LensReading which lens this session carries.
All contracts

CTR-EDW · version 2

Enterprise Data Warehouse — Conformed fact rows

The honest caveat on this contract: a conformed row is already an interpretation. Where a figure resolves to the warehouse rather than to the ledger, the lineage surface says so.

RatifiedIn contract

What the contract says

These are the terms the connector is generated against. A source that stops matching them is a drift finding, not a silent data change.

Entity
Conformed fact row · inbound
Primary key
factKey
Declared volume
2,400,000 records a day
Freshness commitment
4 h
Owner
N. Ferreira · Data Platform
Vendor
Snowflake · critical criticality · modeled
Cardinality
One conformed fact to many source rows. The warehouse has already collapsed the grain, which means the platform inherits its conforming decisions and cannot audit behind them.
Write-back rule
None. The warehouse is downstream of the systems of record and the platform never writes upstream through it.

Review state

Neither review is performed by the team that wrote the contract.

Security reviewApproved
R. Osei · Jul 16, 2026
Privacy reviewApproved
Helena Marsh · Aug 3, 2026

Validation and drift

Drift is only detectable on a contract that has been read against a sample.

Last validatedAug 10, 2026 · 8 d ago
100,000 records read in the sample.
Validated against a 100,000-row sample. The largest validation run performed.

The field list

5 fields · 1 key · 0 personal. The validation rule on each row is what admits or rejects a record at the staging boundary.

FieldSourceTargetTypeFlagsDriftValidation
Fact keyFACT_KEYfact.refstring(64)
keyinternal
In contractWarehouse surrogate. Stable across reloads, which the warehouse team warrants in the contract.
GrainGRAINfact.grainstring(40)
internal
In contractDeclared per subject area. A grain change is a contract-breaking event and must bump the version.
Measure valueMEASUREfact.valuedecimal(18,4)
internal
In contractNull is not permitted. A null measure means the conforming logic failed upstream.
Effective dateEFF_DTfact.effectiveOndate
internal
In contractWarehouse effective date, not source posting date. The two differ and the contract names the difference.
Load idLOAD_IDfact.loadRefstring(24)
internal
In contractIdempotence key. A repeated load id is suppressed rather than re-applied.

What has passed the staging boundary against this contract

1 batch · 44,100 arrived · 42,860 admitted · 1,240 quarantined · 0 rejected

BatchArrivedInAdmittedQuarantinedRejectedResultGate
STG-0021Aug 18, 06:0044,10042,8601,2400Partly admitted42,860 admitted · 1,240 quarantined · 0 rejected outright.

Rejects raised against these batches

1 reason covering 1,240 records. Each one names a field, what was observed and what the contract expected.

Load identifier already applied and suppressed1,240 records · 3 h oldCleared
Observed: A load id replayed by the grain change
Expected: A load id not previously applied
STG-0021 · field loadId · queue quarantine · N. Ferreira

Replay windows

1 window recorded against this contract.

RPL-011Completed
Scheduled backfill · Aug 13, 2026 to Aug 14, 2026
A subject area was regrained from daily to hourly and the prior day was re-read at the new grain.
88,400 replayed · 88,400 held at the idempotence key · 0 duplicates created

Write-back routes into this system

No write-back route is designed into this system.

What goes stale when this feed does

3 asserted figures depend on Enterprise Data Warehouse. The rule on each row is what the surface does when the feed misses its commitment.

FigureWhere it is shownRoleOn staleNote
Transactions processed each weekAI GBS towersprimarySuppress the figureThe weekly count is read from the warehouse. Without it there is no figure to show, so the tile is suppressed rather than shown stale.
Cost per transactionAI GBS towerscontributingMark it staleThe denominator is the warehouse transaction count.
Traced facts behind unit decisionsData lineageprimaryMark it staleLineage is traced through the warehouse. A stale warehouse means the trace ends at yesterday.
What this contract is, and what it is not

A contract, a validation rule, a freshness commitment and a replay path are real mechanisms and they are what this wave builds. What they are not is a connection. No connector on this estate is attached to a customer instance, so every batch, every replay and every write-back attempt recorded here ran against a modeled extract or a sandbox, and the registry keeps saying modeled.

The last validation read 100,000 records 8 days ago, which is 4% of a single day at the declared volume. A sample proves the shape of what it read and nothing beyond it.