hydraulics, and swab and surge on a trip.
Where it refuses to help
It will not invent a limit
A rate window is only as wide as the constraints the case carries. An unset limit or an absent pore or fracture curve binds nothing and is reported as unset — the engine supplies no operating limit of its own.
It will not lump the standpipe pressure
Every leg is reported separately, including the ones that read zero. A motor with no pressure-loss curve shows as a zero in the breakdown, so a missing input stays visible instead of disappearing into a total that happens to look plausible.
It will not promote an experimental model
Rotation, cuttings loading, tool-joint minor losses and circulating temperature are each off by default and labelled experimental. The temperature model states on screen how far it has been validated before anyone relies on it.
It will not hide which correlation produced a number
The report lists every correlation active in the case, with its source and status, taken from one register. A model that is switched off does not appear, so the list is what was used rather than what is available.
ECD and ESD profilesMPD back-pressure studiesRate window and bit hydraulicsSwab and surge screening
03Pressurized mud cap drilling
Anadysi
Models how fast gas rises under a mud cap, by every published basis at once, and the bullhead rate and cycle that keep it where it belongs.
Pressurized mud cap drilling (PMCD) trades a loss zone for a gas problem. Gas that enters the annulus rises under the mud cap, the surface pressure climbs, and the gas is bullheaded back. How fast it rises sets the bullhead rate and the length of every cycle — and the published answers differ several-fold, from a field rule to an idealised Taylor bubble. A plan built on one of them, chosen without seeing the others, rests on a number nobody can defend.
01
Describe the well
Geometry and string, the mud cap and sacrificial fluids, the loss zone and the pressure limits — built up on screen as the case is entered.
02
Model gas migration
Every published basis runs on the same well and the same clock: a field rule, the Taylor-bubble family, drift-flux in an annulus and dispersed bubbles. The planning basis is chosen with the spread in view, and the bullhead rate that beats it follows.
03
Plan the cycles
The cycle itself — allowable surface-pressure rise, real-gas expansion on the way up, bullhead volumes — with the input uncertainty carried through to P10, P50 and P90 rather than a single answer.
Where it refuses to help
It will not use a velocity nobody can derive
An in-house migration velocity with no published derivation is not built in. An engineer who owns such a number may enter it, and it is labelled as an input everywhere it is used.
It will not let the ceiling become the plan
The Taylor-bubble bases describe one idealised bubble filling the hole — an upper bound, several times faster than field migration. They are shown, and labelled on screen as not a planning number.
It will not hide the safety margin
There is no default margin applied out of sight. The engineer sets it, and the panel states what it means for the bullhead rate.
It will not assume the annulus is a pipe
Where a basis is used outside the geometry it was derived for — a pipe-flow bubble in an annulus — it is marked as an assumption on screen, not adopted silently.
Gas migration modellingBullhead rate selectionPMCD cycle planningUncertainty (P10 / P50 / P90)
04Underbalanced drilling
Grafikos
Turns a well-flow simulator’s injection sweep into the underbalanced operating window — and into the report a team signs off on.
An underbalanced design is a sweep: dozens of gas and liquid injection rates, each a full multiphase simulation. Which combinations keep the well underbalanced, clean the hole, run the motor and stay inside the surface limits is scattered across the case reports — and is usually assembled by hand in a spreadsheet, once, under deadline, where nobody can check it afterwards.
01
Import the study
Summary workbooks, text summaries and the per-case reports of a sweep are read directly, units carried into every channel, one curve per injection case.
02
Draw the window
Bottom-hole pressure against gas injection for each liquid rate, closed by the limit curves — minimum liquid velocity and motor throughput — between the reservoir and target pressures, with the dew point marked. Hole cleaning and velocity are drawn against depth, section by section.
03
Report the study
Every template, for every run, exported in one pass as reopenable case files and PDF or PowerPoint reports, with a per-case summary of the criteria each case fails.
Where it refuses to help
It will not draw a window the limits do not support
A liquid rate whose curve never meets a limit closes the window there, rather than being stretched to fit. The shaded region is only what every criterion allows.
It will not apply a criterion to the wrong section
Hole-cleaning and velocity criteria differ between vertical, inclined and horizontal hole. Each section is judged against its own minimum, and an interval excluded from the criterion is marked as excluded, not quietly passed.
It will not lose a failing case in the summary
The results table lists, case by case, which criteria fail — including gas returns above the surface limit and bottom-hole pressure below the dew point.
It will not leave a unit to be guessed
Units travel with every imported channel, and a conversion — scfm to MMscf/d, gpm to bbl/d — is declared in the template that makes it, not applied out of sight.
Underbalanced operating windowHole-cleaning criteria by sectionMultiphase study reportingEngineering data plotting
A workflow worth improving?