epic: codebase survey findings (prioritized backlog)

[Bccorb]

Tracking issue for the codebase survey (README review + full source survey). Ordered by priority; lower issue number roughly tracks higher priority. Relationships note where one issue fixes or unblocks another (GitHub shows the reverse links on each referenced issue).

P0 - launch-critical correctness

• bug: global (multi-region) EC2/RDS/Lambda fetches ignore the selected AWS profile #11 Global EC2/RDS/Lambda fetches ignore the selected profile
• bug: cross-region describe rebuilds SDK config without the active profile #12 Cross-region describe drops the active profile
• bug: findings are recomputed from stale cross-service data and mislabeled with the current region #13 Findings recomputed from stale cross-service data, mislabeled region
• bug: app panics on STS/ECS failures (unwrap on get_caller_identity, list/describe clusters) #14 Panics on STS/ECS unwrap
• bug: cost cache is a single global file, not keyed by profile or region (cross-account leak) #15 Cost cache not keyed by profile/region (cross-account leak)

P1 - error honesty, robustness, key bugs

• feat: add an in-UI error/notification surface (eprintln is invisible in the alternate screen) #16 Add an in-UI error surface (enables the honesty fixes below)
• bug: list fetchers turn any error into an empty, success-looking table #17 List fetchers swallow errors into empty tables (needs feat: add an in-UI error/notification surface (eprintln is invisible in the alternate screen) #16)
• bug: Cost Explorer errors render as $0 spend and a flat chart #18 Cost Explorer errors render as $0 (needs feat: add an in-UI error/notification surface (eprintln is invisible in the alternate screen) #16)
• bug: region discovery silently collapses to us-east-1 on failure #19 Region discovery collapses to us-east-1 (needs feat: add an in-UI error/notification surface (eprintln is invisible in the alternate screen) #16)
• bug: access-denied is detected by string-matching and misclassifies many auth errors #20 Access-denied string-matching misclassifies auth errors
• bug: global region selection is lost on restart #21 Global selection lost on restart
• bug: no fetcher paginates, so large accounts get silently truncated inventories #22 No pagination -> truncated inventories
• bug: numeric view-switch keys fire while an overlay is open, and key guards are inconsistent #23 Numeric keys fire under overlays; inconsistent guards

P2 - engineering health, scale, triage model

• perf: AWS refresh blocks the draw loop and the phased progress UI is dead #24 Refresh blocks the draw loop; dead phase UI
• perf: findings/cost-savings refresh fetches ~11 services sequentially #25 Sequential fetches should be concurrent (needs perf: AWS refresh blocks the draw loop and the phased progress UI is dead #24)
• feat: cache service inventories with a TTL, keyed by profile+region (fast navigation) #26 Inventory cache TTL keyed by profile+region -> closes bug: cost cache is a single global file, not keyed by profile or region (cross-account leak) #15
• perf: N+1 serial describe calls for target-group health and SQS attributes #27 N+1 serial describes
• reliability: configure SDK retry/backoff and timeouts for large accounts #28 No retry/backoff (compounds perf: N+1 serial describe calls for target-group health and SQS attributes #27)
• tech-debt: replace per-service match arms with a service trait/registry #29 Service trait/registry -> fixes the match drift + SQS gap
• tech-debt: extract a shared list-view table/scroll/selection helper #30 Shared list-table helper -> unblocks tech-debt: give each view its own selection/scroll state (or clamp centrally) #31, feat: in-view resource search/filter and column sorting #39, ux: manage long-value truncation and standardize empty states #40, tech-debt: small cleanups (duplicated table widths, misspelled model file, magic item count) #44
• tech-debt: give each view its own selection/scroll state (or clamp centrally) #31 Per-view selection state (via tech-debt: extract a shared list-view table/scroll/selection helper #30)
• tech-debt: consolidate the three clients_for_region copies and the two aggregation reducers #32 Consolidate clients_for_region + aggregation (home for bug: global (multi-region) EC2/RDS/Lambda fetches ignore the selected AWS profile #11/bug: cross-region describe rebuilds SDK config without the active profile #12)
• tech-debt: drive numeric shortcuts and help from the command registry (single source) #33 Shortcuts+help from one registry (folds in bug: numeric view-switch keys fire while an overlay is open, and key guards are inconsistent #23)
• tech-debt: refactor rebuild_findings into a rule registry #34 rebuild_findings rule registry (enables test: add coverage for finding rules, model predicates, cost parsing, and region/profile logic #38)
• feat: carry resource tags/ownership on all service models #35 Tags on all models -> unblocks feat: per-client tag focus filter across service views #8 and feat: add high-signal services and findings (S3, IAM, EBS/EIP/snapshots, RDS security, ECS incidents) #42
• feat: expand the Finding model for a real triage queue (id, pivot, cost, severity, dedup/snooze) #36 Expand Finding model (id/pivot/cost/severity/dedup/snooze); relates bug: findings are recomputed from stale cross-service data and mislabeled with the current region #13 - partial: stable id and severity scale done, cost done for the 2 rules a list price can answer; executable pivot and acknowledge/snooze remain
• reliability: version and validate the persisted config; stop silently wiping preferences #37 Config schema version + validation
• test: add coverage for finding rules, model predicates, cost parsing, and region/profile logic #38 Test coverage (after tech-debt: refactor rebuild_findings into a rule registry #34, feat: expand the Finding model for a real triage queue (id, pivot, cost, severity, dedup/snooze) #36) - premise restated: the suite grew from 3 to 140 tests, but model predicates and cost parsing are still untested

