AVEVA Historian keeps one set of stored values per tag, yet a trend, a report and an SQL query can get quite different numbers back from it. The retrieval mode decides what comes back: the stored points, one value per cycle, or a value calculated over each cycle. This article follows AVEVA's Historian 2026 documentation. What a historian does in general: What is a process historian; AVEVA Historian next to the PI System: AVEVA Historian vs PI System.
1. What is stored: storage modes and deadbands
Retrieval can only return what storage kept. Each tag uses one of four storage modes: no storage, forced storage (every collected value), cyclic storage (values at a fixed interval) or delta storage (only changed values, timed to 1 ms). Delta storage can be narrowed with deadbands on the tag's Storage tab in the Operations Control Management Console:
- Time deadband: the minimum time in milliseconds between stored values; 0 stores every change.
- Value deadband: a percentage of the tag's engineering-unit range; 0 turns it off.
- Swinging door (rate) deadband, for analog tags: a point is stored when the slope changes by more than the set percentage. A value deadband, if also set, is applied first, and a deadband override period stores the last received point once that time has passed since the last stored one.
The swinging door never invents values: only points actually received are written to disk. It keeps the value just before a change of rate, so peaks stay defined, and AVEVA positions it for steadily rising and falling signals such as a tank level rather than noisy ones. A storage change applies only to data stamped from the change onward, and since version 8.0 all history is stored in UTC.
2. The retrieval modes at a glance
Historian versions before 9.0 offer only cyclic and delta; full, interpolated, best fit, average, minimum, maximum, integral, slope, counter and ValueState came with Historian 9.0, RoundTrip with 10.0, predictive retrieval with 11.6.14101 and bounding value retrieval with 17.3.100. In SQL the mode is set with wwRetrievalMode. Cyclic is the default for the analog tables, including the summary tables; delta for the discrete and string tables and the History table.
| Mode | Returns | Tag types |
|---|---|---|
| Cyclic | the value at each cycle boundary, or the last one before it | all |
| Delta | stored values that changed | all |
| Full | every stored value, repeats included | all |
| Interpolated | a value interpolated at each boundary | analog |
| Best fit | first, last, minimum, maximum and first non-Good value per cycle | analog, analog summary |
| Average | time-weighted average per cycle | analog, analog summary |
| Minimum, Maximum | lowest or highest stored value per cycle | analog |
| Integral | area under the curve per cycle, as a total | analog |
| Slope | change per second between consecutive points | analog |
| Counter | increase per cycle, rollovers included | analog, discrete |
| ValueState | time spent in each state per cycle | integer, discrete, string, state summary |
| RoundTrip | time between leading edges of the same state | integer analog, discrete, string |
Other tag types fall back to cyclic or delta results, or return no rows, depending on the mode.
3. Raw values: delta and full
Delta returns the stored rows whose value changed; full returns every stored row. AVEVA calls full retrieval the best representation of the recorded process, and the heaviest load on server, network and client. Both return an initial value: without a point exactly at the start, the last point before it is returned at the start time with Quality 133, if the start condition uses >= rather than >. Delta can be thinned at query time with wwTimeDeadband or wwValueDeadband (a percentage of full scale); inside a value deadband, a change of quality is still returned.
4. One value per cycle: cyclic and interpolated
Cycle-based modes split the query period by a cycle count (wwCycleCount) or a resolution in milliseconds (wwResolution). With neither, Historian uses 100 cycles; with both, the cycle count is ignored.
Cyclic returns, at each boundary, the value stored there or the last one before it. It is fast, but AVEVA warns that gaps and spikes between boundaries do not show. Interpolated calculates a value at the boundary when none is stored there, for smoother trends of slow tags. It follows the tag's interpolation setting, so a stair-step tag, or any non-analog tag, gives the same result as cyclic.
5. Best fit for long trends
Best fit returns up to five stored points per cycle: first, last, minimum, maximum, each at its own time, and the first value with non-Good quality. A point that fills two roles is returned once. AVEVA's example: a week of five-second data gives 120,960 values in delta mode and about 300 in best fit.
6. Calculated values: average, minimum, maximum, integral
Average is time-weighted: each value counts for as long as it lasted, and NULL gaps do not count. AVEVA states that average retrieval with a cycle count of 1 is much faster than the SQL Server AVG aggregate and usually more accurate, since AVG is a plain statistical average.
Minimum and maximum return the lowest or highest stored value in each cycle at its real time. They need <= for the query end time; with a cycle count of 1 they match SQL MIN and MAX, faster.
Integral scales the same area by the IntegralDivisor of the tag's engineering unit, which converts the rate to units per second. AVEVA's example: a one-minute cycle averaging 3.5 litres per second returns 210.
Watch the first row of any aggregate. Historian calculates a "phantom" cycle that ends at the query start. AVEVA's example asks for an hourly average from 08:00 with a one-hour resolution and gets one row stamped 08:00: with the timestamp rule at End it is the average of 07:00 to 08:00; with wwTimeStampRule = 'Start' it covers 08:00 to 09:00. The StartDateTime column shows each row's cycle start.
7. Rates, counters and states
Slope gives the change per second from the previous point, useful for spotting a value that moves too fast.
