CODE_REVIEW_ISSUES.json

{
  "meta": {
    "project": "LazyTail",
    "review_date": "2026-03-12",
    "reviewers": ["Uncle Bob (Robert C. Martin)", "Kent Beck"],
    "overall_grade": "B-",
    "total_issues": 44
  },
  "issues": [
    {
      "id": "C1",
      "severity": "critical",
      "category": "god-class",
      "title": "App struct is a God Class",
      "file": "src/app/mod.rs",
      "lines": "155-2957",
      "description": "App struct has 25+ fields and 2957 lines mixing 6 responsibilities: tab management, input handling, filter control, source panel navigation, UI rendering state, and event dispatching. apply_event() alone is 132 lines with 40+ match arms.",
      "task": "Extract InputController (input_buffer, input_cursor, input_mode), FilterController (current_filter_mode, regex_error, query_error, pending_filter_at), SourcePanelController (source_panel, side_panel_width), and TabManager (tabs, active_tab, combined_tabs). Target: App < 500 lines, each controller < 300 lines.",
      "status": "done"
    },
    {
      "id": "C2",
      "severity": "critical",
      "category": "god-class",
      "title": "query.rs is a monolithic 1864-line file",
      "file": "src/filter/query.rs",
      "lines": "1-1864",
      "description": "Contains parser, AST types, filter implementation, and validation all in one file. This is 26% of the entire filter module.",
      "task": "Split into query/ directory: query/mod.rs (re-exports), query/ast.rs (FilterQuery, Aggregation types), query/parser.rs (text query parsing), query/validate.rs (validation logic). Each file should be 400-500 lines.",
      "status": "done"
    },
    {
      "id": "C3",
      "severity": "critical",
      "category": "god-class",
      "title": "MCP tools.rs is a God Object",
      "file": "src/mcp/tools.rs",
      "lines": "1-2233",
      "description": "All 6 MCP tool implementations plus helpers, response stripping, and tests in a single 2233-line file. Three nearly identical ANSI-stripping functions violate DRY.",
      "task": "Split into per-tool files: search_impl.rs, lines_impl.rs, stats_impl.rs, render_impl.rs. Extract trait AnsiStrippable for response stripping to eliminate the three duplicate strip_*_response() functions.",
      "status": "done"
    },
    {
      "id": "C4",
      "severity": "critical",
      "category": "bug",
      "title": "Theme loading has infinite recursion vulnerability",
      "file": "src/theme/loader.rs",
      "lines": "66",
      "description": "The resolve_named() function accepts a visited: &mut Vec<String> parameter for cycle detection but never checks it before recursing. If a theme references itself via 'base: my-theme-name', infinite recursion and stack overflow occur.",
      "task": "Add visited.contains(name) check at the top of resolve_named() before the recursive call to load_theme_file(). Return a ConfigError::CycleDetected if the theme name is already in the visited list.",
      "status": "open"
    },
    {
      "id": "H1",
      "severity": "high",
      "category": "long-method",
      "title": "main.rs has multiple oversized functions",
      "file": "src/main.rs",
      "lines": "118-1191",
      "description": "main() is ~258 lines handling 5 concerns. collect_file_events() is ~135 lines. collect_input_events() is ~110 lines. run_app_with_discovery() is ~113 lines. Raw Duration and Instant values scattered as timing state (primitive obsession).",
      "task": "Extract main() into setup_cli(), detect_mode(), run_mode(). Create EventLoop or TimeTracker struct to hold timing state (last_status_refresh, last_file_poll). Extract collect_file_events() into smaller focused functions. Target: main() < 50 lines, no function > 80 lines.",
      "status": "open"
    },
    {
      "id": "H2",
      "severity": "high",
      "category": "long-method",
      "title": "render_sources_list() is a 246-line God Function with DRY violations",
      "file": "src/tui/side_panel.rs",
      "lines": "109-354",
      "description": "Handles category headers, combined items, regular items, AND overflow overlay. Contains duplicated line-building logic between combined items (lines 181-209) and regular items (lines 261-288) — identical truncation and metadata appending patterns.",
      "task": "Extract append_truncated_metadata() helper to eliminate DRY violation. Split into render_category_header(), render_combined_item(), render_regular_item(), render_overflow_overlay(). Target: each function < 60 lines.",
      "status": "open"
    },
    {
      "id": "H3",
      "severity": "high",
