TerraMetrix Ensemble Explorer · User guide Open new workspace ↗
WORKFLOW AND INTERPRETATION

From ensemble uncertainty to targets for review

Use the four modules together to inspect simulation responses, assess history match, explore spatial uncertainty and identify areas worth investigating. The workspace keeps opened modules and their results available while you switch.

1. Start with a consistent analysis basis

  1. Open the workspace and select an ensemble folder in your first module. File access is read-only; simulation files are processed locally in the browser.
  2. Open another module and choose Use saved ensemble. This reuses the saved folder choice after the browser's permission check; it does not automatically load the other module's analyses.
  3. Use Workspace overview to check opened sessions, activity and selected folder names. Vector and ScoreCard share one session. Map and Well have separate sessions and settings.
  4. Keep dates, actual restart dates, cases, K intervals, coordinates and units consistent where the comparison requires it. The workspace does not synchronize these settings or transfer results between modules.

Opened modules retain their results and caches when you switch inside the workspace. An unopened module loads only on first use. The shared activity indicator also reports work continuing in a hidden module. Activity animation offers System, On and Off; System follows the device's reduced-motion preference.

Different selected folder names prompt a source check. Equal names do not verify identical folders or models. “Folder selected” does not mean every analysis is loaded. Inspect each module's provenance and result status.

2. Vector Explorer: understand model responses

Select a folder containing SMSPEC/UNSMRY pairs. Inspect the discovered realizations, choose a summary vector and realization, then load the vector or all realizations. Check the vector's keyword, entity, unit and available time range before interpreting the chart.

Use individual curves to inspect outliers and the ensemble envelope to see response spread. The final-value histogram describes the loaded endpoints; check complete-to-end counts and expected end date before comparing values. A spread is a model-ensemble description, not a guarantee of forecast accuracy.

Next: use ScoreCard to examine the available history-match evidence, or Map to inspect the spatial properties behind differing responses.

Choose Field, Region, Group, Well or Other / system, then search for a vector parameter. For Region, the full keyword and region family are displayed, followed by a searchable region number. For Well and Group, choose a parameter and a matching name. Only choices present in the loaded SMSPEC are offered, including history vectors.

Choose Use vector to update the selected vector, then use the existing single-realization or ensemble load controls. Browsing choices does not read UNSMRY files or change existing results. The source index and unit are shown before selection. Unknown region families are explicitly labelled; family names are never inferred from DATA files.

Browse all vectors · advanced retains the original searchable catalog for system or unusual entries. It is collapsed by default and its rows are rendered only when opened. This navigation changes neither calculations nor caches.

3. ScoreCard Explorer: inspect history-match evidence

Use the searchable, multi-select Entities and QC prefixes menus. Selected items remain visible as removable chips; typing a new search does not discard them. Choose Field, numbered Region entries, named Group entries or Well entries, together with matching prefixes such as FOPR or ROPR. QC lists only simulation/history pairs that exist in the loaded SMSPEC. All selects the full entity catalogue, independent of the search; None clears the corresponding selection. Field can be selected on its own. Choose entities first: the QC prefix menu offers only prefixes with at least one simulation/history pair for those entities, with a scoped pair count. Previously selected prefixes that lose all pairs remain marked until removed. A status message identifies missing entity/prefix combinations (up to six examples); only existing pairs enter QC, as before. Defaults uses available prefixes for the current entities; with no entities selected, no new prefixes are offered. The entity catalogue stays independent of the prefix selection. Forecast menus are unaffected and can include vectors without history. Check the selected pair count before running QC. If the enhanced menu is unavailable, a native multiple-selection list remains usable.

Select entities and vector prefixes that have matching history vectors, such as WOPR/WOPRH. Review category weights and the maximum normalized misfit, then run the score card. Compare the ranking, vector/category scores and history coverage rather than relying on a total score alone.

Acceptance requires every selected vector pair to have a finite normalized misfit at or below the chosen threshold. It is distinct from the weighted total score. Results depend on the selected observations, entities, prefixes and weights; changing that basis can change the ordering.

Understand weights and acceptance

The QC setup follows three bands: select Entities and QC prefixes at the top, review the full-width weight preview below, then set acceptance and run QC in the bottom row. Panels stack on narrow screens. Input ranges and QC calculations retain their existing behaviour.

