Enterprise Agent Deployment Pitfall: YAML Import Replaces — Not Merges — Shared Memory Rules
ThoughtSpot Spotter 3 saves conversation corrections as shared memory rules, but its YAML migration API replaces the target model's entire rule set instead of merging, risking loss of existing business logic like financial definitions when deploying new fixes.
ThoughtSpot's Spotter 3 Data Agent allows users with write permission to correct query misunderstandings during conversation, saving those corrections as shared memory rules attached to a data model. In a retail example, a user asked for "best-selling products" and Spotter returned a ranking by sales amount; the business team corrected it to rank by quantity sold, and Spotter stored this as a shared rule for all users of that model.
Field Selection Determines Business Alignment
The retail data model contains both sales (revenue) and quantity purchased (units sold) columns. Spotter initially chose sales because "selling" linguistically resembles "sales," producing a valid but business-misaligned ranking. A high-priced coat may top the revenue list while a low-margin shirt leads in units. The article emphasizes that even with correct table names and executable queries, picking the wrong metric deviates from business intent.
Shared Memory Captures Business Conventions and Query Recipes
Beyond simple term-to-metric mappings, Spotter 3 also supports query recipes derived from validated Liveboards — predefined field sets, granularities, and filters for recurring report types. As teams accumulate rules across domains (store, product, finance), the article notes a documented behavior: even if the underlying data model later changes column names or metric definitions, old memories remain attached to the model. Teams must verify not only that rules are saved but that they still apply to the current schema.
YAML Migration API Uses Full Replacement, Not Merge
Starting in version 26.8.0.cl, ThoughtSpot provides a Beta migration API to export shared memory from one environment (e.g., dev) as YAML and import it into another (e.g., prod). The import operation replaces the target model's entire shared memory with the file's contents. In a staged scenario, the production model already holds two rules: "net sales excludes refunded amounts" and "week starts on Monday." The dev export contains only the new "best-selling defaults to quantity" rule. Importing that file removes the two existing production rules.
Pre-Deployment Diff and Dry-Run Validation
To avoid silent rule loss, the article recommends exporting the production model's current shared memory as a baseline, merging dev-validated rules into that file, and reviewing the diff — only the new sales rule should appear as a change. ThoughtSpot's migration API offers a dry_run endpoint that returns the target model's current rule count, planned deletions, and planned insertions. Submitting only the new rule shows two deletions and one insertion; submitting all three rules shows zero deletions and three insertions (re-writing the existing ones).
Automation must not rely solely on HTTP 200; the response body includes a status field that can read VALIDATION_FAILED despite a 200 code, indicating structural or reference errors in the YAML.
Post-Deployment Query Verification with Divergent Test Data
Even a successful dry_run and import do not guarantee semantic correctness. Field references in rules may resolve in dev but map differently in prod (e.g., a "refunded" flag column vs. an order-status derivation). The article prescribes re-running the corrected query ("best-selling products") against a sample where revenue and unit rankings differ, confirming Spotter uses quantity purchased. It also demands re-testing previously working queries like "this week net sales" with cross-week data containing refunds, verifying week-start and refund-exclusion logic. If test data happens to align (e.g., no refunds that week, or revenue and unit rankings coincidentally match), a broken rule can still produce seemingly correct results — so test cases must expose the semantic boundaries.
Reference sources: ThoughtSpot official case study on Spotter metric selection; ThoughtSpot docs on conversation-generated shared memory; ThoughtSpot docs on Spotter memory and query recipes; ThoughtSpot developer docs on Spotter memory migration API; ThoughtSpot official docs on cross-environment shared memory migration.
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