P3 - UX and product backlog, cleanups

• feat: in-view resource search/filter and column sorting #39 In-view search/filter/sort (via tech-debt: extract a shared list-view table/scroll/selection helper #30) - partial: row filter done, column sorting remains
• ux: manage long-value truncation and standardize empty states #40 Truncation + empty-state UX (via tech-debt: extract a shared list-view table/scroll/selection helper #30)
• bug: footer key hints drift from real handlers; SelectSshKey overlay scroll arms are no-ops #41 Footer hint drift + overlay no-op arms (via tech-debt: drive numeric shortcuts and help from the command registry (single source) #33)
• feat: add high-signal services and findings (S3, IAM, EBS/EIP/snapshots, RDS security, ECS incidents) #42 High-signal services & findings: S3, IAM, EBS/EIP/snapshots, RDS security, ECS (via feat: carry resource tags/ownership on all service models #35)
• bug: ECS overview shows placeholder CPU/memory values #43 ECS placeholder CPU/memory
• tech-debt: small cleanups (duplicated table widths, misspelled model file, magic item count) #44 Small cleanups (via tech-debt: extract a shared list-view table/scroll/selection helper #30)

Spawned from this epic

• feat: remember each view's selection and scroll position across navigation #71 Remember each view's selection and scroll position across navigation
• tech-debt: normalize per-service fetch dispatch in refresh.rs #73 Normalize the per-service fetch dispatch in refresh.rs (carved out of tech-debt: replace per-service match arms with a service trait/registry #29, which covered row actions only)

Key dependency chains

• feat: cache service inventories with a TTL, keyed by profile+region (fast navigation) #26 closes bug: cost cache is a single global file, not keyed by profile or region (cross-account leak) #15 (unified profile/region-keyed cache)
• tech-debt: extract a shared list-view table/scroll/selection helper #30 unblocks tech-debt: give each view its own selection/scroll state (or clamp centrally) #31, feat: in-view resource search/filter and column sorting #39, ux: manage long-value truncation and standardize empty states #40, tech-debt: small cleanups (duplicated table widths, misspelled model file, magic item count) #44
• feat: add an in-UI error/notification surface (eprintln is invisible in the alternate screen) #16 unblocks bug: list fetchers turn any error into an empty, success-looking table #17, bug: Cost Explorer errors render as $0 spend and a flat chart #18, bug: region discovery silently collapses to us-east-1 on failure #19
• tech-debt: replace per-service match arms with a service trait/registry #29 + tech-debt: drive numeric shortcuts and help from the command registry (single source) #33 remove the per-service / per-key drift class (incl. SQS d/o gap, bug: numeric view-switch keys fire while an overlay is open, and key guards are inconsistent #23)
• feat: carry resource tags/ownership on all service models #35 unblocks feat: per-client tag focus filter across service views #8 (existing tag-filter issue) and feat: add high-signal services and findings (S3, IAM, EBS/EIP/snapshots, RDS security, ECS incidents) #42
• tech-debt: refactor rebuild_findings into a rule registry #34 enables test: add coverage for finding rules, model predicates, cost parsing, and region/profile logic #38; tech-debt: consolidate the three clients_for_region copies and the two aggregation reducers #32 is the home for bug: global (multi-region) EC2/RDS/Lambda fetches ignore the selected AWS profile #11/bug: cross-region describe rebuilds SDK config without the active profile #12

Status note

The #31 line above reads "per-view selection state", but that issue centralized
selection clamping into the shared list helper rather than storing selection per
view. Remembering a selection across navigation is #71, which is still open.