Weights for total score adjusts the existing relative-priority sliders. Percentages beside the sliders and the coloured bar show a Preview of normalised category shares for the selected QC vectors. Four included categories with priority 1 each have 25% each. If only Production and Pressure are selected, priorities 0.4 and 0.3 give approximately 57.1% and 42.9%; unselected categories show Not included. A single included category has 100% regardless of its positive priority value. Slider positions remain relative priorities, not independent percentage targets.

These are category weights, not separate weights for every vector. Selecting several production vectors still gives Production one category share; the original within-category score calculation is unchanged.

Equal priorities sets the four original adjustable priorities to 1 without running QC. The original 0–1 values remain inspectable under Relative priority values. If every included priority is zero, the existing calculation uses equal shares among included categories; the preview explains this fallback. The existing Other category has a fixed zero priority and no new slider. When selected, its share is shown explicitly. It is not changed by Equal priorities, but can receive a share through the existing all-zero fallback.

The preview uses selected category metadata only and opens no simulation files. Actual category availability is confirmed during QC: a selected category with no finite scores may be absent from the completed weights. The Weights used in last completed QC bar beside the results reads the existing saved normalised weights directly. It is not recomputed from current form inputs. Pending edits, including changes made while QC is running, remain preview-only until a new analysis uses them. Percent labels are rounded; bar segments use the unrounded shares. Changing ensemble clears the displayed completed weights.

Acceptance criteria is a separate block. Weights affect total-score ranking, while every selected QC vector must have a valid misfit at or below the limit. Setting a category to zero priority does not remove its vectors from acceptance testing. To change which vectors are tested, change the QC selection. Adjusting the weight preview does not change existing scores, acceptance or forecasts; run the score card to apply a new setup.

Read the QC heatmap: relative score and acceptance

The default Relative score view keeps the existing within-ensemble ranking colours. Green means better ranked for that vector; red means worse ranked. These colours do not test the misfit limit. A low relative score can still pass the absolute misfit criterion. A high total score cannot compensate for a failed selected vector.

The separate Status column shows the existing Accepted/Rejected result. On rejected rows, ×n counts finite vector misfits that fail the limit, and ?n counts unavailable misfits. A vector cell with × and an amber frame fails the saved limit; ? means no valid misfit. These markings apply only to vector columns. Every selected vector must have a valid misfit less than or equal to the limit. Equality passes. Category and Total cells show the original aggregate relative scores and have no independent acceptance threshold.

Hover a vector cell to see its actual normalized misfit, the limit from the completed QC run, Pass/Fail/Unavailable, relative score and row failure counts. For example, with a limit of 0.75, a misfit of 0.70 passes even if that vector's relative score is low. A misfit of 0.80 fails even when other vectors give the realization a high total score. In a single-category selection, the category and Total scores can be identical.

Choose Misfit / limit to show distance from the acceptance boundary instead of relative rank. A ratio below 1 is within the limit, exactly 1 passes, and a ratio above 1 fails. Category and Total columns are hidden because their scores have no comparable misfit limit. Colours saturate at 2, but finite cell values and hover ratios are not truncated. With a zero limit, exact zero misfit passes; positive misfit has an infinite ratio, an × marker and a failed status, distinct from unavailable data. A non-finite or negative limit has no ratio view. The Status series uses explicit colours rather than the numeric colour scale.

Changing the view or Top rows redraws existing results without running QC, reading simulation files, or changing acceptance, scores, weights or forecasts. Pending edits to the setup do not replace the threshold stored in the completed result. On narrow screens scroll the heatmap horizontally; cell symbols and status remain readable. Exported charts retain the visible status and markers; the chart-data acceptance series is explicitly labelled 1 accepted / 0 rejected.

Under Representative forecast cases, first choose Entity type: Field, Region, Group or Well. For Region, choose a Region family, then a Region number and a Forecast vector. Every navigation menu is searchable: type a family tag, a number such as 200, or a vector such as ROPT. Only numbers in the selected family and vectors for the selected number appear, so hundreds of numbers from several families no longer share one list. Numbers are ordered numerically. Group and Well show a searchable name menu; Field goes directly to its vectors.