Counter returns how much a counter rose in each cycle, using the tag's rollover value. AVEVA's example: with a rollover value of 10,000, a reading of 9,900 followed by 100 means 200 units. A counter reset by hand needs a rollover value of 0, and a counter deadband (a percentage of the range) separates resets from rollovers and filters reversals. AVEVA also notes that CTU counter tags imported into Historian default to signed integers, which roll over to the negative limit; set as unsigned, they roll over to 0.
ValueState returns the time spent in each state per cycle, chosen with wwStateCalc: Minimum, Maximum, Average, Total or Percent, each also as a Contained version that ignores occurrences cut by a cycle boundary. AVEVA's examples are how long a machine ran or a valve stayed open, How run and stop times enter OEE: OEE calculation. RoundTrip measures the time between consecutive leading edges of the same state, such as one production cycle to the next.
8. Predictive and bounding values
Predictive retrieval (wwFilter = 'SLR()') fits a simple linear regression to stored values, for example to estimate when a tank will run empty after communication is lost. Bounding value retrieval returns the first value at or before a given time, or the first after it.
9. Query options that change the result
An option set in a mode where it does not apply is ignored. The ones that most often explain a surprising number:
- Quality rule (
wwQualityRule): GOOD leaves out values with uncertain OPC quality; EXTENDED uses them; OPTIMISTIC lets a calculation mixing good and NULL values return a number, and the PercentGood column then shows how much of the cycle had good data. - Interpolation type (
wwInterpolationType): Stairstep or Linear, set system-wide, per analog tag or in the query, which overrides the others. AVEVA recommends linear in general and stair-step for exceptions such as setpoints. - Timestamp rule (
wwTimeStampRule): Start or End of the cycle, or the server default. - History version (
wwVersion, Latest by default) and time zone (wwTimeZone, by default the Historian computer's).
In the Historian Client Trend application they come as retrieval styles, such as
What we check first
This is TDL's practice, not an AVEVA requirement. When a trend or report shows a number nobody trusts, we check:
- The mode, resolution and quality rule that the trend style, report or query really uses.
- The tag's storage mode, deadbands and interpolation type, and when they last changed.
- For hourly and shift reports, the timestamp rule, the phantom-cycle first row and the time zone.
- For counters, the rollover value and whether the PLC counter came in as a signed integer.
- A short window in full retrieval, to compare calculated values with what is stored.
How TDL approaches it
TDL Industrial Automation is a System Integrator member of the AVEVA Partner Ecosystem, and our AVEVA service page lists AVEVA Historian among the AVEVA software we build, extend and maintain. For the SEHC Admin Building AHU and OAC monitoring we integrated a historian and automated reports with the AVEVA System Platform SCADA. When data stops arriving, see AVEVA Historian troubleshooting.
Frequently asked questions
Which AVEVA Historian retrieval mode suits a trend over a long period?
Best fit is AVEVA's compromise for this: it returns the first, last, minimum and maximum value and the first non-Good value of each cycle, so a week of five-second data comes back as about 300 values instead of 120,960. Cyclic is faster but can miss spikes between cycle boundaries.
Why does the 08:00 row of an hourly average report show the hour before?
The first row of an aggregate query comes from the phantom cycle that ends at the query start. With the timestamp rule at End, a row stamped 08:00 holds the average of 07:00 to 08:00; with wwTimeStampRule set to Start, it covers 08:00 to 09:00.
What is the difference between cyclic and interpolated retrieval?
Both return one value per cycle boundary. Cyclic returns the value stored there or the last one before it; interpolated calculates a value from the stored points on either side, but only for analog tags using linear interpolation; otherwise it gives the same result as cyclic.
How does AVEVA Historian count a counter that rolls over?
Counter retrieval uses the rollover value configured for the tag: with a rollover value of 10,000, a reading of 9,900 followed by 100 counts as 200 units. A counter reset by hand needs a rollover value of 0.
Does the swinging door deadband create values that were never measured?
No. AVEVA states that only points actually received from the data source are written to disk; the swinging door only decides which of them to keep.
Sources
- AVEVA, Welcome to AVEVA Historian, Historian 2026 documentation
- AVEVA, Storage modes, About delta storage mode, Time and value deadbands for delta storage, Swinging door deadband for delta storage, Benefits of the swinging door deadband, Additional options that affect the swinging door deadband and Edit storage information for an analog tag
- AVEVA, Explore retrieval modes, Cyclic retrieval, Delta retrieval, Delta retrieval - initial values, Full retrieval, Interpolated retrieval and Best fit retrieval
- AVEVA, Average retrieval, Minimum retrieval, Maximum retrieval, Integral retrieval and Integral retrieval - how it works
- AVEVA, Slope retrieval, Counter retrieval, Counter retrieval - calculations for a manually reset counter, Counter retrieval - using a counter deadband, ValueState retrieval, RoundTrip retrieval, Predictive retrieval (wwFilter) and Bounding value retrieval
- AVEVA, Which options apply to which retrieval modes?, Cycle count (wwCycleCount), About phantom cycles, Value deadband (wwValueDeadband), Time deadband (wwTimeDeadband), History version (wwVersion), Interpolation type (wwInterpolationType), TimeStamp rule (wwTimeStampRule), Time zone (wwTimeZone), Quality rule (wwQualityRule) and State calculation (wwStateCalc)
- AVEVA, Work with retrieval styles, Use the standard retrieval styles and Retrieval styles, application settings, and tag settings, Historian Client Trend
Read 15 November 2026. This article paraphrases the sources and does not quote them; read the original documents before designing or working on a system.