      "category": "fragile-code",
      "title": "Span index management uses magic numbers",
      "file": "src/tui/log_view.rs",
      "lines": "414-452",
      "description": "apply_expanded_bg() uses hardcoded indices (0, 1, 2, 3) to locate spans built by build_item(). Values like sep_idx = if has_source_tag { 2 } else { 1 } are brittle — if span insertion order changes in build_item(), this breaks silently with no compiler warning.",
      "task": "Introduce a SpanRole enum (SourceTag, LineNumber, Separator, Content) or a LineSpans struct that tracks span positions semantically. Replace index arithmetic with role-based lookups in apply_expanded_bg().",
      "status": "open"
    },
    {
      "id": "H4",
      "severity": "high",
      "category": "long-method",
      "title": "preset.rs compile() is 175 lines mixing concerns",
      "file": "src/renderer/preset.rs",
      "lines": "150-325",
      "description": "Single function mixes parser resolution, layout entry compilation, and field validation. Multiple nested match statements. High cognitive load and hard to test individual compilation phases.",
      "task": "Split compile() into compile_parser(), compile_layout(), and validate_fields(). Each should be independently testable. Target: compile() becomes a 20-line orchestrator calling sub-functions.",
      "status": "open"
    },
    {
      "id": "H5",
      "severity": "high",
      "category": "long-method",
      "title": "Web handle_request() is a 348-line monolith",
      "file": "src/web/handlers.rs",
      "lines": "24-372",
      "description": "All HTTP routing in a single match statement with 10+ arms, each with 10-50 lines of nested logic. Changes to one endpoint risk breaking others. Weak error categorization for multi-step operations like source deletion.",
      "task": "Extract per-endpoint handler functions: handle_get_api_lines(), handle_post_api_filter(), handle_get_api_events(), etc. The main handle_request() should be a thin router dispatching to these handlers. Target: router < 40 lines, each handler < 60 lines.",
      "status": "open"
    },
    {
      "id": "H6",
      "severity": "high",
      "category": "long-method",
      "title": "file_reader::read_line_at() is 102 lines with 4 code paths",
      "file": "src/reader/file_reader.rs",
      "lines": "268-369",
      "description": "Contains 4 separate access strategies (sequential fast path, columnar offsets, columnar tail, sparse index fallback) in a single method. Violates Uncle Bob's rule that methods should be small enough to understand at a glance.",
      "task": "Extract into try_sequential_path(), try_columnar_path(), try_columnar_tail_path(), try_sparse_path(). Main method becomes ~20 lines calling each strategy in priority order.",
      "status": "open"
    },
    {
      "id": "M1",
      "severity": "medium",
      "category": "dry-violation",
      "title": "Repeated mmap loading pattern in streaming_filter.rs",
      "file": "src/filter/streaming_filter.rs",
      "lines": "66-661",
      "description": "All 5 *_impl() functions repeat identical mmap setup: File::open, metadata check for empty, unsafe Mmap::map. ~15 lines duplicated 5 times.",
      "task": "Extract fn load_mmap(path: &Path) -> Result<(Mmap, u64)> that handles open, empty check, and mapping. Use in all 5 impl functions. Saves ~60 lines of duplication.",
      "status": "open"
    },
    {
      "id": "M2",
      "severity": "medium",
      "category": "long-parameter-list",
      "title": "stream_filter_range_impl() takes 8 parameters",
      "file": "src/filter/streaming_filter.rs",
      "lines": "432",
      "description": "Parameters: path, filter, start_line, end_line, start_byte_offset, bitmap, tx, cancel. Uncle Bob recommends max 3 parameters.",
      "task": "Introduce RangeFilterOptions struct grouping start_line, end_line, start_byte_offset, bitmap. Reduces to 4 params: path, filter, options, cancel (tx moves into options or return type).",
      "status": "open"
    },
    {
      "id": "M3",
      "severity": "medium",
      "category": "long-parameter-list",
      "title": "process_filter_shared() takes 7 parameters",
      "file": "src/filter/engine.rs",
      "lines": "122-129",
      "description": "Parameters: reader, filter, tx, batch_size, start_line, end_line, cancel. Also a 106-line method mixing lock acquisition, batch filtering, progress reporting, and synchronization.",
      "task": "Extract ProcessFilterOptions { batch_size, start_line, end_line }. Split method into read_batch(), filter_batch(), send_progress(). Target: each sub-method < 30 lines.",