The selection line shows the complete family, number and raw vector with units. Changing family keeps the number and forecast quantity when available; otherwise the menus select an available option, which you should check before generating cases. Only entities and vectors present in the loaded SMSPEC appear. Region identity uses NUMS; group and well identity uses WGNAMES. Forecast selection does not require a matching history vector and does not change which realizations passed QC. Navigation uses already loaded metadata and opens no simulation files.

Region identity: the full summary keyword and SMSPEC NUMS identify the selected series. Recognized standard regional quantities (for example ROPR and ROPT) use FIPNUM · Region n. Qualified quantities such as ROPT_REG use FIP family REG · Region n; REG is the reported summary tag, which may abbreviate a longer DATA family name. Different families with the same number remain separate in QC and forecast menus. The explicit NUM suffix is treated as the standard FIPNUM family. Original vector names remain visible in selections and chart titles.

Family labels recognize explicitly padded five-character quantity slots and their family tags independently of the list of known quantities. For example, RRPV_PRE and RRPV_REG are labelled PRE and REG even though quantity/history recognition remains conservative and unchanged. Unrecognized bare names and ambiguous unpadded forms are temporarily hidden from the representative forecast choices rather than guessed as FIPNUM. The source metadata is retained and these vectors remain inspectable in Vector Explorer with their raw keyword and NUMS. This display filter does not change history-match QC or remove any source values. They are not automatically classified as history solely because a trailing family character is H. No DATA comments, geographical names or region arrays are read to invent labels. For example, a geological or geographical region name requires a separately verified name mapping.

The representative forecast cases describe the selected forecast vector within the accepted QC realizations. History-match quality does not establish a unique geological model or prove that a proposed well adds production.

Next: inspect spatial differences in Map and retain an explicit, representative realization selection for opportunity screening. ScoreCard acceptance is not automatically applied to Map or Well.

Explore shortlist cases across another vector

After generating a shortlist, use Explore selected cases across other vectors below the forecast comparison. Comparison vectors do not need matching history. Search the comparison-vector menu and choose First, Last, Maximum, Mean or Sum, then click Compare distribution. No comparison files are read automatically. FOIP with First is the initial default when available; otherwise the shortlist vector with Last is suggested. This is a separate comparison: it preserves REAL identities, case labels, QC scores and the original shortlist.

The grey empirical cumulative distribution shows all discovered realizations with a finite comparison aggregate, including QC-rejected cases. The green curve shows only QC-accepted realizations with a finite comparison aggregate. Each population has its own valid-value denominator. Counts show valid/total values for the whole ensemble, QC accepted and shortlist labels. The coloured markers show existing shortlist cases on the whole-ensemble curve; the table also gives their position within the accepted population.

A rank of 75% means that 75% of valid values in that population are lower than or equal to the case value. Ties and the case itself are included. This ascending empirical rank is not a probability of success and does not rename a case: a P90 selected from FOPT can occupy any FOIP rank. Repeated REALs retain all shortlist labels and can share a marker location; hover identifies the shared labels. There is no fitted distribution or density estimated from the few shortlisted cases.

Aggregation reuses the existing scalar definitions over the entire loaded series, including history. First/Last skip non-finite reports, retain zero and show each shortlisted case's actual report date. First FOIP is therefore the first valid loaded value, not automatically forecast-start FOIP; inspect report dates when comparing initial states. Mean weights reports equally and Sum is not a time integral. Comparison availability requires a finite aggregate, not original forecast completeness: a case with incomplete production forecasts can still provide a valid initial FOIP background value. Missing comparisons remain visible in the shortlist table and in the unavailable-values disclosure; missing ranks are shown as dashes.

Compare distribution explicitly evaluates the chosen vector for the discovered ensemble through the shared readers, six-worker concurrency and existing caches. It retains only scalar comparison rows and plot data beyond those shared caches. Pending edits leave the displayed basis unchanged until rerun. Module switching keeps the result; a new shortlist, QC result or ensemble resets it. Source-changing actions are guarded while reads finish, including saved-ensemble reuse, to protect cache ownership. Nothing is persisted and no composite multi-vector score or new shortlist is calculated.

Forecast versus comparison value

After completing a distribution comparison, use Forecast versus comparison value below its table. X reuses the exact scalar values from that completed comparison. Y uses the forecast vector that produced the shortlist. For example, use FOIP/First for X and FOPT/Last for Y. Choose Y aggregation and click Plot relationship; no Y evaluation runs automatically. Scalar shortlists default to their original aggregation; whole-curve shortlists default to Last for a physical Y value, not curve-distance RMSE. Both axis titles include the vector, units and aggregation.

