The V2 recommendation catalog

V1                              V2
─────────────────────────────────────────────────────────────
The Rule interface you met      A catalog-driven engine:
in M2.1: Go rules under         capability YAML under
internal/rules/, each with      providers/snowflake/*.yaml
Evaluate(), MetricsAware,       plus signal definitions in
PricingAware                    signals/*/snowflake-*.yaml

455 distinct RC-* ids           141 catalogs / ~508
553 registered instances        capabilities, of which 6
                                catalogs are Snowflake

WAREHOUSES  (the majority of recoverable spend)
  idle / suspend              a warehouse burning credits
                              it does not need to
  resize down                 oversized for its workload
  resize up                   spilling to disk, where the
                              spill dominates runtime
  multi-cluster min / max     min too high, or max wider
                              than concurrency needs
  consolidate                 several warehouses doing one
                              warehouse's work
  auto-suspend guardrail      threshold long or disabled
  statement timeout guardrail runaway queries uncapped
  resource-monitor guardrail  no credit ceiling in place
  tag guardrail               warehouse not attributable

TABLES
  retention                   Time Travel longer than the
                              table's value justifies
  transient                   a rebuildable table paying
                              for Fail-safe
  archive                     cold data on hot storage
  suspend clustering          automatic clustering running
                              on a table that no longer
                              benefits
  drop search-optimization    a search-optimization service
                              nothing queries that way

MATERIALIZED VIEWS            maintenance cost against
                              actual query benefit

SNOWPIPE                      file-batching: many small
                              files ingested individually
                              cost far more than the same
                              volume batched

ORPHAN STAGES                 stages nothing loads from

ACCOUNT LEVEL (advisory)
  capacity contract           consumption pattern versus
                              the committed contract
  edition downgrade           paying for an edition whose
                              features are unused

FIRST, ALWAYS: the warehouse signals.
  Compute is the majority of a typical Snowflake bill, and
  resize-down plus consolidate are the two levers with the
  largest single-action impact.

SECOND: the guardrails.
  auto-suspend, statement timeout and resource monitor do
  not save money today. They cap the worst case, which on
  a credit-based model is unbounded. A runaway query on an
  X-Large with no statement timeout is the Snowflake
  equivalent of a forgotten override.

THIRD: Snowpipe file-batching.
  Frequently the largest surprise, because serverless spend
  sits outside the warehouse figure most teams watch.

FOURTH: the table signals.
  Individually small, collectively material on a large
  account, and the safest to act on.

LAST: the two account-level advisories.
  Capacity contract and edition downgrade are procurement
  conversations, not engineering actions. They carry no
  one-click anything and belong in a renewal discussion.

1. WHICH WINDOW did the signal measure?
   Snowflake workloads are frequently weekly rather than
   daily. A resize-down measured over a week that
   contained no month-end close will under-size.

2. IS THE WAREHOUSE SHARED?
   The consolidate signal is the one most likely to be
   right about the money and wrong about the politics.
   Two teams sharing one warehouse lose the ability to
   attribute their own spend.

ACCOUNT: 14 warehouses, ~4,200 tables, $41,200/mo compute

OPEN SNOWFLAKE FINDINGS: 31

WAREHOUSE SIGNALS                          POTENTIAL
  resize down    WH_ETL_MAIN  L -> M          $4,100/mo
  resize down    WH_BI_PROD   XL -> L         $3,800/mo
  consolidate    WH_ADHOC_1 + WH_ADHOC_2      $1,200/mo
  idle/suspend   WH_MIGRATION_TMP             $890/mo
  multi-cluster  WH_BI_PROD min 2 -> 1        $700/mo
  ──────────────────────────────────────────────────────
                                             $10,690/mo

SNOWPIPE
  file-batching on the events pipe            $2,900/mo
  (18M files/day averaging 4 KB. Batching to
   ~64 MB targets is an ingestion change, not
   a Snowflake setting.)

TABLE SIGNALS (11 findings)                   $1,340/mo
  mostly retention on staging tables carrying
  7-day Time Travel they do not need

GUARDRAILS (4 findings)                            $0
  auto-suspend disabled on WH_ADHOC_2
  no statement timeout on WH_ETL_MAIN
  no resource monitor on the account
  WH_MIGRATION_TMP has no cost-centre tag

ACCOUNT ADVISORY (2 findings)                      $0
  edition downgrade: Business Critical features
  show no usage in 90 days -> renewal question
  capacity contract: consumption 23% under commit

WHAT THE TEAM DID
  Week 1: the two resize-downs and the idle warehouse.
          $8,790/mo, three actions, no code changes.
  Week 1: ALL FOUR guardrails, despite $0 savings.
          The missing statement timeout on WH_ETL_MAIN was
          the one that mattered: a runaway query there had
          burned $2,300 in one weekend two months earlier,
          which is how the account got audited at all.
  Week 3: Snowpipe batching, after a data-eng sprint.
  Deferred: consolidate. Both ad-hoc warehouses belong to
          teams who want their own spend visible.
          Correct call: $1,200/mo is not worth losing
          attribution over.

1. Filter Recommendations to the RC-21xx range.
   Total findings: ______   Total savings: $______

2. Split them:
   warehouse signals   ______ findings  $______
   table signals       ______ findings  $______
   guardrails          ______ findings  $0
   account advisory    ______ findings  $0

3. Take your largest resize-down. What window did the
   signal measure?                      ______ days
   Did that window contain a month-end close?   Y / N
   If N, widen before acting.

4. Check all four guardrails regardless of savings:
   auto-suspend set tight everywhere?           Y / N
   statement timeout set on ETL warehouses?     Y / N
   resource monitor on the account?             Y / N
   warehouses carry a cost-centre tag?          Y / N
   Any N is worth fixing this week.

5. If a consolidate signal appears, name the teams
   involved and decide the attribution question BEFORE
   the cost question:
   ______________________________________________