      "status": "open"
    },
    {
      "id": "M4",
      "severity": "medium",
      "category": "tell-dont-ask",
      "title": "TUI rendering reaches deep into App/Tab/LogSource chains",
      "file": "src/tui/mod.rs",
      "lines": "46-54",
      "description": "Rendering functions check app.active_tab().source.mode == ViewMode::Aggregation and duplicate tab resolution logic (active_combined vs active_tab). Violates Tell-Don't-Ask principle.",
      "task": "Add App accessor methods: current_render_tab() -> &TabState, is_aggregation_active() -> bool. Use these in tui/mod.rs, tui/log_view.rs, and handlers/input.rs instead of reaching into internals.",
      "status": "open"
    },
    {
      "id": "M5",
      "severity": "medium",
      "category": "inconsistent-error-handling",
      "title": "Inconsistent error handling patterns across modules",
      "file": "multiple",
      "lines": "various",
      "description": "Five different error handling patterns used inconsistently: silent let _ = send(), silent _ => continue, eprintln + break (capture.rs), poisoned mutex recovery (web/handlers.rs), eprintln + return (watcher/dir.rs). No documented error philosophy.",
      "task": "Establish error handling guidelines: (1) propagate via Result at system boundaries, (2) log-and-continue only in background processing with explicit // deliberate: ... comment, (3) never silently recover from poisoned mutex without logging. Add one-line comments to all existing silent-skip patterns.",
      "status": "open"
    },
    {
      "id": "M6",
      "severity": "medium",
      "category": "tech-debt",
      "title": "TabState has redundant scroll_position/selected_line fields",
      "file": "src/app/tab.rs",
      "lines": "87-88",
      "description": "scroll_position and selected_line on TabState are manually synced copies of Viewport state via sync_from_viewport(). Comment acknowledges this is 'for compatibility'. Creates risk of desync bugs.",
      "task": "Remove scroll_position and selected_line from TabState. Update all callers to use tab.viewport.scroll_position() and tab.viewport.selected_line() directly. Remove sync_from_viewport() method.",
      "status": "open"
    },
    {
      "id": "M7",
      "severity": "medium",
      "category": "misplaced-code",
      "title": "451 lines of integration tests in filter/mod.rs",
      "file": "src/filter/mod.rs",
      "lines": "381-823",
      "description": "Module root contains 451 lines of integration tests (index_filter_integration_tests) embedded via #[cfg(test)]. This bloats the production module from ~100 to 823 lines.",
      "task": "Move integration tests to tests/filter_integration.rs. Keep only unit tests for FilterMode and FilterHistoryEntry in the module. Target: mod.rs < 200 lines.",
      "status": "open"
    },
    {
      "id": "M8",
      "severity": "medium",
      "category": "extensibility",
      "title": "SeverityCounts uses individual fields instead of array",
      "file": "src/index/builder.rs",
      "lines": "27-36",
      "description": "SeverityCounts has individual trace/debug/info/warn/error/fatal fields. add_severity() uses a 6-arm match. Adding a new severity level requires changes in 3+ places.",
      "task": "Consider using an array indexed by severity enum variant: counts: [u64; Severity::COUNT]. add_severity becomes counts[severity as usize] += 1. Alternatively, keep current design if no new severity levels are planned.",
      "status": "open"
    },
    {
      "id": "M9",
      "severity": "medium",
      "category": "primitive-obsession",
      "title": "Aggregation uses raw tuple instead of named struct",
      "file": "src/filter/aggregation.rs",
      "lines": "45-51",
      "description": "HashMap<Vec<String>, (usize, Vec<usize>)> uses an unnamed tuple where group.0 is count and group.1 is line_indices. Obscures meaning.",
      "task": "Replace tuple with struct GroupAccumulator { count: usize, line_indices: Vec<usize> }. Makes code self-documenting: entry.count += 1 instead of entry.0 += 1.",
      "status": "open"
    },
    {
      "id": "M10",
      "severity": "medium",
      "category": "long-method",
      "title": "render_log_view() is 171 lines mixing 4 concerns",
      "file": "src/tui/log_view.rs",
      "lines": "70-240",
      "description": "Mixes setup/context building (~40 lines), viewport calculation (~25 lines), main item-building loop (~60 lines), and widget assembly (~20 lines). Closure captures 6+ variables implicitly.",
