Repository navigation
feat: improve data app starters and chart exploration - #162
Conversation
Give finding apps a primary trend, a business breakdown, and secondary detail data. Keep the original absolute date and group scope visible, distinguish exploration from display changes, and reset all controls together. Bind captions to each displayed result so refresh failures cannot relabel retained data. Keep metric queries semantic and preserve the existing group drill. Share responsive compositions across hosted and local starters, and let widget controls wrap before compressing heading words. Cover the four starters through public browser workflows.
Guide finding and metric authors to choose evidence from the available data without imposing a fixed section sequence. Explain current-data queries with fixed dates and preserve semantic aggregation for rates and averages. Keep responsive implementation guidance in the SDK documentation: use existing grids first, adapt custom layouts through ancestor container queries, and verify narrow embeds inside wide browser windows.
Remove the duplicate finding and metric UX policy from the SDK authoring guide. Keep API mechanics with the SDK and evidence/composition decisions with the app brief and skills. Preserve the container-layout documentation and link to it from the composition mechanics.
Share the gallery's pinned mock and seed data with public consumer tests. Use unequal segment populations to verify conversion numerators, denominators, calendar comparisons, additive revenue by day, and weighted non-additive metrics through executed SQL. Exercise four test-only scenarios with the existing browser harness and public SDK interfaces. Check filters, displayed scope, finding reset, and comparison disabled without inventing a previous value. Keep the starters focused and document Docker as a test prerequisite.
Avoid copying the original dates, reporting zone, and selected group into prose that can drift when agents adapt a starter. Derive the initial label from the same UTC calendar bounds and filter defaults used by the controls. Use the existing date formatter and filter description API in hosted and local finding templates. Preserve queries, displayed-result scope, reset behavior, and responsive composition.
Move finding details into URL-backed Overview and Data tabs and remove the persistent scope paragraph. Keep exploration notices and reset controls conditional on changed filters, while each result retains its own scope. Use labeled rankings for categorical breakdowns, sort metric segments by sample count, and show counts beside the interactive group labels. Replace implementation instructions in preview headers with user-facing descriptions. Keep hosted and local templates aligned. Cover keyboard tab navigation, reset behavior, stale-result scope, and metric group drilling in browser tests.
Replace the finding's loose exploration text and reset button with Notice, a responsive surface that explains the original scope and presents the return action together. Keep baseline labels derived from the declared defaults. Remove repetitive scope captions from the finding and metric starter widgets. Failed-refresh feedback still identifies the displayed population, while the chart retains its prior values until replacement data is ready. Export Notice through the authoring and UI entry points with explicit styles, theme tokens, a polite live announcement, and the generated API reference. Keep hosted and local starters aligned and cover reset, tabs, drilling, and failed-refresh scope through the public browser workflows.
Handle empty aggregate rows and unavailable comparison periods explicitly. Keep measured zeroes valid and exercise selectable empty periods through the public rate Finding workflow.
Define the sample metric by its additive aggregation rather than observed fixture values. Identify the grouped measure in chart titles, compact group labels for narrow embeds, and remove authoring instructions from user-visible context. Keep hosted and local compositions aligned.
Open the dated finding on March 9–15 with its previous-period comparison. Show the trend across both windows, compare their totals, and break the change down by member. Keep exact counts and deltas in the Data tab. Resolve every query window from the same declared filter. Preserve missing member counts as unavailable, retain measured zeroes, and omit comparison columns and series when comparison is off. Keep hosted and local query, operation, and UI composition aligned. Describe the data and derived change in user-facing evidence context rather than showing authoring instructions. Validate the default totals, group scope, member changes, reset, and comparison-off paths through public SQL and browser workflows.
Replace the local connectivity screen with the same self-contained sample exploration as the hosted starter. Show a filtered total and group counts, use the system appearance, and keep data definitions separate from observed fixture values and authoring instructions. Treat an unmatched selection as unavailable rather than measured zero. Let the primary view own the overview's empty and initial-error states while the total keeps its own query and result. Omit presentation and export setup from the default app; keep those public workflows in a dedicated fixture. Make bridge connection probes explicit fixture-owned queries, update gallery query handling and copied-app hot reload, and retain regression coverage for CSV/presentation snapshots during failed filter changes. Update local file ownership and authoring documentation for the complete starter.
Prevent the SDK build from inlining NODE_ENV into serveLocalApp. Production previews can disable HMR while keeping Bun JSX output aligned with the selected React runtime. Exercise the compiled server with a production React page and a working appearance control.
Lead with an interactive bar chart of four ranked groups, with a shared total and an exact-value table. Keep hosted and local starters aligned and preserve measured zero. Treat prior empty results as loading or error during new requests, so nullable metric selectors never receive absent content as stale data. Verify chart filtering and empty-result recovery through both runtimes.
