Home › Learn › Hard problems by layer › Semantic to UI
Hard problems by layer · 10 of 10
Semantic to UI: hard problems and anti-patterns
The hard challenges and common anti-patterns of semantic to ui in a data platform, with how each is solved and tested. This hop must serve dashboards, conversation interfaces, agents and apps the same certified numbers, cheaply and visibly fresh.
Where we are
Semantic to UI
What this hop must achieve. Serve dashboards, conversation interfaces, agents and apps the same certified numbers, cheaply and visibly fresh.
This is a hop: the step that moves data from one layer to the next. Handled by: semantic layer generators, observability, agents, the platform app.
The hard part
7 challenges, and how they are solved
Every example comes from the tyre world, a synthetic tyre company whose fourteen simulated days carry planted problems with known right answers, so each fix is tested, not just described.
| Challenge | What goes wrong | Tyre-world example | How we solve it | Verdict | Status |
|---|---|---|---|---|---|
| Dashboards re-querying raw tables UI-01 | Every open and every click scans large raw tables, and the warehouse bill grows with viewers. | A monitoring dashboard over raw platform telemetry. | Dashboards read certified definitions over compact summary tables, with stored results refreshed when data changes. assurance rules COST-04, SERV-04. Shift-left checks that guard it: SL-COST-CODE-06, SL-COST-CONFIG-01, SL-COST-CONFIG-02, SL-COST-CONFIG-05, SL-COST-CONFIG-06. | solved | designed |
| A conversation interface roaming raw tables UI-02 | Plain-language questions produce heavy or wrong queries on raw data. | Asking for OTIF straight from silver. | The conversation interface sees only certified definitions and verified functions for frequent questions. assurance rules SERV-06. Shift-left checks that guard it: SL-SEM-07. | solved | building |
| One number, many tools UI-03 | The dashboard, the conversation interface and an agent disagree. | OTIF south November. | All of them read the same certified definitions; known answers are tested. assurance rules SERV-04. | solved | building |
| Agents writing their own queries UI-04 | An agent's ad-hoc query is untested, unpermissioned and can be wrong. | A triage agent explaining a failed run. | Agents call typed tools with fixed, tested queries; anything they propose goes to the owner's gate. assurance rules SERV-06, GATE-03; decisions ADR-023 Agents propose, gates decide; agents follow the same kernel and adapter split. Shift-left checks that guard it: SL-SEM-07. | solved | designed |
| The cost of looking UI-05 | Watching and asking cost more than the pipelines themselves. | A monitoring template and plain-language questions over raw telemetry. | Summary tables, stored results and verified functions under monitoring and conversation screens. assurance rules COST-04. Shift-left checks that guard it: SL-COST-CODE-06, SL-COST-CONFIG-01, SL-COST-CONFIG-02, SL-COST-CONFIG-05, SL-COST-CONFIG-06. | solved | designed |
| Stale screens UI-06 | A cached screen shows yesterday's number as today's. | A dashboard cached before a restatement. | Refresh on data change; freshness and the provisional flag shown on the screen. assurance rules PERF-02, SERV-02. Shift-left checks that guard it: SL-COST-CODE-23, SL-COST-CONFIG-05, SL-COST-CONFIG-06. | solved | designed |
| Rebuilding what the platform gives UI-07 | A custom portal duplicates dashboards and catalog screens and falls behind. | Viewing dashboards in a custom app. | Native screens for viewing; our own app only for actions (alerts, approvals, quarantine release, recovery, declarations). assurance rules GATE-03. | people decide | designed |
What not to do
Common anti-patterns
An anti-pattern is a habit that looks reasonable and keeps causing the same damage. These are the ones seen most often at this hop.
| Anti-pattern | Why it hurts | Do this instead |
|---|---|---|
| Business logic in BI tools or views | The same metric gets computed differently in each dashboard, and nobody can test it. | Put calculations in gold build() or in the semantic layer's metric definitions. |
| Pointing a conversational query tool at raw silver tables | The tool then guesses joins and meanings and gives plausible but wrong answers. | Point it at metric definitions and described gold tables. |
| Agents querying tables directly | Untested, unpermissioned queries; answers that differ from the dashboard. | Typed tools over certified definitions, with proposals going to a gate. |
| A custom portal that rebuilds native dashboards | Twice the work, and it drifts from the governed screens. | Link out to native screens; build only screens that take actions. |
FAQ
Questions people ask
What are the hardest problems in the semantic to ui step?
Dashboards re-querying raw tables; A conversation interface roaming raw tables; One number, many tools; Agents writing their own queries; The cost of looking.
What is the most common mistake at semantic to ui?
Business logic in BI tools or views. Put calculations in gold build() or in the semantic layer's metric definitions.
Next
Where to go next
↑ All layers · All 69 challenges, filterable · Shift left: fail fast on data code · Medallion architecture