      "task": "Extract build_render_context(), calculate_line_heights() (replace closure with LineHeightCalculator struct), build_list_items(). Target: render_log_view() becomes a 30-line orchestrator.",
      "status": "open"
    },
    {
      "id": "M11",
      "severity": "medium",
      "category": "dry-violation",
      "title": "Constructor duplication in tab.rs",
      "file": "src/app/tab.rs",
      "lines": "150-344",
      "description": "Four constructors (new, from_stdin, from_discovered_source, from_config_source) repeat: index reader opening, file size reading, index size calculation, watcher setup, reader Arc<Mutex<>> wrapping.",
      "task": "Extract fn open_file_source(path: &Path, watch: bool) -> Result<FileSourceSetup> that returns a struct with reader, index_reader, file_size, index_size, watcher. All constructors use this helper.",
      "status": "open"
    },
    {
      "id": "M12",
      "severity": "medium",
      "category": "coupling",
      "title": "Closure over-capture in log_view line height calculation",
      "file": "src/tui/log_view.rs",
      "lines": "123-141",
      "description": "Line height closure captures 6+ variables from outer scope (line_indices, expanded_lines, content_width, line_wrap, raw_mode, reader_guard). Dependencies are implicit, making the code hard to reason about.",
      "task": "Replace closure with an explicit LineHeightCalculator struct that takes all dependencies in its constructor. Pass calculator to resolve_with_heights() instead of the closure.",
      "status": "open"
    },
    {
      "id": "L1",
      "severity": "low",
      "category": "dry-violation",
      "title": "String truncation logic duplicated across TUI modules",
      "file": "src/tui/",
      "lines": "various",
      "description": "Truncation-with-ellipsis logic appears in log_view.rs (format_source_tag), help.rs (name truncation in confirm dialog), and side_panel.rs (item rendering). Each reimplements width calculation and truncation.",
      "task": "Extract a shared utility function truncate_with_ellipsis(s: &str, max_width: usize) -> String in a tui utilities module. Use across all three files.",
      "status": "open"
    },
    {
      "id": "L2",
      "severity": "low",
      "category": "coupling",
      "title": "combined_reader::source_info() coupled to ratatui Color",
      "file": "src/reader/combined_reader.rs",
      "lines": "167-170",
      "description": "source_info() takes &[ratatui::style::Color] and returns Option<(&str, Color)>. Domain reader shouldn't depend on TUI framework types.",
      "task": "Return Option<(&str, usize)> with source_id index. Let the TUI caller look up the color from the source_id. Removes ratatui dependency from reader module.",
      "status": "open"
    },
    {
      "id": "L3",
      "severity": "low",
      "category": "data-integrity",
      "title": "Session context eviction is arbitrary, not LRU",
      "file": "src/session.rs",
      "lines": "113-121",
      "description": "When session contexts exceed MAX_CONTEXTS (100), entries are removed using HashMap iteration order which is arbitrary. No timestamp-based or LRU eviction.",
      "task": "Add a last_used timestamp to session entries. When evicting, sort by last_used and remove oldest. Alternatively, use an IndexMap to preserve insertion order.",
      "status": "open"
    },
    {
      "id": "L4",
      "severity": "low",
      "category": "safety",
      "title": "Silent u32 overflow in sparse_index.rs",
      "file": "src/reader/sparse_index.rs",
      "lines": "84",
      "description": "line_number as u32 silently truncates if line_number >= 2^32 (4 billion lines). Comment says 'supports up to 4 billion lines' but no runtime check.",
      "task": "Add a debug_assert!(line_number <= u32::MAX as usize) or return an error if the line number exceeds u32 range. Document the 4-billion-line limit at the type level.",
      "status": "open"
    },
    {
      "id": "L5",
      "severity": "low",
      "category": "fragile-code",
      "title": "Config error message parsing uses magic string offset",
      "file": "src/config/error.rs",
      "lines": "139-166",
      "description": "Parses serde-saphyr error messages with rfind(' at line ') and magic offset + 9. If the YAML library changes its error format, this silently returns (None, None).",
      "task": "Add a comment explaining the dependency on serde-saphyr's error format. Consider adding a test that validates the parsing against a known error message to catch format changes.",
      "status": "open"
    },
    {
      "id": "L6",
      "severity": "low",
      "category": "error-handling",
