Pilot
Operator validation
One asset. One decision question. A result that can be a no. A validation is an experiment designed to establish, to an engineering standard, whether the method works on a specific well population — and what result would establish that it does not.
Who it is for
Producing-well portfolios with a review question worth answering.
- Producing-well portfolios — operators with a well population and a decision question that decline analysis can legitimately inform: which wells warrant engineering attention, what the do-nothing baseline is, where the data record itself is the problem.
- Production, reservoir and asset teams — the engineers who will own the verdicts. A validation supplies evidence to a named engineer; it does not replace one.
- Operators seeking broader review coverage and clearer prioritisation — and prepared to bring the wells expected to fail as well as the wells expected to work. The census — the full list of wells in scope — is declared before any data is touched.
- Not for reserves evaluation, regulatory filing support, or any process that needs a number for every well. The method fails closed by design.
What is needed
Start from the history you already have.
Only the first column is required. Everything else raises what can be concluded, and its absence is printed on the dossier rather than hidden.
Identity and history
- Well and completion identity: a stable well identifier and the reporting entity volumes are filed against, with what each entity physically is
- Production history: monthly or finer oil, gas and water volumes per well or completion, with unit system and measurement basis declared per source
- Period status, first production date and extract date, so every result can be pinned to a data vintage
- The decision the validation is meant to inform, stated before the data is touched
- Named engineers to confirm identity links and events, adjudicate segmentation and approve publication
Operating context
- Producing days or hours on production per month, and downtime with start, end and reason code
- Operating events and reports: shut-ins and restarts, workovers, stimulations, lift, choke and equipment changes, facility constraints — structured or as remarks, tickets and well files
- Whether volumes were directly measured or allocated, and by what method, plus amendment history
Economics and surveillance
- Economic assumptions for value screening: one declared condition set — price basis, incremental cash operating cost, taxes, ownership and effective date — owned by a named person
- Flowing and static pressures, well tests, injection volumes, water-cut history and completion intervals — recorded on the dossier; they change what it can state, not what the platform currently computes
- The operator's existing review record, captured before the first run, so a differential measure can be reported
What ActionWell runs
Declared before it runs. Reproducible after.
A pre-run declaration fixes scope, census, gates and every policy parameter before any run touches the data. Later changes are logged amendments.
Data-quality assessment
Ingestion into an immutable raw layer with lineage; identity resolved over time; a data-quality profile and eligibility verdict for every well, every run, with the gate and reason for any exclusion.
Production-history reconciliation
Reported volumes reconciled against filed totals; zeros kept by cause, gaps preserved, negatives quarantined; conditioning that changes numbers only reversibly.
Baseline analysis
Segmentation into operating regimes with candidate boundaries at recorded changes; multi-candidate decline fitting under declared bounds; plausibility checks; out-of-sample backtests at several past origins; the expected baseline per accepted well with its conditions and uncertainty by component.
Evidence assembly
Deviation against the published baseline as new production arrives, with operating-event associations carried as hypotheses beside the alternatives not excluded.
Materiality and economic screening
Materiality is reported per well as the list of conditions cleared or failed; economic limit and identified value only where a complete condition set exists, with price, cost and policy sensitivities reported separately.
Engineer-review workflow
Review queues on technical signals, each with its criterion printed; engineer review with recorded verdicts; publication pinned to data vintage, code version and policy version after a reproduction test; standing monitoring from first publication.
What the operator receives
Every well accounted for, including the ones that fail.
- Portfolio coverage assessment — every entity in scope traced through each gate to publication, by count and by production weight, with the gate that removed each well named.
- Data-quality and method-eligibility view — unexplained zeros, quarantined negatives, restated months, identity resolved only through an unconfirmed link, unit or basis changes: the data problems the operator could fix, separated from the regime problems nobody can.
- Ranked review queue — wells that warrant attention with the reason each is there and, for every well not queued, the named reason. No combined score.
- Selected well dossiers — for wells in scope, including those that came out not eligible: the outcome, the failing item, the evidence that would resolve it and when to re-evaluate.
- Baseline and uncertainty — expected baselines with their conditions, rate basis and policy versions printed; the backtest record per horizon with the benchmark named; calibration measured on the operator's own population, or an explicit “not yet measurable”.
- Validation log — who reviewed, what was amended as versions, dissent, and the amendment log, so the reader can see nothing was moved after the fact.
- Agreed validation findings — the conclusion, positive or negative, on the front page, with the conditions that would change the answer: better data, a different asset, longer history, finer cadence, a different decision scope.
- The negative record — every well suppressed, every model rejected, every group too small to measure, every metric not computable, and why.
Decision boundaries
Stated before the first number, so they cannot move after it.
- Demo values are not customer results. Every product screen on this site shows an illustrative demo portfolio. No operator deployment exists yet and no operator result is claimed.
- Decline-curve analysis does not prove causality by itself. A temporal association is not causality; attribution is a named engineer's act, with evidence the calculation does not supply.
- Identified value is not realised production. It is a conditional, unrealised estimate under one assumed condition set. Realised uplift is measured only against a do-nothing baseline frozen before the job, and the two are never summed.
- Material actions require engineer validation. Every automated candidate — segmentation, association, exception — is adjudicated by an engineer before it changes a published result. A candidate surfaced by the system is not an approved action; the system does not recommend interventions.
- No reserves conclusion. ActionWell does not book or classify reserves and is never accountable for a reserves conclusion.
- No economics without the operator's own inputs. A rate cutoff is not an economic limit, and wells are never compared under split conditions.
- No performance figure before it is measured here. Nothing measured on one asset is quoted as a product characteristic.
- Responsibility stays with the accepting engineer. Nothing reaches the operator without a named engineer's scoped approval, and responsibility for results does not transfer to the tool.
How a validation is judged
A validation that cannot produce a no cannot produce a trustworthy yes.
The conclusion — positive or negative — appears on the front page, with the coverage funnel, the metric register and the negative record published regardless. A negative result is a valid, designed-for outcome. It may be that decline analysis is not the right instrument for this asset; that coverage collapses on data defects; that the fitted models do not beat a trivial benchmark on this history; or that the queue adds nothing over the engineer. Each is stated plainly rather than tuned away, and a no about this asset is not a no about the method.
No fixed duration is promised, because forward-measured calibration at long horizons cannot be accelerated, and no result is guaranteed, because a validation that reports outcome measures early is not honest.
Commercial model
Subscription per covered producing well.
ActionWell is offered as a subscription per covered producing well, with portfolio terms reflecting scope, data readiness, and deployment requirements.
Scope a validation
Validate ActionWell on an operating portfolio.
Bring a defined well population and the production history already available. ActionWell will establish coverage, surface data limitations, construct the baseline, and produce a governed review queue for engineer validation.