Exit scan: BigQuery

Scope

• Platform: BigQuery
• Scope: projects lab-a (data); regions EU, europe-west1, US
• Privilege level: 2 (+ queries (normalized forms), reads per table, policies)
• SQL mode: read-only
• Reads: 18 done out of 22
• Content read: 9 tables, 12 objects, 3 query families
• Hidden schemas or datasets read: _tmp
• Versions: canot 0.1.0, duckdb 1.5.6, sqlglot 30.22.0

Capabilities

This report is an inventory: inventoried, neither exported nor replayed.

BigQuery connector: tested; recorded test: canot drill, 2026-10-13: public dataset thelook_ecommerce (docs/ecarts.md).

What Canot cannot prove here: the declared forms (P1, P2) cover tables and queries; objects to rebuild besides tables are counted, not proven.

Observed costs

Window: 180 days. On-demand and reserved: two billing models, never added together.

On demand

Reserved

On demand and reserved do not add up: two billing models, two bases.

Storage

Movements

2 frozen, 7 live, 0 recreated, 0 not determined.

recreated: DDL less than 7 days old with no later write; live: write less than 30 days old; frozen: older last write; not determined: writes not read. Based on: B03, B04, B05, B07, B10, B11 ….

What does not leave with the copy

Objects to rebuild or describe

Routines per language: JAVASCRIPT 1, SQL 1.

Columns with a platform-specific type (3)

Billed history

Dependencies

External dependencies (F01)

Hidden external dependencies (F02)

Writes outside the platform (F03)

External consumers (F09)

Tools seen in queries

Query replayability

Based on: 3/3 families, from 2026-04-20 to 2026-10-17, sqlglot 30.22.0, duckdb 1.5.6.

accepted ≠ identical result. Translated to DuckDB, not compared.

Most executed families (normalized forms)

Ranges

Planning estimates: each range carries its assumptions. A canot drill replaces the assumed throughput with a measurement.

Findings

Each finding is labeled "To review", with no severity level. It states what is observed and what is not.

F01 · External dependencyTo review

2 findings

F02 · Hidden external dependencyTo review

1 finding

F03 · Write outside the platformTo review

2 findings

F05 · Storage with no observed readTo review

2 findings

F06 · Billed history that does not transferTo review

1 finding

F08 · Dynamic SQLTo review

1 finding

F09 · External consumersTo review

1 finding

F10 · Security to rebuildTo review

2 findings

F11 · Scheduled chain and its failuresTo review

4 findings

F12 · Incomplete readTo review

3 findings

No finding for: F04, F07. An absence of findings only holds for the reads done (see F12 and the method).

What this report does not quantify

• Average billed storage: not provided by collection.
• Storage cost: your contract price not read.
• Reserved, in dollars: your contract price not read.
• Copy read cost (Storage Read API, network egress): your contract price not read.
• Your contract rate: your contract price not read.
• Destination storage, running both platforms in parallel, operations after the exit: depend on your contract and your architecture.
• Identical results of translated queries: accepted ≠ identical result: translated to DuckDB, not compared; a canot drill compares the results.
• Cost projections provided by the platform: not used: they cannot be verified from the scan.
• BigQuery free tier: shared by the billing account, cannot be attributed per project.
• Reads through the Storage Read API: missing from INFORMATION_SCHEMA.JOBS.
• Billed bytes of queries under a row access policy: NULL value in INFORMATION_SCHEMA.JOBS.

Using this report for your regulatory work (GDPR, DORA, NIS2, Data Act, SecNumCloud, EBA)

These elements can support your own work on GDPR, DORA, NIS2, the Data Act, SecNumCloud or the EBA outsourcing guidelines. They are evidence, not a finding: this report is not a certification, it is not legal advice, and it does not say whether your contracts or your organization meet any regulation.

References as of 2026-10-19. Check the current texts: they evolve, and their application depends on your situation.

Method

18 reads done out of 22. Each read is recorded with its status and its reason; a value that was not read stays empty, never 0.

… and 2 more in reads.csv.

Privileges used

Level 2: bigquery.jobs.create, jobs.listAll, datasets.get, tables.list, tables.get, routines.list, routines.get, models.list, models.getMetadata, rowAccessPolicies.list, transfers.get, connections.list, dataPolicies.list (optional). Never requested: tables.getData, bigquery.user and jobUser roles.

Cost of the scan

Safeguards

• Every SQL statement sent goes through a read-only guard: only SELECT, SHOW, DESCRIBE and ALTER SESSION SET statements go out. No write, no API activation, no parameter change.
• Query users and roles pseudonymized with HMAC, using a salt drawn for this scan and never written; tag values hashed; locations reduced to the bucket.
• Raw SQL text is never written: only normalized forms, with literals masked, leave collection.

Rules

• Movement: recreated: DDL less than 7 days old with no later write; live: write less than 30 days old; frozen: older last write; not determined: writes not read.
• Storage with no read: actual window of the view read, compared with 180 days; the blind spots of each view are listed in the finding.
• Copy time: assumed throughput of 20 to 80 MB/s.
• List prices are dated and sourced; they do not replace your contract price.