      "title": "Capture mode swallows indexing errors",
      "file": "src/capture.rs",
      "lines": "145-147",
      "description": "Index errors during capture are logged via eprintln('Warning: failed to index line') and silently continued. Could cascade into corrupted index that causes later analysis failures.",
      "task": "Add a counter for index errors. If errors exceed a threshold (e.g., 10), log a prominent warning that the index may be incomplete. Document the design decision to continue on index failure.",
      "status": "open"
    },
    {
      "id": "L7",
      "severity": "low",
      "category": "srp-violation",
      "title": "build_missing_indexes() mixes progress reporting with logic",
      "file": "src/source.rs",
      "lines": "506-558",
      "description": "Function mixes filtering (which sources need indexes), progress reporting (eprintln calls), and index building orchestration. Progress format '[i/count]' duplicated 4 times.",
      "task": "Extract progress reporting into a callback parameter or a ProgressReporter trait. Extract the common '[i/count]' format into a helper. Keep build_missing_indexes() focused on orchestration.",
      "status": "open"
    },
    {
      "id": "L8",
      "severity": "low",
      "category": "data-clump",
      "title": "Drill-down state fields should be grouped",
      "file": "src/log_source.rs",
      "lines": "113-117",
      "description": "drill_down_aggregation and drill_down_pattern on FilterConfig always travel together but aren't in their own type. Name 'drill_down_aggregation' is also unclear.",
      "task": "Group into struct DrillDownState { parent_result: AggregationResult, parent_pattern: String }. Replace the two Option fields with a single Option<DrillDownState>.",
      "status": "open"
    },
    {
      "id": "L9",
      "severity": "low",
      "category": "naming",
      "title": "TabState::stream_writer is misleadingly named",
      "file": "src/app/tab.rs",
      "lines": "100",
      "description": "stream_writer is actually the handle to the reader that appends lines, not a writer. The name suggests it writes to a stream, but it writes to the in-memory buffer.",
      "task": "Rename to stream_appender or streamable_reader_handle to clarify its purpose.",
      "status": "open"
    },
    {
      "id": "L10",
      "severity": "low",
      "category": "documentation",
      "title": "Index builder batch size is undocumented magic number",
      "file": "src/index/builder.rs",
      "lines": "18",
      "description": "const BATCH: usize = 1024 has no comment explaining why 1024 was chosen. Is it empirically tuned? A cache-line optimization? Unclear.",
      "task": "Add a comment explaining the rationale for the 1024 batch size (e.g., 'Empirically tuned for balance between flush frequency and I/O overhead').",
      "status": "open"
    },
    {
      "id": "L11",
      "severity": "low",
      "category": "feature-envy",
      "title": "TabState directly manipulates marker file directory structure",
      "file": "src/app/tab.rs",
      "lines": "404-414",
      "description": "TabState::refresh_source_status() knows about .lazytail/data/ and .lazytail/sources/ directory structure, navigating parent directories to find marker files. This is source.rs's responsibility.",
      "task": "Move this logic to LogSource::refresh_status() or call a source.rs function that encapsulates the directory traversal.",
      "status": "open"
    },
    {
      "id": "C5",
      "severity": "critical",
      "category": "architecture-violation",
      "title": "Event-driven architecture bypassed for inactive tabs, combined tabs, and streams",
      "file": "src/main.rs",
      "lines": "751-1028",
      "description": "ADR-001 and CLAUDE.md state 'all state transitions go through App::apply_event()' but main.rs bypasses this in 3 places: (1) collect_file_events() calls tab.apply_file_modification() directly on inactive tabs (line 823) and mutates combined tab fields directly (lines 829-872), (2) collect_filter_progress() calls tab.apply_filter_event() directly on inactive tabs (lines 914-920), (3) collect_stream_events() admits in its docstring it 'modifies tabs directly rather than returning events' (lines 978-1028). AppEvent::StreamData and AppEvent::StreamComplete exist but are #[allow(dead_code)].",
      "task": "Document these as intentional fast-path exceptions in CLAUDE.md with performance justification. Optionally wire up AppEvent::StreamData/StreamComplete to unify the stream path. Add explicit '// FAST-PATH: bypasses event system for performance' comments at each bypass site.",
