Tableau / Advanced: quality and governed delivery / Lesson 13

Extracts, live connections and freshness

Choose data-access and refresh contracts from source capability, usable-data age and ownership rather than equating live with always current.

About 4 min readContent reviewed 2026-10-28

On this page

Lesson goals

Compare extract copies with live source queries.
Separate full/incremental strategy from source completeness.
Define hosted/private-network connectivity responsibilities.

Identify the data-copy boundary

An extract stores selected data separately from the original source, commonly in Hyper format, while a live connection queries a supported source under its connection behavior. An extract can support analysis without the original connection during local use, but its freshness depends on processing. Live still depends on source update timing, query latency, caching and access. Neither label guarantees a user-visible result is current. Define source cutoff, expected query response and permitted data population separately. The first record contrasts an extract 60 with a later source 75 as synthetic states, not captured engine results or a performance comparison.

text Synthetic extract/source cutoffs

extract data cutoff T1:Sales60
later source cutoff T2:Sales75
separate timestamps:source readiness, extract completion, view observation
review:which copy and which permitted population answered the query?

The numbers are authored teaching values. No extract file, live connector or hosted source is opened.

Define refresh and change policy

Full refresh and supported incremental configurations have different processing scope. An incremental key or range design must account for late rows, corrections, duplicates and deletions according to the selected feature/version. Do not promise that every incremental strategy repairs every historical change. Extract filters and aggregation can also exclude data or remove detail permanently from that copy. Record retained columns, grain and refresh rules. A successful task cannot restore facts that the source had not yet supplied, and a recent completion timestamp is not the same as the newest usable business event.

Separate local and hosted connectivity

A Desktop connection working on an author’s machine does not establish that Tableau Cloud or Server has an equivalent path and identity. Private-network connectivity can involve supported Tableau Bridge scenarios for Cloud, while Server has its own deployment and connection responsibilities. Connector and authentication support must be checked for the exact source and environment. This course installs no Bridge client, gateway or server and reads no credentials. Ownership should cover source mapping, scheduled processing, credential lifecycle and recovery so a personal laptop or one author’s memory does not become the unreviewed operational dependency.

Measure usable-data age

The second program subtracts a synthetic usable-data cutoff from an observation time. A refresh can finish at 10:00 using 09:20 data and be observed at 10:10, producing 50 minutes of data age. Explicit UTC timestamps avoid an unstated locale assumption in this tiny model. The program does not query refresh history or prove source completeness. Pair freshness with counts and amount reconciliation: a newer timestamp can conceal missing rows. A reader-facing last updated label should state whether it means processing completion or usable data cutoff, because those support different decisions.

python Synthetic usable-data age

from datetime import datetime, timezone
observed = datetime(2026, 2, 2, 10, 10, tzinfo=timezone.utc)
cutoff = datetime(2026, 2, 2, 9, 20, tzinfo=timezone.utc)
print(int((observed - cutoff).total_seconds() // 60))

Expected output

50 minutes is timestamp arithmetic only. No Tableau task, source or service is queried.

Review access-pattern evidence

A handoff records source identity, live/extract choice, extract grain, change policy, source cutoff, task result, connectivity owner and unresolved quality issues. Licenses, schedules and feature limits depend on actual entitlement and current documentation; no universal frequency or compression promise belongs in the contract. This course executes only UTC subtraction and no connector or refresh. A later authorized review must validate actual data access and processing separately from worksheet calculations. Choose the access pattern from requirements and evidence rather than claiming all live views are real time or all extracts are safely independent backups.

Common errors and fixes

A live label is reported as guaranteed real time

Why it happens: Source readiness, caching and query behavior were ignored.
How to fix it: Define and verify the actual usable-data freshness path.
Prevent it next time: Keep access mode and cutoff evidence separate.

An old correction is assumed covered by every incremental refresh

Why it happens: Change handling was not part of the refresh design.
How to fix it: Use the supported correction/backfill strategy for the exact configuration.
Prevent it next time: Test late rows and historical updates explicitly.

Desktop success is treated as hosted connectivity proof

Why it happens: Different network paths and identities were conflated.
How to fix it: Review approved hosted source mapping and ownership evidence.
Prevent it next time: Validate local and hosted access independently.

Check your understanding

1. Why can extract 60 and later source 75 both be arithmetically correct?

Show worked solution

They can represent different cutoffs. Decide whether the extract meets the promised freshness and coverage before calling its calculation wrong. Verify source readiness and actual processing separately. A correct total over stale or filtered data can still be unsuitable for a current decision.

2. What does 50 minutes in the age program not establish?