Each background point represents one REAL with finite X and Y values. Green points are QC-accepted cases; grey points are the remaining ensemble cases. These background groups are disjoint. Existing shortlist cases are highlighted using their original labels and colours; repeated roles may overlap. Hover reveals REAL, X/Y values, QC status, shared shortlist labels and First/Last report dates. The shortlist table lists the same paired values and missing status. Paired/total counts are shown for the whole ensemble and accepted population; missing X or Y is disclosed by REAL without replacing it with zero. Y is not read for cases with no valid X.

The axes show physical scalar values, not cumulative percentages. Use the cloud to see whether higher FOPT tends to accompany higher initial FOIP, and to compare production responses for cases with similar initial volume. A case's original P90/P50/P10 label does not give its rank on either axis. No regression, correlation coefficient, recovery factor, causal claim or new representative-case selection is calculated. Cases can have different available report dates, and the plot adds no completeness/date-alignment filter; inspect dates, especially with Last.

The entire loaded series, including history, is used with the existing aggregation definitions. Pending Y changes preserve the completed plot until rerun. A new completed X comparison, new shortlist/QC or ensemble clears the relationship result. Module navigation retains and reflows both plots. Existing readers, six-worker concurrency and caches are reused, with source replacement guarded until in-flight Y reads finish. The CDF and its ranks remain unchanged.

Understand QC Reference and forecast eligibility

QC REFERENCE · BEST ELIGIBLE CASE replaces the ambiguous Best history match shortlist label. It is the highest QC score among accepted cases eligible for the selected forecast vector and method. The ranking chart and top QC summary still describe the scored/QC population. A higher-scoring accepted case may be unavailable for a particular forecast, so QC Reference need not be the overall QC leader. Its selection algorithm has not changed.

After Generate shortlist, open Forecast eligibility below the status/warning. The table lists every accepted QC case, sorted by QC score, with its last valid date/time, expected end, forecast eligibility and reason. It identifies the QC Reference and counts higher-scoring accepted cases excluded from that forecast. Expected end is the latest valid end among successfully evaluated forecast series, rather than a separately configured simulation date. Missing summary pairs, evaluation failures and missing usable aggregate values are distinguished from series ending early. A dash means the date is unavailable, not zero.

Whole-curve mode additionally explains missing/invalid reports, incompatible time grids or an unavailable comparison. This is an explanation of the method's actual eligible-case list; the table does not alter eligibility or QC acceptance. Existing warnings count successfully evaluated forecast series; this table starts from all QC-accepted cases and can therefore also show cases without a usable forecast value. It describes the last evaluated settings and is cleared on a new forecast evaluation, QC run or ensemble replacement. Expanding the table and changing display/settings read no simulation files. The table is held in page memory only.

First value and whole-curve forecast selection

Single-value summary remains the default. Last uses the last finite report, First the first finite report (including a valid zero), Maximum the highest, Mean the arithmetic mean of finite reports, and Sum their unweighted sum. Mean gives each report equal weight; it is not time-weighted. Sum is not a time integral: summing rates does not produce a volume. First can describe initial OIP, but refers to the start of the loaded simulation, not automatically to forecast start. Scalar mode retains the existing complete-to-end population rule.

Closest to percentile curve compares entire loaded curves, including any historical period. First choose the vector, then choose this selection method and generate the shortlist. This version has no separate forecast-start or date-range filter. The result states its start/end, timestamp count and eligible population. It requires TIME or YEARS and at least two complete accepted curves with finite values at every report, strictly increasing times and the same time grid. The longest valid period, then densest grid, defines the comparison basis; REAL breaks ties. Curves with gaps, shorter periods or different grids are excluded and counted. There is no interpolation, extrapolation or automatic shortening to fit missing curves. QC acceptance is unaffected.

At every shared timestamp, the existing quantile convention gives P90 = q10 (low), P50 = q50 and P10 = q90 (high). For each candidate, squared errors to each target curve are integrated using trapezoidal time weights, divided by the period length, then square-rooted. This time-weighted RMSE has the vector's units; lower is closer. Irregular reporting frequency does not give each report equal influence. Each target independently selects the closest eligible actual realization; ties prefer higher QC score, then lower REAL. The same realization can therefore represent several targets or also be Reference. A curve representative is not necessarily a percentile of final production or recovery.

