fix: harden dashboard refresh flow and complete cleanup pass

cmd/demo/main.go

@@ -6,6 +6,7 @@ import (
 	"io"
 	"log"
 	"os"
+	"sync/atomic"
 	"time"
 
 	tea "github.com/charmbracelet/bubbletea"
@@ -39,6 +40,7 @@ func main() {
 
 	ctx, cancel := context.WithCancel(context.Background())
 	defer cancel()
+	var snapshotRequestID atomic.Uint64
 
 	refreshAll := func() {
 		snaps := make(map[string]core.UsageSnapshot, len(accounts))
@@ -61,7 +63,11 @@ func main() {
 			}
 			snaps[acct.ID] = snap
 		}
-		p.Send(tui.SnapshotsMsg(snaps))
+		p.Send(tui.SnapshotsMsg{
+			Snapshots:  snaps,
+			TimeWindow: core.TimeWindow30d,
+			RequestID:  snapshotRequestID.Add(1),
+		})
 	}
 
 	go func() {

cmd/openusage/dashboard.go

@@ -14,6 +14,7 @@ import (
 	"github.com/janekbaraniewski/openusage/internal/config"
 	"github.com/janekbaraniewski/openusage/internal/core"
 	"github.com/janekbaraniewski/openusage/internal/daemon"
+	"github.com/janekbaraniewski/openusage/internal/dashboardapp"
 	"github.com/janekbaraniewski/openusage/internal/tui"
 	"github.com/janekbaraniewski/openusage/internal/version"
 )
@@ -31,6 +32,9 @@ func runDashboard(cfg config.Config) {
 
 	timeWindow := core.ParseTimeWindow(cfg.Data.TimeWindow)
 
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
 	model := tui.NewModel(
 		cfg.UI.WarnThreshold,
 		cfg.UI.CritThreshold,
@@ -39,6 +43,7 @@ func runDashboard(cfg config.Config) {
 		cachedAccounts,
 		timeWindow,
 	)
+	model.SetServices(dashboardapp.NewService(ctx))
 
 	socketPath := daemon.ResolveSocketPath()
 
@@ -47,12 +52,10 @@ func runDashboard(cfg config.Config) {
 		socketPath,
 		verbose,
 	)
-	viewRuntime.SetTimeWindow(string(timeWindow))
-
-	ctx, cancel := context.WithCancel(context.Background())
-	defer cancel()
+	viewRuntime.SetTimeWindow(timeWindow)
 
 	var program *tea.Program
+	dispatcher := &snapshotDispatcher{}
 
 	model.SetOnAddAccount(func(acct core.AccountConfig) {
 		if strings.TrimSpace(acct.ID) == "" || strings.TrimSpace(acct.Provider) == "" {
@@ -96,16 +99,11 @@ func runDashboard(cfg config.Config) {
 		}
 	})
 
-	model.SetOnRefresh(func() {
-		go func() {
-			snaps := viewRuntime.ReadWithFallback(ctx)
-			if len(snaps) > 0 && program != nil {
-				program.Send(tui.SnapshotsMsg(snaps))
-			}
-		}()
+	model.SetOnRefresh(func(window core.TimeWindow) {
+		dispatcher.refresh(ctx, viewRuntime, window)
 	})
 
-	model.SetOnTimeWindowChange(func(tw string) {
+	model.SetOnTimeWindowChange(func(tw core.TimeWindow) {
 		viewRuntime.SetTimeWindow(tw)
 	})
 
@@ -118,6 +116,7 @@ func runDashboard(cfg config.Config) {
 	})
 
 	program = tea.NewProgram(model, tea.WithAltScreen(), tea.WithMouseCellMotion())
+	dispatcher.bind(program)
 
 	go func() {
 		runStartupUpdateCheck(
@@ -139,8 +138,8 @@ func runDashboard(cfg config.Config) {
 		ctx,
 		viewRuntime,
 		interval,
-		func(snaps map[string]core.UsageSnapshot) {
-			program.Send(tui.SnapshotsMsg(snaps))
+		func(frame daemon.SnapshotFrame) {
+			dispatcher.dispatch(frame)
 		},
 		func(state daemon.DaemonState) {
 			program.Send(mapDaemonState(state))