Offer the available groups and an explicit All groups selection instead of exact-match free-text search. Keep chart actions, typed operation inputs, and query parameters aligned in hosted and local starters. Verify group selection, measured zero, reset, and refresh after an empty result. Static checks, the browser suite, packaged consumers, and copied starter checks pass.
Replace generic sample counts with a fictional September infrastructure cost investigation. Services open scoped customer allocations while keeping the overview available, with query parameters and additive cost definitions accessible through inspection. Keep hosted and local starters aligned. Store allocations in integer cents, format the derived USD amounts consistently, and remove the redundant search and summary table. Verify the service-to-customer journey, scoped evidence, zero, empty and failed requests, production rendering, and phone navigation. Repository checks, packaged consumers, and copied starter builds pass.
Support count, ratio, and currency scales on interactive bars. Keep grid lines subtle, emphasize zero, and measure the axis gutter for aligned interactions. Give category labels room on narrow screens and expose scroll anchoring while preserving the latest-sample default.
Show three charts and three summary metrics from fictional weekly cost allocations. Add month and customer filters, a service or customer breakdown, and customer details that inherit the active scope. Keep hosted and local starters aligned, document placeholder facts at their source, and verify filtering, derivations, and drill-down through public browser workflows.
|
Reviewed Merge is still blocked by conflicts in the local standalone/finding |
Merge origin/main without rewriting the published exploration branch. Derive starter evidence from operation query names and pass the local smoke fixture through the shared app declaration contract. Preserve the scoped dashboards and finding comparisons while incorporating the current component and focus behavior. Verify the complete package and starter checks on the merged result.
albert20260301
left a comment
There was a problem hiding this comment.
Reviewed the integration merge at 4909bdc. The three conflict resolutions retain the expanded starter queries through the shared operation query-name API and keep the current local-server contract; the merged main changes preserve the focus-ring fixes.
Local static checks, build, 131 focused tests across four files, and all three starter gates pass. Required CI is now green, including all five Docker-backed business-evidence tests. This resolves the conflicts and verification gaps in my earlier comment. The prior chart, narrow-embed and theme verification covers the unchanged feature code.
Distinguish independent query results from related metrics sharing a summary. Frame sample-count snippets as illustrative examples and use the current choice-filter and operation-client pattern. Clarify shared detail contexts and remove unused summaryViews from both cost starters, reserving eager summary subscriptions for page stories and exports.
Describe filter declarations as shared by browser operations and controls, removing the obsolete claim that starter operations run on the server.
Identify fictional sources, currency, coverage, and fixed demo choices at their replacement points. Explain shared filters, service pivots, and finding reset behavior using the actual code contracts. Remove generic layout instructions and the obsolete one-query-per-widget rule. Keep hosted and local variants aligned without changing application logic or sample values.
albert20260301
left a comment
There was a problem hiding this comment.
Reviewed the nine-file delta through 1b2b00f. The authoring guides now match the operation registry and shared-result contracts; removing the unused summaryViews entry leaves the cost summary loaded by its own visible section. Local static checks, build, six cost/starter browser tests and all three starter gates pass. Required CI is green on this head.
🤖 I have created a release *beep* *boop* --- ## [0.78.0](v0.77.0...v0.78.0) (2026-11-19) ### ⚠ BREAKING CHANGES * **apps:** unify local setup with shared app declarations ([#159](#159)) * run embedded page actions from the host ([#151](#151)) ### Features * improve data app starters and chart exploration ([#162](#162)) ([c57a557](c57a557)) * run embedded page actions from the host ([#151](#151)) ([8e496b8](8e496b8)) * show horizontal grids by default on Cartesian charts ([#157](#157)) ([bfd12a7](bfd12a7)) ### Bug Fixes * correct comparison math, date range edge cases, and stale docs ([#154](#154)) ([6ec7c6b](6ec7c6b)) * polish component states, focus rings, and alignment ([#161](#161)) ([202074e](202074e)) * resolve packed manifest dependency before publication ([#158](#158)) ([75f57eb](75f57eb)) * stacked bar tooltip inspection across sparse hourly data ([#156](#156)) ([e917ff0](e917ff0)) ### Code Refactoring * **apps:** unify local setup with shared app declarations ([#159](#159)) ([dec99a2](dec99a2)) --- This PR was generated with [Release Please](https://github.com/googleapis/release-please). See [documentation](https://github.com/googleapis/release-please#release-please).
Give hosted and local apps matching starters for three distinct jobs: investigating a dated Finding, exploring a semantic metric, and answering a standalone business question.
Use container-based layouts across page and embedded widths. Add a shared contextual notice and numeric bar-chart axes and gridlines. Preserve comparison-label capitalization, distinguish empty evidence from stale populated results, and read the local Bun server mode at runtime. Keep product composition rules in the authoring skills rather than duplicating them in SDK documentation.
Business-evidence fixtures execute SQL against DuckDB to check rates, additive and non-additive metrics, and evidence without a time dimension. Existing apps keep their saved source; the new compositions apply when creating or updating apps from these starters.