Reference is the best QC score among eligible aligned curves and has distance zero to itself. Alternative is the curve farthest from Reference among the remaining top QC candidates, using the same time-weighted RMSE. Its distance is to Reference, not to a percentile. The table identifies the target of every distance. The comparison plot uses the same eligible population and target percentiles as curve selection. Switching method is a pending edit until Generate shortlist is clicked; results describe the method actually used. Readers and caches are reused.

4. Map Explorer: investigate spatial uncertainty

Select the reservoir property, K/region filters, realization selection, restart settings and map resolution, then build maps. Inspect the map QC information and legend units. Compare mean, median and percentile maps; use the section/profile tools to examine changes along a line.

Aggregation depends on the property. A saturation, permeability, hydrocarbon column and volume are different quantities with different allowed calculations. Read the property description beside the controls. Difference maps are last restart minus first restart and can be negative. Model XY coordinates do not by themselves establish a geographic coordinate reference system.

Find actual realizations close to a percentile map

P90, P50 and P10 are calculated independently in each map cell from the finite values across compatible realizations. The resulting percentile surface normally does not belong to one realization, and its sum is not generally the percentile of total volume. In this module P90 is the statistical 10th percentile, P50 the median and P10 the statistical 90th percentile; low values are not automatically favourable for every property.

  1. Build maps as usual, check units, geometry and processed cases, and select P90, P50 or P10.
  2. Expand Representative realizations below the map and click Find closest realizations. The top three cards rank actual realization maps by area-weighted root-mean-square error (RMSE) from the selected percentile surface. Lower RMSE means a closer spatial match, in the same parameter units. It does not rank history match, forecast or commercial value.
  3. Use Show map or Show difference on a card. The map's view buttons switch between the percentile surface, the selected actual realization's derived map and realization minus percentile. Percentile and actual-case views share a colour scale; difference uses a scale centred on zero. Positive difference means the case is above the percentile surface. Zoom and pan are preserved.

Ranking uses the whole completed map, independent of zoom, on the same finite cells for every compatible case. At least 80% of visible target-map area must be shared by all cases; otherwise no ranking is produced. Missing values are not filled. The cards report compared area; difference maps leave unassessed cells blank. Native mode compares the existing logical I/J positions using reference-grid XY areas: it does not verify common physical geometry or remap different grids. Regular XY reuses the existing projection and sample mapping. Check whether the completed maps represent comparable geology and coordinates.

The cross-section/profile continues to show ensemble statistics, not the selected case or its difference. Changing the statistic clears its candidates; changing setup or ensemble invalidates the comparison until maps are rebuilt. A superseded comparison cannot publish old results. Comparison uses only loaded derived maps and does not open simulation files. Only the latest completed set can be retained, subject to a separate 64 MiB map/comparison memory budget; larger ensembles retain their normal map functionality but cannot use this comparison in this version.

Next: check whether potential opportunities are consistent with pressure, saturation, geometry and geological interpretation before taking them to Well's screening and review tools.

5. Well Opportunity Explorer: screen, compare, challenge

After selecting the folder, use Find opportunities for reservoir screening and target review, or Well diagnostics for named-well forecasts, classification, completion properties and the reservoir/production pilot. The ensemble selection is shared; switching these views retains the existing inputs and results. Screening does not require well discovery or completion loading.

Within Find opportunities, the shortcut bar stays available while you scroll. Use Criteria to return to the setup, Map to inspect spatial evidence, Compare targets to compare quantities, Candidate review to read the interpretation and Shortlist to revisit saved records. These links move within the same loaded result and do not run an analysis. The comparison table now follows the map, before the candidate card and review notes.

  1. Under Analysis basis, choose a reservoir date, a geologically comparable K interval and the geometry mode. Each case uses its latest restart at or before the selected date; inspect the actual dates in Case provenance.
  2. Start with a manageable case selection. Under Screening criteria, set justified reservoir thresholds, then the support/coverage requirements. Initial values are illustrative and editable. Blank optional fields disable that criterion.
  3. Check included/excluded cases, coordinates and missing data before interpreting targets. Increase to a larger representative ensemble after validating a few values against your model viewer.
  4. Use the map to locate patterns, the comparison table to contrast quantities, and candidate cards to document the evidence and the next checks.
  5. Challenge interesting areas with individual cases, single layers, narrower intervals and alternative justified criteria/map widths. Check geology, faults, communication, pressure, water/gas and a feasible well concept.