      "status": "open"
    },
    {
      "id": "H7",
      "severity": "high",
      "category": "documentation-mismatch",
      "title": "process_event() is a dead pass-through despite documentation claims",
      "file": "src/main.rs",
      "lines": "1185-1187",
      "description": "CLAUDE.md states 'Side effects are in process_event() in main.rs' but process_event() is a single-line function that calls app.apply_event(event) with zero side-effect logic. Side effects actually happen in collect_*() functions and within apply_event(). The documentation describes an idealized architecture that was never implemented or was refactored away.",
      "task": "Update CLAUDE.md and ARCHITECTURE.md to accurately describe where side effects occur. Either remove the process_event() wrapper or document that it exists as a future extension point.",
      "status": "open"
    },
    {
      "id": "H8",
      "severity": "high",
      "category": "dependency-inversion",
      "title": "Domain core log_source.rs imports from adapter module app/",
      "file": "src/log_source.rs",
      "lines": "1",
      "description": "LogSource is documented as 'domain core shared across TUI/Web/MCP' (ADR-014) but imports FilterState and ViewMode from crate::app. This inverts the hexagonal architecture dependency direction — domain code depends on adapter code. FilterConfig.state: FilterState and LogSource.mode: ViewMode embed adapter-defined types in domain structs. This forces filter_orchestrator.rs and web/mod.rs to also import from app/ to work with these types.",
      "task": "Move FilterState and ViewMode definitions from src/app/mod.rs to src/log_source.rs (or src/filter/mod.rs for FilterState). Have app/mod.rs re-export them for backward compatibility. This restores correct dependency direction: adapters import from domain, not vice versa.",
      "status": "open"
    },
    {
      "id": "H9",
      "severity": "high",
      "category": "solid-violation",
      "title": "LogReader::as_any() violates LSP and ISP",
      "file": "src/reader/mod.rs",
      "lines": "21",
      "description": "CLAUDE.md documents LogReader as having 3 methods but it has 4 — the undocumented as_any() -> &dyn Any enables runtime downcasting. tui/log_view.rs (lines 255,269,305) downcasts LogReader to CombinedReader to access source_info(), severity(), and renderer_names(). This breaks Liskov Substitution — callers must know the concrete type. It also violates Interface Segregation by forcing all LogReader implementations to carry reflection support they don't need.",
      "task": "Option A: Extract as_any() into a separate ReaderReflection trait. Option B: Add source_info() with a default no-op to LogReader trait. Option C: Pass combined-reader metadata through a separate data channel to the rendering layer rather than downcasting. Update CLAUDE.md to document actual method count.",
      "status": "open"
    },
    {
      "id": "H10",
      "severity": "high",
      "category": "data-loss",
      "title": "Web adapter missing 'unknown' severity count field",
      "file": "src/web/mod.rs",
      "lines": "74-82",
      "description": "Web's SeverityCountsView struct has 6 fields (trace, debug, info, warn, error, fatal) but is missing the 'unknown' field. MCP's SeverityCountsInfo (src/mcp/types.rs:329-337) correctly includes all 7 fields. The source of truth SeverityCounts (src/index/checkpoint.rs:12-22) has 7 fields including unknown. Web clients silently lose unknown severity count data.",
      "task": "Add 'unknown: u64' field to SeverityCountsView in src/web/mod.rs and populate it from the checkpoint data in src/web/state.rs.",
      "status": "open"
    },
    {
      "id": "M13",
      "severity": "medium",
      "category": "adapter-parity",
      "title": "Web adapter missing renderer preset support",
      "file": "src/web/handlers.rs",
      "lines": "98-119",
      "description": "MCP adapter (src/mcp/tools.rs:296-355) includes full renderer preset support via PresetRegistry and render_line_info(). Web adapter returns raw line content only — LineRow struct has no 'rendered' field. Same domain data surfaces different capabilities depending on which adapter is used.",
      "task": "Either add optional preset rendering to Web's /api/lines endpoint (add rendered field to LineRow, pass PresetRegistry to web handler) or document this as an intentional design decision with rationale.",
      "status": "open"
    },
    {
      "id": "M14",
      "severity": "medium",
      "category": "dry-violation",
