Direct answer
Drilling data is not just a sequence of sensor readings ordered by timestamp. A useful drilling record must also answer where the bit was, which wellbore was active, what operation was occurring, which BHA and hole section applied, which values were measured or calculated, and when the context became authoritative.
That distinction is not academic. A generic time-series table can plot standpipe pressure or hookload. It cannot, by itself, tell an engineer whether two intervals are comparable, whether a pressure change occurred while pumps were stable, or whether a survey belongs to the original hole or a sidetrack.
One well has several analytical axes
Time remains essential. It preserves sequence, duration, transitions, and event timing. But drilling questions frequently use other axes:
| Axis | Typical question | Common failure if context is missing |
|---|---|---|
| Event time | What happened before the alarm? | Ingest time is mistaken for measurement time. |
| Measured depth | How did ROP change through the lateral? | Off-bottom time is treated as depth exposure. |
| TVD and trajectory | Where is the well relative to target or offsets? | Interpolated geometry is presented as a measured survey. |
| Rig state | Was this value recorded while drilling, circulating, or tripping? | Incompatible operating conditions are averaged together. |
| BHA/run | Which equipment and model assumptions applied? | Performance is assigned to the wrong run. |
| Wellbore/branch | Did this station belong to the original hole or a sidetrack? | Separate trajectories are flattened into one path. |
The axes are related, but none is a safe replacement for another.
A timestamp does not define operational meaning
Suppose two rows have the same WOB, RPM, flow rate, and ROP. One may represent stable rotary drilling. The other may occur immediately after returning to bottom, before the system has stabilized. The raw values look similar; the operational meaning does not.
This is why rig-state classification is stateful. Current values, recent transitions, persistence, equipment state, and data quality all matter. An analytical system should retain both the underlying measurements and the interpreted state, including confidence and the version of the classifier that produced it.
The same principle applies to events. An alarm is not merely a timestamp. It has an onset, evidence window, peak, recovery, applicable channels, state transitions, model residuals, quality conditions, and eventually a human response.
Depth is not a second timestamp
Measured depth often repeats. During a connection, circulation, reaming, or a trip, many time samples can belong to the same depth. Bit depth can move backward. Hole depth may remain constant while bit depth changes substantially.
Reducing that history to one row per depth requires an explicit analytical rule:
- Which rig states are eligible?
- Is the value an average, median, last observation, distribution, or extrema envelope?
- How much time or footage contributed?
- Was the depth measured or interpolated?
- Which source and unit were used?
Without those answers, a depth curve may look precise while hiding how it was constructed. The next article, Two Clocks, One Well, examines those alignment choices directly.
Identity is part of the measurement
In real operations, a well name is not a durable database key. Names can repeat, change, or vary by data provider. A physical project can include a pilot hole, one or more sidetracks, plug-backs, and separate planned and as-drilled trajectories.
A reliable result therefore needs an identity envelope. At minimum it should carry:
- organization and authorization scope;
- site well identifier and source-system identifier;
- wellbore or sidetrack identifier;
- run or BHA when relevant;
- time or depth interval and timezone;
- source database or stream;
- source and canonical units;
- data-as-of time;
- calculation or model version;
- quality and lineage status.
This envelope belongs on APIs, exports, engineering curves, reports, and AI answers. It is the difference between “the query returned a number” and “the system can explain exactly what the number represents.”
Measured, calculated, and inferred values are different evidence
Drilling applications routinely combine three evidence types:
- Measured: a sensor or source system supplied the value.
- Calculated: a deterministic method transformed validated inputs.
- Inferred: a classifier, statistical model, or engineering belief estimated a state.
These categories should remain visible. A modeled pressure is not a pressure sensor. A rig-state belief is not a manually verified activity. An interpolated survey station is not a new survey.
The distinction becomes even more important when an LLM explains results. The language layer may choose a deterministic tool and explain its output, but it should not blur measured and inferred evidence or recreate the calculation from prose.
Live and historical data are two stages of one record
Live systems optimize for low latency and recent operational queries. Historical systems optimize for retention, reprocessing, and analytical scans. The transition between them introduces late records, duplicates, corrections, and recalculated channels.
The architecture needs a declared reconciliation policy:
- identify records by source and event time;
- make ingestion idempotent;
- preserve the original value when a canonical conversion is added;
- version calculated outputs;
- record archive cutover and data-as-of state;
- expose gaps instead of silently filling them;
- make reprocessing reproducible.
A dashboard and a post-well query should not disagree merely because they read different storage tiers.
A practical drilling-data contract
Before an analytical result is trusted, it should be possible to answer the following:
- Which authorized well and wellbore does this represent?
- Is the interval defined by time, MD, TVD, event, run, or state?
- Which observations contributed, and which were excluded?
- What are the source and display units?
- Is the value measured, calculated, or inferred?
- How fresh is the source?
- Which calculation, model, or classifier version was used?
- What quality limitations remain?
That contract is the foundation for real-time monitoring, offset comparison, engineering models, directional analytics, reporting, and evidence-grounded AI. It also explains why DrillingMetrics treats time traces, depth-aligned offsets, rig states, surveys, engineering calculations, and wellbore construction as connected parts of the same platform rather than unrelated charts.
Where to go next
- Read Two Clocks, One Well for time/depth alignment rules.
- Read The Average Trap before ranking wells or runs.
- Explore real-time drilling analytics and historical offset analysis to see how the contexts become operational workflows.