Exact cells compares one K layer and requires physically matching cell corners. Geographical XY · K interval allocates source-cell volumes by 3D overlap onto a common reference XY map. It can include differing grids, but does not correlate stratigraphy or transform unrelated coordinate systems. The same K numbers must still represent comparable geology.

After screening, Analysis summary records the last completed analysis. Expand its settings to check the ensemble folder, requested date, actual restart range, geometry, layers, case selection, thresholds and result counts. Editing the form does not rewrite this snapshot: its status explains whether criteria must be applied or the analysis basis must be loaded again. A cancelled or failed run leaves a historical summary, not a current result. Edit criteria returns to the setup without reading files. Changing map colours, zoom or table sorting does not change the analysis summary.

Expand How this analysis is calculated beside the screening controls to read the current geometry method. Collapsing this explanation does not change the settings or results.

XY interval PORV and oil pore volume are summed. Saturations and pressure use allocated pore-volume weights; PERMX uses a geometric mean weighted by geometric overlap volume. Criteria apply to the resulting interval aggregates, not independently to every contributing source cell.

6. Read the opportunity map

The map is useful for spatial interpretation as well as data checks. Map variable can show support, median/q10 oil pore volume, XY active bulk thickness or reference-completion distance. Always read the current legend: colour does not have one fixed meaning.

What you seeInterpretation
SupportThe fraction of all selected cases meeting every enabled criterion in an area. For 25 selected cases, 21 passing means 84% support. It is not commercial success probability.
CoverageThe fraction with usable enabled inputs or a known inactive observation. In XY mode, every selected K must contribute positive geometric volume. Missing enabled inputs in any contributing active cell make the observation incomplete. Coverage is not the fraction of map area occupied by reservoir.
Target outlineThe area meets the support and coverage thresholds and has positive support. Changing the displayed variable, palette or scale does not change its target status.
Missing or inactive observationsInsufficient coverage, no active observations and a missing displayed value have separate legend states. Missing data are not zero oil. Known inactivity is a known non-target.
Geometric regionsGroups of adjacent target areas. Member areas may pass in different subsets of the ensemble. A region is not a demonstrated hydraulic compartment or a joint success probability.
Well symbolsAmber shows recorded reference completions in the analysis K interval; blue shows other K layers. Filled/hollow denotes open/closed. These are completion-cell centres from the reference realization, not surveyed trajectories or wellheads.
Distance and hatchingHorizontal centre-to-centre distance to recorded reference completions, including open and closed connections. The visual radius marks proximity without filtering targets. Unknown distance is not “outside”. Vertical separation, drainage and hydraulic communication are not evaluated.

Drag to pan, wheel or use +/− to zoom, and use Fit map to reset the view. Click a mapped area or a target region to inspect its evidence. Well visibility does not change the distance calculations.

Automatic colour scales use the loaded map rather than just the zoomed view. To compare runs visually, choose the same variable, palette and explicit minimum/maximum. Set both bounds; values beyond them receive the end colour. Changing variable clears the bounds because the unit may change.

7. What does “XY map width” mean?

In geographical XY mode, this is the side length of one square map area, in metres. It controls the aggregation grid, not the width of the whole map, zoom, a search radius or a well drainage area.

At 400 m, one interior map square covers 400 × 400 = 160,000 m². At 500 m it covers 250,000 m². Only the actual source-cell overlap contributes; boundary areas are not filled by extrapolation. Within the selected K interval, overlapping contributions are combined for each square and each realization.

A larger width combines more geometry in each square and generally produces fewer, coarser areas. A smaller width resolves more spatial detail and generally requires more work. Because volume is additive and weighted properties are aggregated over the square, changing width can alter volumes, criterion outcomes, support, target counts and region boundaries.

Do not compare per-area oil volumes or target counts as if they were independent of resolution. Hold map width and K interval fixed for a direct comparison. For a sensitivity test, ask whether the geographic opportunity and its geological explanation persist, rather than expecting identical values. A positive contribution from each K still does not establish net pay or hydraulic continuity.