      "title": "Web adapter duplicates query validation before FilterOrchestrator",
      "file": "src/web/handlers.rs",
      "lines": "182-193",
      "description": "Web handler pre-validates regex and query patterns (calling RegexFilter::new() and query::parse_query()) before passing to FilterOrchestrator::trigger(), which performs the same validation internally during filter construction. MCP does not pre-validate. This violates DRY and means validation error messages may diverge.",
      "task": "Remove pre-validation from web handler. Let FilterOrchestrator::trigger() handle all validation and return errors that the web handler can translate to HTTP error responses.",
      "status": "open"
    },
    {
      "id": "M15",
      "severity": "medium",
      "category": "documentation-mismatch",
      "title": "Follow-mode jumps don't work for inactive tabs despite ADR-009 claim",
      "file": "src/app/mod.rs",
      "lines": "1493-1497",
      "description": "ADR-009 states 'Follow mode jumps happen for inactive tabs if enabled' but jump_to_end() only fires for the active tab during filter completion in handle_filter_progress_event(). Inactive tabs with follow_mode=true do not auto-scroll when their filter completes or when new lines arrive.",
      "task": "Either implement follow-mode jumps for inactive tabs (add jump_to_end() call in the inactive tab branch of collect_filter_progress()) or update ADR-009 to remove the incorrect claim.",
      "status": "open"
    },
    {
      "id": "M16",
      "severity": "medium",
      "category": "side-effect",
      "title": "Render function mutates TabState during rendering",
      "file": "src/tui/log_view.rs",
      "lines": "149-150",
      "description": "render_log_view() takes &mut App and mutates tab.scroll_position and tab.selected_line during rendering. Render functions should be pure reads of application state. This makes the render pass side-effectful, complicates reasoning about state transitions, and could cause issues if rendering is called multiple times or cached.",
      "task": "Have render_log_view() return the new scroll_position and selected_line as a struct. Apply the updates in the main event loop after rendering completes. Alternatively, resolve this with M6 by removing the redundant fields entirely.",
      "status": "open"
    },
    {
      "id": "L12",
      "severity": "low",
      "category": "misplaced-code",
      "title": "FilterOrchestrator lives outside the filter/ module",
      "file": "src/filter_orchestrator.rs",
      "lines": "1-150",
      "description": "FilterOrchestrator is at src/filter_orchestrator.rs (crate root) but logically belongs in src/filter/orchestrator.rs alongside other filter infrastructure. ARCHITECTURE.md lists it under the filter/ module map but it actually lives at the crate root.",
      "task": "Move src/filter_orchestrator.rs to src/filter/orchestrator.rs. Add pub mod orchestrator to src/filter/mod.rs. Update imports across the codebase.",
      "status": "open"
    },
    {
      "id": "L13",
      "severity": "low",
      "category": "dead-code",
      "title": "Dead AppEvent::StreamData and AppEvent::StreamComplete variants",
      "file": "src/app/event.rs",
      "lines": "57-62",
      "description": "AppEvent::StreamData and AppEvent::StreamComplete exist in the event enum but are marked #[allow(dead_code)] and never constructed. Stream events bypass the event system entirely (see C5). These represent an abandoned attempt to unify stream handling with the event architecture.",
      "task": "Either wire these up by having collect_stream_events() return them as events (fixing C5), or remove them to reduce confusion. If removed, add a comment explaining why streams bypass the event system.",
      "status": "open"
    },
    {
      "id": "L14",
      "severity": "low",
      "category": "documentation-mismatch",
      "title": "ADR-005 scrolloff documentation says 3 but code has uncommitted change to 0",
      "file": "src/app/viewport.rs",
      "lines": "13",
      "description": "ADR-005 explicitly states 'Edge padding (scrolloff=3)' as a design decision. The working directory shows DEFAULT_EDGE_PADDING = 0, which is an uncommitted modification from the documented value of 3. If the change is intentional, the ADR is stale. If not, the change should be reverted.",
      "task": "Either commit the change and update ADR-005 to document scrolloff=0 with the rationale for removing edge padding, or revert the uncommitted change to restore scrolloff=3.",
      "status": "open"
    }
  ]
}