cmd/openusage/snapshot_dispatcher.go

@@ -0,0 +1,43 @@
+package main
+
+import (
+	"context"
+	"sync/atomic"
+
+	tea "github.com/charmbracelet/bubbletea"
+	"github.com/janekbaraniewski/openusage/internal/core"
+	"github.com/janekbaraniewski/openusage/internal/daemon"
+	"github.com/janekbaraniewski/openusage/internal/tui"
+)
+
+type snapshotDispatcher struct {
+	program *tea.Program
+	nextID  atomic.Uint64
+}
+
+func (d *snapshotDispatcher) bind(program *tea.Program) {
+	d.program = program
+}
+
+func (d *snapshotDispatcher) dispatch(frame daemon.SnapshotFrame) {
+	d.send(frame, d.nextID.Add(1))
+}
+
+func (d *snapshotDispatcher) refresh(ctx context.Context, rt *daemon.ViewRuntime, window core.TimeWindow) {
+	requestID := d.nextID.Add(1)
+	go func() {
+		frame := rt.ReadWithFallbackForWindow(ctx, window)
+		d.send(frame, requestID)
+	}()
+}
+
+func (d *snapshotDispatcher) send(frame daemon.SnapshotFrame, requestID uint64) {
+	if d == nil || d.program == nil || len(frame.Snapshots) == 0 {
+		return
+	}
+	d.program.Send(tui.SnapshotsMsg{
+		Snapshots:  frame.Snapshots,
+		TimeWindow: frame.TimeWindow,
+		RequestID:  requestID,
+	})
+}

cmd/openusage/telemetry.go

@@ -12,8 +12,6 @@ import (
 	"github.com/janekbaraniewski/openusage/internal/daemon"
 	"github.com/janekbaraniewski/openusage/internal/detect"
 	"github.com/janekbaraniewski/openusage/internal/integrations"
-	"github.com/janekbaraniewski/openusage/internal/providers"
-	"github.com/janekbaraniewski/openusage/internal/providers/shared"
 	"github.com/janekbaraniewski/openusage/internal/telemetry"
 	"github.com/spf13/cobra"
 )
@@ -52,9 +50,6 @@ func newTelemetryHookCommand() *cobra.Command {
 		Args: cobra.ExactArgs(1),
 		RunE: func(_ *cobra.Command, args []string) error {
 			sourceName := args[0]
-			if _, ok := providers.TelemetrySourceBySystem(sourceName); !ok {
-				return fmt.Errorf("unknown hook source %q", sourceName)
-			}
 
 			payload, err := io.ReadAll(os.Stdin)
 			if err != nil {
@@ -90,7 +85,7 @@ func newTelemetryHookCommand() *cobra.Command {
 				daemonErr = err
 			}
 
-			result, err := ingestHookLocally(
+			result, err := daemon.IngestHookLocally(
 				ctx,
 				sourceName,
 				strings.TrimSpace(accountID),
@@ -148,107 +143,6 @@ func newTelemetryHookCommand() *cobra.Command {
 	return cmd
 }
 
