- SKILL.md Step 3 now recommends SNOWFLAKE_LEARNING_DB.PUBLIC (pre-provisioned
  learning environment) as the default, with custom DB/schema as alternative
- run_snippet.py --db default changed from CORTEX_SNIPPETS to SNOWFLAKE_LEARNING_DB
- Skill notes SNOWFLAKE_LEARNING_ROLE and SNOWFLAKE_LEARNING_WH for users on
  the learning environment path

.... Generated with [Cortex Code](https://docs.snowflake.com/en/user-guide/cortex-code/cortex-code)

Co-Authored-By: Cortex Code <noreply@snowflake.com>

Step 3 now captures TARGET_CONNECTION before deployment so tutorials
can run against any Snowflake account (not just the active CoCo account).
All snowflake_sql_execute calls and run_snippet.py use that connection.

.... Generated with [Cortex Code](https://docs.snowflake.com/en/user-guide/cortex-code/cortex-code)

Co-Authored-By: Cortex Code <noreply@snowflake.com>

- SKILL.md Act 1 no longer presents the comparison table in the main flow;
  instead adds a prompt hint: "Ask me 'Tell me about other approaches'"
- New Step 5b handler responds to that prompt with the full table + narration
- Added Tableau row to 11 snippet READMEs (asof_join, derived_metrics,
  entity_facts, multi_fact_table, multi_path_metrics, range_join,
  scoped_dataset, semi_additive_metric, shared_degenerate_dimension,
  variables, window_metrics)
- time_intelligence, materialization, tags already had Tableau coverage

.... Generated with [Cortex Code](https://docs.snowflake.com/en/user-guide/cortex-code/cortex-code)

Co-Authored-By: Cortex Code <noreply@snowflake.com>

Adds two new patterns (20 total):
- fact_as_relationship_key: computed FK fact joining sales to fiscal_quarters via CONCAT(YEAR, '-Q', QUARTER) — no physical FK column needed
- system_explain_semantic_query: SYSTEM$EXPLAIN_SEMANTIC_QUERY demos on a support ticket SV, including PRIVATE fact diagnosis and pre-flight error debugging

.... Generated with [Cortex Code](https://docs.snowflake.com/en/user-guide/cortex-code/cortex-code)

Co-Authored-By: Cortex Code <noreply@snowflake.com>

Demonstrates that SEMANTIC_VIEW() uses caller rights (not owner rights
like standard views): users need SELECT on both the SV and all base
tables. Uses a 4-role pattern (SV_OWNER, SV_CREATOR, SV_USER,
SV_USER_NO_BASE_SELECT) with separate SV and DATA schemas to show a
succeeding and a deliberately failing query side-by-side.

Requires ACCOUNTADMIN. Creates dedicated SV_CALLER_TEST
database/warehouse/roles with a cleanup block.

.... Generated with [Cortex Code](https://docs.snowflake.com/en/user-guide/cortex-code/cortex-code)

Co-Authored-By: Cortex Code <noreply@snowflake.com>

SVs (like views) use owner rights by default. The trick is making the
SV owner a role with no base table access — so owner rights alone can't
reach the data, and the caller must supply their own table privileges.
Updated comparison table and four-role pattern table accordingly.

.... Generated with [Cortex Code](https://docs.snowflake.com/en/user-guide/cortex-code/cortex-code)

Co-Authored-By: Cortex Code <noreply@snowflake.com>

- New snippet: row_access_policies (22nd pattern) — demonstrates the NULL-row
  problem when a RAP is applied to the dimension table only, and two workarounds:
  inner-join helper view and applying the RAP directly to the fact table
- SKILL.md: probe-first SNOWFLAKE_LEARNING_DB check; direct snowflake_sql_execute
  execution instead of run_snippet.py; mandatory cleanup offer at end of tutorial;
  access-control snippets use existing warehouse instead of creating a dedicated one

.... Generated with [Cortex Code](https://docs.snowflake.com/en/user-guide/cortex-code/cortex-code)

Co-Authored-By: Cortex Code <noreply@snowflake.com>

One physical DIM_DATE aliased twice — order_date_dim and ship_date_dim — giving
each role independent logical dimension names (order_year, ship_year, etc.) with
no USING clause needed. Demonstrates revenue-by-order-month vs revenue-by-ship-month
and fulfillment lag cross-tab analysis.

Also fixes SV DIMENSIONS syntax throughout: entity.logical_name AS physical_column.

.... Generated with [Cortex Code](https://docs.snowflake.com/en/user-guide/cortex-code/cortex-code)

Co-Authored-By: Cortex Code <noreply@snowflake.com>

…SING syntax

- New snippet: accumulating_snapshot — Kimball Accumulating Snapshot Fact Table
  modeling a loan origination funnel (application → review → decision → funding);
  one DIM_DATE alias, four milestone relationships, USING per stage metric;
  derived conversion rate metrics (review_rate, decision_rate, funding_rate)
- Fix multi_path_metrics/semantic_view.sql: USING clause was inverted
  (AS expression USING rel → USING rel AS expression)
- Update SKILL.md: 23 → 24 patterns, add accumulating_snapshot to table

.... Generated with [Cortex Code](https://docs.snowflake.com/en/user-guide/cortex-code/cortex-code)

Co-Authored-By: Cortex Code <noreply@snowflake.com>

Four diagnostic scenarios with exact error messages and corrected models:
1. Ambiguous path relationship — two date FKs, no USING → query-time error
2. Fan trap — header-level metric routed through bridge to fine-grain dim
3. Table with no relationship — orphaned table in TABLES clause
4. Duplicate logical names (deploy error) + overlapping synonyms (CA ambiguity)

Includes pre-deployment checklist, diagnostic cheat sheet, and Cortex Analyst
ambiguity demo comparing broken vs fixed synonym sets.