8. Use the table to compare evidence

Compare target areas has one row per original target area/cell. Sorting is high-to-low for one chosen quantity, not a composite well ranking. Table focus and pagination do not filter the map or the full screening CSV.

QuantityRead it with
Oil pore volumeAllocated PORV × SOIL in native reservoir volume units. Compare q10, median, q90 and n/selected. This is not stock-tank, mobile, recoverable or incremental oil.
Active bulk thicknessActive geometric overlap volume divided by the whole XY square's area, in metres. It is not net pay. A small value may reflect thin layers, partial areal occupancy or both. Available in XY mode.
n/selectedEach property's own valid observation count. Candidate quantiles include valid cases that fail screening criteria and known inactive zeros for volume/thickness. Missing/unavailable observations are omitted, not zero-filled. Other property distributions can have different counts.
q10–q90Ascending sample quantiles: q10 is low and q90 is high. With few realizations these are coarse descriptions, not confidence limits. A high median with a low q10 merits inspection of the individual cases.
Nearest completionCompare analysis K and all K distances, their reference date, well and connection identity. A nearby completion in another layer deserves a communication check; a distant one does not prove an area is undrained.

Use support and coverage to understand consistency and available evidence, then compare volume, spread, thickness and well context together. Click the area button to connect a table row to its map location. Inspect realization evidence before explaining an outlier or a missing value.

Region oil volume is summed within each complete realization before computing its distribution. Do not add area medians to obtain a region median. The region may have fewer complete observations than one selected area.

9. Turn an interesting area into a review record

  1. Open Candidate review for a region or selected area. Read “What the screen shows”, “Attention points” and “Next checks”. The card uses fixed rules over loaded evidence; suggested checks are not completed analyses.
  2. Keep Region overview separate from Selected area only. A regional distance minimum may belong to a different area from the one selected.
  3. Save an individual target to the Shortlist with a name, why it is interesting and what remains uncertain. The saved snapshot retains its original analysis basis; later screenings do not update it.
  4. Use Suggest circles · largest target regions for an editable communication outline. Suggestions follow geometric target-footprint area, not oil-volume ranking, and repeated suggestions preserve existing edits. Save form changes explicitly.
  5. Export the candidate card and the relevant map/shortlist evidence for review with colleagues. Document geological checks and the proposed next decision.

A circle can include non-target space and existing wells. It is a review note, not a drainage radius, recommended trajectory or new target boundary. Moving, resizing or recolouring a circle does not change the screening.

Keep a portable record

Shortlists and circle notes are local to this browser and website. Export JSON before clearing site data or moving browser. PNG records the current map view with its legend and notes; JSON records analysis/view basis and evidence, not the full simulator dataset. This version does not import these JSON records. Export screening CSV for all mapped areas, and shortlist CSV for a review summary. Overlapping shortlist entries are not independent volumes and must not be added.

For each candidate record the source/reference, selected cases, requested/actual dates, K interval, map width, criteria, support, coverage and property sample counts. Include geological reasoning and the next check, not just a favourable number.

10. When a result is unexpected

No targets, or low coverage
Inspect case exclusions and selected-area evidence. Distinguish missing inputs, unavailable geometry and known inactivity. Check whether every selected K contributes. Do not lower coverage merely to make targets appear.
Different results between runs
Compare folder/source metadata, case selection, date/actual restarts, K interval, map width and every criterion. “First N” uses discovery order and is not a calibrated representative sample.
Only a few wells are visible
Only recorded reference completions in the chosen scope can be positioned. Enable other-K context if appropriate. A well without recorded completions cannot be located by these symbols.
Coordinates seem wrong
Inspect Coordinate check and compare a known cell/well with Map Explorer or your model viewer. Model X/Y are in metres here; no geographic CRS or EPSG code is inferred.
Saved ensemble is unavailable
Use the module's folder-selection button. Browser permissions or storage policy may prevent reuse. Saving a choice does not grant permanent access or save live numerical results.
A large analysis reaches a limit
Use fewer cases, a narrower K interval or a coarser XY map, as appropriate to the study. Preserve the analysis basis when comparing. Close unused workspace tabs to release their live memory; refreshing loses their live results.

Norwegian opportunity-finder reading guide · Use your browser's Print command to save this English guide as PDF.