-func ingestHookLocally(
-	ctx context.Context,
-	sourceName string,
-	accountID string,
-	payload []byte,
-	dbPath string,
-	spoolDir string,
-	spoolOnly bool,
-) (daemon.HookResponse, error) {
-	source, ok := providers.TelemetrySourceBySystem(sourceName)
-	if !ok {
-		return daemon.HookResponse{}, fmt.Errorf("unknown hook source %q", sourceName)
-	}
-	reqs, err := telemetry.ParseSourceHookPayload(source, payload, shared.TelemetryCollectOptions{}, accountID)
-	if err != nil {
-		return daemon.HookResponse{}, fmt.Errorf("parse hook payload: %w", err)
-	}
-	resp := daemon.HookResponse{
-		Source:   sourceName,
-		Enqueued: len(reqs),
-	}
-	if len(reqs) == 0 {
-		return resp, nil
-	}
-
-	if strings.TrimSpace(dbPath) == "" {
-		resolved, resolveErr := telemetry.DefaultDBPath()
-		if resolveErr != nil {
-			return daemon.HookResponse{}, fmt.Errorf("resolve telemetry db path: %w", resolveErr)
-		}
-		dbPath = resolved
-	}
-	if strings.TrimSpace(spoolDir) == "" {
-		resolved, resolveErr := telemetry.DefaultSpoolDir()
-		if resolveErr != nil {
-			return daemon.HookResponse{}, fmt.Errorf("resolve telemetry spool dir: %w", resolveErr)
-		}
-		spoolDir = resolved
-	}
-
-	store, err := telemetry.OpenStore(dbPath)
-	if err != nil {
-		return daemon.HookResponse{}, fmt.Errorf("open telemetry store: %w", err)
-	}
-	defer store.Close()
-
-	pipeline := telemetry.NewPipeline(store, telemetry.NewSpool(spoolDir))
-	if spoolOnly {
-		enqueued, enqueueErr := pipeline.EnqueueRequests(reqs)
-		if enqueueErr != nil {
-			return daemon.HookResponse{}, fmt.Errorf("enqueue to telemetry spool: %w", enqueueErr)
-		}
-		resp.Enqueued = enqueued
-		return resp, nil
-	}
-
-	retries := make([]telemetry.IngestRequest, 0, len(reqs))
-	var firstIngestErr error
-	for _, req := range reqs {
-		resp.Processed++
-		result, ingestErr := store.Ingest(ctx, req)
-		if ingestErr != nil {
-			if firstIngestErr == nil {
-				firstIngestErr = ingestErr
-			}
-			retries = append(retries, req)
-			continue
-		}
-		if result.Deduped {
-			resp.Deduped++
-		} else {
-			resp.Ingested++
-		}
-	}
-
-	if len(retries) == 0 {
-		return resp, nil
-	}
-	if firstIngestErr != nil {
-		resp.Warnings = append(resp.Warnings, fmt.Sprintf("direct ingest failed for %d event(s): %v", len(retries), firstIngestErr))
-	}
-
-	enqueued, enqueueErr := pipeline.EnqueueRequests(retries)
-	if enqueueErr != nil {
-		resp.Failed += len(retries)
-		resp.Warnings = append(resp.Warnings, fmt.Sprintf("retry enqueue failed: %v", enqueueErr))
-		return resp, nil
-	}
-	flush, warnings := daemon.FlushInBatches(ctx, pipeline, enqueued)
-	resp.Processed += flush.Processed
-	resp.Ingested += flush.Ingested
-	resp.Deduped += flush.Deduped
-	resp.Failed += flush.Failed
-	resp.Warnings = append(resp.Warnings, warnings...)
-
-	if remaining := len(retries) - flush.Processed; remaining > 0 {
-		resp.Warnings = append(resp.Warnings, fmt.Sprintf("%d event(s) remain queued in spool", remaining))
-	}
-	return resp, nil
-}
-
 func newTelemetryDaemonCommand() *cobra.Command {
 	var (
 		socketPath      string