.... Generated with [Cortex Code](https://docs.snowflake.com/en/user-guide/cortex-code/cortex-code)

Co-Authored-By: Cortex Code <noreply@snowflake.com>

…itive

Scenario 5a: Reversed relationship direction — deploy-time error when FK is on
the RHS of REFERENCES (not a PK/UK). Engine catches it immediately.

Scenario 5b: Wrong cardinality (lying PK declaration) — the most dangerous
diagnostic. Declaring PRIMARY KEY on an FK column (e.g. DEAL_ID on DEAL_ITEMS)
tells the engine the relationship is 1:1, disabling the fan trap guard. The
exact query that errors on a correctly-declared model runs silently and inflates
numbers by the average items-per-parent count. Detection: compare SV total to
raw SELECT SUM() on the table.

Scenario 6: Forgotten semi-additive behavior — checklist-only, no DDL. Snapshot
metrics (balance, headcount, inventory) should never use SUM across time.
References semi_additive_metric snippet for the NON ADDITIVE BY fix.

Also: add DELETE FROM guards to seed_data.sql to prevent double-insert inflation.

.... Generated with [Cortex Code](https://docs.snowflake.com/en/user-guide/cortex-code/cortex-code)

Co-Authored-By: Cortex Code <noreply@snowflake.com>

Each snippet now has both semantic_view.sql (DDL) and semantic_view.yaml (YAML)
giving users a choice of authoring format. SKILL.md updated to ask the format
question early and branch Tutorial/Apply steps accordingly.

YAML highlights:
- Full parity for 18 patterns with no DDL-only limitations
- USING → using_relationships, NON ADDITIVE BY → non_additive_dimensions,
  PRIVATE → access_modifier: private_access, AI_VERIFIED_QUERIES → verified_queries
- 7 patterns flagged with DDL-only notes (ASOF/range joins, VARIABLES, WITH TAG,
  MAX_STALENESS, inline SQL subqueries, caller-rights roles)
- verify_only=TRUE in SYSTEM$CREATE_SEMANTIC_VIEW_FROM_YAML documented as
  pre-deployment dry-run (closes sv_diagnostics "no dry-run" gap)
- SYSTEM$READ_YAML_FROM_SEMANTIC_VIEW documented for exporting existing DDL SVs

SKILL.md changes:
- Step 0b: ask authoring format (DDL vs YAML) before any deployment
- YAML/DDL feature map table for quick reference
- Tutorial Step 4/7: branch on AUTHORING_FORMAT for file reading and deployment
- Apply Steps A2/A3/A5: branch for adapted DDL vs adapted YAML generation

.... Generated with [Cortex Code](https://docs.snowflake.com/en/user-guide/cortex-code/cortex-code)

Co-Authored-By: Cortex Code <noreply@snowflake.com>

Track 1 (canonical exports from deployed SVs):
- range_join, accumulating_snapshot, role_playing_dimensions, sv_diagnostics
  replaced with ground-truth YAML from SYSTEM$READ_YAML_FROM_SEMANTIC_VIEW
- range_join: range join IS fully supported in YAML via type: range +
  right_range: {start_column, end_column} and constraints: [{distinct_range}]
- AI_SQL_GENERATION maps to module_custom_instructions: sql_generation: in YAML
  (previously flagged as DDL-only — correction applied to range_join)

Track 2 (spot-check 3 key YAML patterns):
- semi_additive_metric: non_additive_dimensions syntax valid ✓
- ai_metadata: verified_queries syntax valid ✓
- entity_facts: fixed value_segment dimension expr (must reference private fact
  by logical name: lifetime_value, not SUM(orders.AMOUNT)) ✓
  fixed lifetime_value fact expr (SUM(orders.order_amount) not SUM(orders.AMOUNT)) ✓

.... Generated with [Cortex Code](https://docs.snowflake.com/en/user-guide/cortex-code/cortex-code)

Co-Authored-By: Cortex Code <noreply@snowflake.com>

AI_SQL_GENERATION is NOT DDL-only — it maps to
module_custom_instructions: sql_generation: in YAML.
Removed from DDL-only list in SKILL.md; added to YAML ↔ DDL feature map.
Fixed comment in 7 snippet YAML files that incorrectly said DDL-only.

BETWEEN EXCLUSIVE range join is also NOT DDL-only — it maps to
type: range + right_range + constraints: [{distinct_range}] in YAML.
Added to YAML ↔ DDL feature map; removed from DDL-only list.
ASOF remains DDL-only.

.... Generated with [Cortex Code](https://docs.snowflake.com/en/user-guide/cortex-code/cortex-code)

Co-Authored-By: Cortex Code <noreply@snowflake.com>