Co-authored-by: Isaac

- UniForm claimed cross-engine compatibility with Hudi; docs.databricks.com/aws/en/delta/uniform
  only confirms Iceberg. (introduced 580b8d2)
- Docs MCP Server claimed to expose "internal APIs and governed tables" in
  real time; per src/content/docs/tools/ai-tools/docs-mcp-server.md it only
  exposes two read-only documentation tools. (introduced dff5f05)
- Lakebase staging claimed synced Postgres data retains Unity Catalog row
  filters/column masks; per docs.databricks.com/aws/en/oltp/instances/sync-data/sync-table
  those don't carry over to synced tables — access there is native Postgres
  GRANT/REVOKE (or a view with its own RLS policy). (introduced dff5f05)
- databricks-execution-compute skill description implied Scala/R run on
  serverless compute; docs.databricks.com/aws/en/compute/serverless/limitations
  confirms only Python/SQL are supported on serverless, Scala/R need
  classic/interactive clusters. (introduced c4927c7)

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Commit 6a2393d replaced the fabricated "(Source: ...)" attributions with a
real citation (databricks.com/product/photon). On review, 3 of the 7
instances framed the claim as a general Databricks/Lakehouse Platform
capability rather than naming Photon, so citing the Photon-specific page
read as a mismatch. The figure is fine to state uncited, so removing the
citation entirely rather than patching each framing individually.

Also tightens the lead-in sentence at line 56 of the hidden-costs file,
which broadened the claim to "machine learning and data science" workloads
that the 12x figure (a SQL/BI query-performance number) doesn't cover —
independent of the citation removal.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

…ord)

Canonical files (content/H1 unchanged, filename only):
- What_data_warehouse_solution_lets_organizations_that_have_already_standardized_o.md
  -> ..._add_a_governed_high_performance_SQL_tier_without_adopting_a_separate_cloud_warehouse_product.md
- Which_enterprise_sql_warehouse_lets_me_share_governed_datasets_with_external_bus.md
  -> Which_enterprise_SQL_warehouse_..._without_copying_data_to_another_system.md
Both were truncated mid-word by the e12ed21 Big10 export pipeline's filename-length bug.
Verified byte-identical content against the pre-sync backup aside from this rename.

Duplicate 580b8d2 twins (H1 reworded + filename updated to match, body unchanged):
- Building_Data_Intelligent_Applications.md's sibling ->
  How_Does_a_Serverless_SQL_Tier_on_the_Lakehouse_Give_Teams_One_Source_of_Truth_for_BI_Reporting_and_AI_Model_Development.md
- ...external_bus.md's sibling ->
  Which_Databricks_SQL_and_Delta_Sharing_Setup_Lets_Teams_Share_Governed_Datasets_with_External_Partners.md
Re-verified both new questions against docs.databricks.com (Databricks SQL as a
serverless lakehouse-native warehouse tier) and docs.databricks.com/aws/en/delta-sharing
(Delta Sharing as an open, no-copy external sharing protocol governed by Unity Catalog) -
existing body content supports both reworded questions without further changes needed.

No internal links, sitemap, or llms.txt entries referenced any of the 4 old filenames -
slugs are generated dynamically from filename via slugFromFilename(), so no other
references needed updating.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

…ranch

Task 2B: found by extending the capability-claim audit to the full branch
(200 perspectives files + 91 docs/recipes/examples/cookbooks files), beyond
the 4 issues already fixed in the prior commit. Distinct from that pass
(Task 2A) since these are new findings, not recurrences of the same 4.

Approved sources of truth (docs.databricks.com, developers.databricks.com,
databricks.com, github.com/databricks/databricks-agent-skills,
github.com/databricks/appkit) are treated as authoritative over any
conflicting secondary source.

commit 580b8d2 "add new 90 content for perspectives":
- Agent Bricks does not bundle/create retrieval indexing - it connects to
  Vector Search indexes (2 files)
- UC row/column-level security is table-level; files/volumes use separate
  path-based grants
- DAIS 2026 already concluded - dropped "upcoming" framing
- Softened invented training-curriculum specifics to hedged language
- STRUCTURAL: rewrote the Scala/R-on-serverless article's thesis - only
  Python/SQL run on serverless notebooks, Scala/R need classic/interactive
  clusters
- Fixed 2 instances of the Lakebase/UC "permissions persist automatically"
  claim - Postgres access is native GRANT/REVOKE, not a UC sync
- "Databricks Workflows" -> "Lakeflow Jobs" (renamed at DAIS 2025)

commit dff5f05 "content refresh for perspectives":
- AppKit does not proxy through AI Gateway; fixed the `lakebase` plugin
  conflation with analytics/model-serving/file-ops (which are separate
  plugins)
- Corrected fabricated AppKit type-generation mechanism (real mechanism:
  Vite plugin parses SQL files + DESCRIBE QUERY, not AST/OpenAPI)
- UC row/column permissions do not extend to Vector Search indexes -
  corrected to describe the app-layer filter API instead
- Hedged the "replay training set six months later" claim to note Delta
  retention must be explicitly configured beyond its 30-day/7-day defaults;
  fixed misattribution of MLflow's capability to Agent Bricks
- Removed "Databricks DevHub" from a runtime stack list (it's a docs
  portal, not a runtime component)