docs/CODEBASE_AUDIT_ACTION_TABLE_2026-03-09.md

@@ -0,0 +1,29 @@
+# Codebase Audit Action Table
+
+Date: 2026-03-09
+Repository: `/Users/janekbaraniewski/Workspace/priv/openusage`
+Branch: `feat/dashboard-race-parser-cleanups`
+
+## Fixed in This Branch
+
+| ID | Status | Area | Evidence | Resolution | Follow-up |
+| --- | --- | --- | --- | --- | --- |
+| R57 | Fixed | Account config contract hardening | `internal/core/provider.go`, `internal/config/config.go`, `internal/daemon/source_collectors.go`, `internal/detect/cursor.go`, `internal/detect/claude_code.go` | Provider-local runtime paths now live behind `ProviderPaths` and `Path`/`SetPath` helpers. Config load normalizes legacy `paths` payloads into the new field, and daemon/detect flows consume the typed path accessors instead of ad hoc provider-specific overloads. | Retain legacy `paths` read compatibility until the persisted config shape can be fully simplified. |
+| R58 | Fixed | TUI settings/detail decomposition | `internal/tui/settings_modal.go`, `internal/tui/settings_modal_input.go`, `internal/tui/detail.go`, `internal/tui/detail_metrics.go`, `internal/tui/detail_analytics_sections.go` | Settings input/update logic and large detail metric/render sections are split out of the remaining coordinator files. The hot TUI files now separate state/input from section rendering much more cleanly. | Only split further if new features start coupling unrelated flows again. |
+| R59 | Fixed | Detail and analytics metric decoding cleanup | `internal/core/analytics_costs.go`, `internal/core/usage_breakdowns_domains.go`, `internal/tui/detail.go`, `internal/tui/detail_analytics_sections.go`, `internal/tui/model_display_info.go` | Remaining burn-rate, language, MCP, and model-cost detection paths now go through shared core helpers instead of renderer-owned metric-prefix checks. UI code consumes shared semantic helpers rather than decoding raw key conventions inline. | Keep new metric-schema additions in `internal/core`, not in TUI renderers. |
+| R60 | Fixed | Render-path caching follow-through | `internal/tui/render_cache.go`, `internal/tui/analytics_cache.go`, `internal/tui/tiles_cache.go`, `internal/tui/model_input.go`, `internal/tui/model_commands.go`, `internal/tui/dashboard_views.go` | Tile, analytics, and detail render paths are now explicitly invalidated on snapshot, window, theme, layout, and selection changes. Detail rendering is cached the same way analytics and tile composition already were, closing the remaining hot-path rebuild gap. | Profile before adding any more caching layers. |
+| R61 | Fixed | Gemini CLI provider decomposition | `internal/providers/gemini_cli/gemini_cli.go`, `internal/providers/gemini_cli/api_usage.go`, `internal/providers/gemini_cli/session_usage.go` | API/quota/account flows and local session aggregation are split out of the coordinator file. The main provider file is now mostly wiring plus fetch orchestration. | Keep future Gemini changes inside the matching helper unit. |
+| R62 | Fixed | Ollama provider decomposition follow-through | `internal/providers/ollama/ollama.go`, `internal/providers/ollama/local_api.go`, `internal/providers/ollama/cloud_api.go`, `internal/providers/ollama/desktop_db.go`, `internal/providers/ollama/desktop_db_settings.go`, `internal/providers/ollama/desktop_db_tokens.go`, `internal/providers/ollama/desktop_db_breakdowns.go` | Ollama’s coordinator, local API, cloud API, and desktop SQLite flows are now separated by concern. The remaining large desktop DB path is split into settings/schema helpers, token estimation, and usage breakdown/daily series helpers. | Keep future SQLite-specific work inside the dedicated desktop DB helper files. |
+| R63 | Fixed | Telemetry and config fixture cleanup | `internal/telemetry/test_helpers_test.go`, `internal/telemetry/usage_view_test.go`, `internal/config/test_helpers_test.go` | Shared store/file helpers now cover the repeated setup patterns in the telemetry and config suites, and `usage_view_test.go` is reduced below the previous monolith threshold. | Apply the same helper pattern to other large suites when they next change. |
+| R64 | Fixed | Runtime-hint rollout follow-through | `internal/core/provider.go`, `internal/detect/codex.go`, `internal/detect/cursor.go`, `internal/detect/ollama.go`, `internal/providers/codex/live_usage.go`, `internal/providers/copilot/copilot.go`, `internal/providers/ollama/request_helpers.go` | Remaining runtime-only config/account hints now flow through `RuntimeHints` and `Hint`/`SetHint` helpers instead of direct provider code reaching into ad hoc `ExtraData` keys for local paths and overrides. | Keep new runtime-only provider hints behind `Hint`/`SetHint` rather than adding more direct map reads. |
+| R65 | Fixed | Provider/session and test-suite decomposition follow-through | `internal/providers/claude_code/conversation_usage.go`, `internal/providers/claude_code/conversation_usage_projection.go`, `internal/providers/copilot/local_data.go`, `internal/providers/copilot/telemetry_session_file.go`, `internal/providers/copilot/copilot_test.go`, `internal/providers/openrouter/openrouter_analytics_test.go`, `internal/providers/openrouter/openrouter_analytics_rollups_test.go`, `internal/providers/zai/zai.go` | The remaining long provider/session paths are now split by parser/projection/aggregation concern, and the last oversized high-churn test suites are divided by scenario family with shared helpers extracted. | Split again only when a specific family regrows into another mixed-responsibility file. |
+
+## Remaining Review State
+
+No active `P1`, `P2`, or `P3` review items remain from this audit. The earlier follow-up rows were either resolved in this branch or explicitly reclassified as optional future design choices rather than outstanding issues.
+
+## Summary
+
+- The original high-risk review items `A1`, `A2`, `A3`, `A4`, `A12`, `A14`, and `A15` are addressed in this branch.
+- The remaining provider/session decomposition, runtime-hint rollout, and large-suite cleanup work is also addressed in this branch.
+- No additional high-confidence correctness bug was found during the follow-up review after the dashboard timeframe race fix.