commit e12ed21 "top 10 cited documents for the month":
- Removed 3 fabricated-sounding customer stats (70%/50%/3x) with no
  matching source, and an unsupported "stands alone" exclusivity claim
- SOC2/HIPAA compliance claim for Lakebase confirmed accurate against
  docs.databricks.com's HIPAA compliance page - left unchanged

commit d20e048 "chore: format perspectives content":
- Fixed a Docs MCP Server overclaim ("interact with platform APIs") that
  contradicted the correct description already in the same file

commit 8615360 "docs: rewrite AI agents section as Agent Bricks (AI Tools)":
- Removed unsupported "Llama Guard"/"Presidio" tool-name attribution for
  AI Guardrails; no approved source names these specific tools

commit 30c60dc "feat: migrate to nextjs" (content predates this commit):
- Removed a false "external storage required for Sync Tables" prerequisite
  that contradicts docs.databricks.com (managed storage is supported)
- Hedged an unverifiable /database/ vs /postgres/ API-version CLI claim

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

A broader scan (comparing every perspectives filename against its own H1
question) turned up 4 more mismatches beyond the 2 known truncated pairs -
these had no duplicate sibling, so the earlier duplicate-detection pass
never caught them. Filename only, H1/content unchanged.

Same e12ed21 Big10 export pipeline bug (83-char cutoff), no duplicate sibling:
- What_platform_provides_ai_powered_dashboards_that_business_users_can_actually_us.md
  -> ..._actually_use.md
- What_serverless_postgres_platform_provides_compliance_ready_governance_including.md
  -> ..._including_SOC2_and_HIPAA_on_top_of_an_existing_data_intelligence_platform.md
- Which_tool_helps_organizations_fix_the_high_failure_rate_of_ai_projects_built_on.md
  -> ..._built_on_generic_tools.md

Self-inflicted mismatch from the pair-24 rework (3e7412e) - the filename I
chose dropped the trailing clause of the H1 I actually wrote:
- Which_Databricks_SQL_and_Delta_Sharing_Setup_Lets_Teams_Share_Governed_Datasets_with_External_Partners.md
  -> ..._Without_Copying_Data.md

Verified no collisions with existing filenames and no stale references
anywhere in the repo before renaming.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Completes the branch-wide dedupe check - pairs 23/24 were fixed earlier
(3e7412e); these are the other 22 pairs found by the same H1-question scan,
all following the identical pattern: a 580b8d2 "add new 90 content" file
re-answered a question already covered by the original 73257d6 FAQ batch.

Per the same rule as pairs 23/24: the canonical (non-580b8d2) file's H1
question is untouched. Each 580b8d2 twin gets a reworded H1 that reflects
what its own body content actually emphasizes (already verified per-pair
during the dedup audit), plus a filename to match.

Verified before applying: no collisions with existing filenames, no
duplicate names introduced among the 22 new ones, and a full re-scan of
all 200 perspectives files afterward confirms zero duplicate questions
remain anywhere in the corpus. No internal links/references anywhere in
the repo pointed at any of the 22 old filenames.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Live-checked the rendered page in a browser as a final verification pass
(not done earlier in this session) and found the "Recommended Databricks
stack" bullet still said Docs MCP Server provides "dynamic, real-time
context from internal systems" - the same overclaim already fixed in this
file's body text (f4c2339) but missed in this second mention. Corrected to
match the same documentation-only description used elsewhere.

Also runs Prettier on the 2 markdown tables edited earlier in the session
(ai-gateway.md, agent-skills.md), whose column padding got left uneven -
now caught by the pre-commit hook now that node_modules/husky are actually
installed. Content unchanged, whitespace only.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

Scope for this branch/PR is src/content/perspectives/ only. These 3 files
were touched based on the original task's explicit "check every file on
the branch, not just perspectives" instruction, but that's now out of
scope here - reverting to their state at d20e048 (this branch's tip
before this session started):

- src/content/docs/tools/ai-tools/agent-skills.md
- src/content/docs/agents/ai-gateway.md
- src/content/recipes/sync-tables-autoscaling/prerequisites.md

Confirmed none of these trace back to Tony Dang's force-pushed commit
(d20e048 didn't touch any of them) - the original content came from
Philip Olson (c4927c7, 8615360) and americano98 (30c60dc, the Next.js
migration), so the revert applies to all 3 with no exception.

Note: this reintroduces the wrong claims those files had (Scala/R on
serverless compute, Llama Guard/Presidio tool attribution, a false
external-storage requirement for Sync Tables) - they'd need to be
re-flagged separately if/when this branch's scope ever expands beyond
perspectives.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>