docs/SYSTEM_REVIEW_DUPLICATION_AND_RESPONSIBILITY_REPORT.md

@@ -0,0 +1,73 @@
+# System Review: Post-Cleanup State
+
+Date: 2026-03-09
+Repository: `/Users/janekbaraniewski/Workspace/priv/openusage`
+Branch: `feat/dashboard-race-parser-cleanups`
+
+## Scope
+
+This report reflects the tree after the dashboard timeframe-race fix, parser consolidation work, daemon/read-model cleanup, provider decomposition, TUI decomposition, render-cache follow-through, runtime-hint cleanup, large-suite splitting, and the final `A1`/`A2`/`A3`/`A4`/`A12`/`A14`/`A15` cleanup pass.
+
+It replaces the earlier “remaining gaps” snapshot. The goal now is to document the actual post-cleanup state, not to preserve stale open items.
+
+## What Is Resolved
+
+The following earlier review themes are materially closed in this branch:
+
+- Dashboard timeframe race and stale snapshot acceptance.
+- Read-model cache dedupe ignoring time window.
+- Stringly typed daemon/telemetry time-window flow.
+- Parser duplication across Cursor, Codex, and Claude Code dashboard/telemetry paths.
+- OpenRouter, Cursor, Claude Code, Codex, Copilot, OpenCode, Z.AI, Gemini CLI, and Ollama monolith concentration in their previously hottest paths.
+- TUI side-effect leakage into persistence, integration install, and provider validation.
+- Major TUI composition concentration in tile/detail/settings code.
+- Remaining detail/analytics metric-prefix parsing pockets that were still living in renderer code.
+- Tile/detail/analytics render-path recomputation on every frame.
+- Account-config runtime-path overload in the hot path.
+- Repeated telemetry/config/provider test setup boilerplate in the most actively changed suites.
+- Remaining runtime-only provider overrides reaching directly into ad hoc `ExtraData` fields.
+- The last oversized high-churn Copilot/OpenRouter test suites.
+
+## Current Findings
+
+### 1. No remaining high-confidence correctness bug surfaced in the follow-up review
+
+After the final cleanup pass and validation run, I did not find another issue on the level of the original dashboard timeframe race. The remaining items are not hidden state-corruption or concurrency defects; they are explicit maintenance tradeoffs.
+
+Validation used for this state:
+
+- `go test ./...`
+- `go vet ./...`
+- `make build`
+
+### 2. The codebase now has clearer responsibility boundaries in the hot areas
+
+The most change-prone areas are no longer concentrated the way they were at the start of the branch:
+
+- TUI render/state work is split across dedicated settings/detail/cache/helper units.
+- Provider-local parsing and fetch logic are split by concern in the previously worst provider files.
+- Daemon hook ingest, HTTP, polling, spool, and read-model paths are separated.
+- Telemetry usage-view query/materialization/projection/aggregate logic is separated.
+
+This reduces review blast radius and makes future concurrency/data-flow work easier to reason about.
+
+### 3. No active audit-priority items remain
+The earlier follow-up list is now closed for the purposes of this review. What remains in the repo are ordinary future refactor options, not unresolved `P1`/`P2`/`P3` findings from this audit.
+
+## References
+
+- [CODEBASE_AUDIT_ACTION_TABLE_2026-03-09.md](/Users/janekbaraniewski/Workspace/priv/openusage/docs/CODEBASE_AUDIT_ACTION_TABLE_2026-03-09.md)
+- [internal/tui/render_cache.go](/Users/janekbaraniewski/Workspace/priv/openusage/internal/tui/render_cache.go)
+- [internal/tui/detail_metrics.go](/Users/janekbaraniewski/Workspace/priv/openusage/internal/tui/detail_metrics.go)
+- [internal/tui/settings_modal_input.go](/Users/janekbaraniewski/Workspace/priv/openusage/internal/tui/settings_modal_input.go)
+- [internal/providers/ollama/desktop_db.go](/Users/janekbaraniewski/Workspace/priv/openusage/internal/providers/ollama/desktop_db.go)
+- [internal/providers/ollama/desktop_db_tokens.go](/Users/janekbaraniewski/Workspace/priv/openusage/internal/providers/ollama/desktop_db_tokens.go)
+- [internal/providers/gemini_cli/api_usage.go](/Users/janekbaraniewski/Workspace/priv/openusage/internal/providers/gemini_cli/api_usage.go)
+- [internal/core/provider.go](/Users/janekbaraniewski/Workspace/priv/openusage/internal/core/provider.go)
+- [internal/telemetry/test_helpers_test.go](/Users/janekbaraniewski/Workspace/priv/openusage/internal/telemetry/test_helpers_test.go)
+
+## Bottom Line
+
+- The original review’s high-priority structural set is addressed.
+- The repo is in materially better shape than at the start of the branch.
+- Remaining items are optional follow-up architecture choices, not outstanding bugs from the review.