refactor(slm-backend): retire the 7 dead SLM monitoring modules whose function is live elsewhere (#16282)

[mrveiss]

Refs #16282. #16305, #16306 and #16307 (its sub-issues) carry the three modules that stay, and performance_monitor, which they still import.

Thinking Path

• tech-debt(slm-backend): 11 of the 12 modules in monitoring/ are imported by no live code #16282 found that 11 of the 12 modules in autobot-slm-backend/monitoring/ have no live importer. monitoring/__init__.py:12 imports only prometheus_metrics.
• I checked each module by reading source against base d5a0c6dcb. For 8 of them, the function already runs elsewhere, in the SLM API or the shared Prometheus recorders.
• The owner decided to wire the other 3 in rather than retire them (slm(monitoring): wire performance_benchmark in as a fleet benchmark, and fix its import of a module that does not exist #16305, slm(reconciler): fold performance_optimizer's tuning logic into the reconciler, then remove the module #16306, company-os: fold business_intelligence_dashboard's cost and ROI analysis into Company OS, and system-wide stats into the SLM #16307). All three import performance_monitor by its bare name, so it stays until the last of them lands (review finding; commit 3). This PR retires the other 7. The README records that whichever sub-issue lands last removes performance_monitor.
• No running path changes. The retired modules have no live importer, and performance_monitor itself can't be imported in the SLM, because it needs config.ConfigManager, which exists only in the main backend (autobot-backend/config/manager.py:51).

Retired (lines on base)	Where its function lives now (base d5a0c6dcb)
advanced_apm_system (877)	middleware/api_request_counter.py:35 with autobot_shared/monitoring/metrics/api_requests.py:39 record_request; api/performance.py:610 get_traces, :743 create_alert_rule
ai_performance_analytics (830)	api/npu.py:457 GET /metrics; in autobot_shared/monitoring/metrics/: performance.py:295 update_npu_metrics, knowledge_base.py:225 record_search, llm_provider.py:26 LLMProviderMetricsRecorder
claude_api_monitor (596)	autobot_shared/monitoring/prometheus_metrics.py:428 record_claude_api_request; autobot_shared/monitoring/metrics/claude_api.py:17 ClaudeAPIMetricsRecorder. It was already marked DEPRECATED (#348).
comprehensive_monitoring_controller (588), monitor_control (680)	main.py:688-689 run the monitoring and performance routers in-process, behind dependencies=_SM, so no separate monitoring process remains to control
metrics_adapter (222)	it wrapped get_metrics_manager(), which callers use directly
performance_dashboard (724)	api/monitoring.py:695 /dashboard, :354 /metrics/fleet

What Changed

Deleted (10 files)

• The 7 modules.
• monitoring/start_monitoring.sh. It launched only the controller (:191, :209) and individual modules (:225-228), none of which can run: the controller and APM/analytics are retired, and the monitor and BI can't be imported.
• monitoring/monitoring_config.yaml. Its only reader was monitor_control.py:186.
• tests/test_claude_api_monitor.py. It exercised only claude_api_monitor.

Baselines

• 4 entries each dropped from scripts/python_file_size_known_large.py and repo_tests/python_file_size_ratchet_baseline.py. performance_monitor's entry stays.
• The SLM performance_dashboard line dropped from pipeline-scripts/hardcoded_values_baseline.txt.
• .flake8:159 (performance_dashboard) dropped.

Operator scripts in autobot-infrastructure/shared/scripts/ (pytest's testpaths reach only shared/scripts/hooks, so CI collects none of these)

• test_phase2_migration.py records Claude calls through the metrics manager instead of ClaudeAPIMonitor.
• test_prometheus_metrics.py: removed the MetricsAdapter block, because its Prometheus writes are already exercised at :61-70. Also removed a duplicate logger assignment and a hardcoded IP.
• test_phase5_cleanup.py: the ClaudeAPIMonitor check now asserts that the file is absent and that record_claude_api_request is still defined. It reads this through the AST, the way the compat check does. Removed the SLM sys.path insertion, which existed only for that import, and the now-unused asyncio, Callable and _emits_deprecation.

Docs

• monitoring/README_PERFORMANCE_MONITORING.md is rewritten. It covers the live equivalents, the 3 modules being wired in, and who removes performance_monitor. The old internal addresses are gone.
• Retirement notes added in CONFIG_REMEDIATION_PLAN.md (the 3 retired-module lines only), system-state.md, CLAUDE_API_OPTIMIZATION_SUITE.md and ssot_config_shell_library_14041.md.
• The test(slm): whole-backend sweep RUN phase — 105 pre-existing runtime failures from sys.modules order-rot and never-run files (post-#11794) #11798 comments in test_api_request_counter.py and test_prometheus_registry_endpoint.py no longer name the retired module.

Commits

1. 547e111bc: the retirement.
2. c0c647809: removed the quick-deploy-verification.sh checks on logs/performance_monitor.log.
3. Keeps performance_monitor.py, its baseline entries, its doc lines and those log checks, so relative to base, commit 2 nets out to nothing.

Not touched (boundary)

• performance_monitor, business_intelligence_dashboard, performance_benchmark and performance_optimizer, their tests, and CostModelConfig (slm(monitoring): wire performance_benchmark in as a fleet benchmark, and fix its import of a module that does not exist #16305, slm(reconciler): fold performance_optimizer's tuning logic into the reconciler, then remove the module #16306, company-os: fold business_intelligence_dashboard's cost and ROI analysis into Company OS, and system-wide stats into the SLM #16307).
• The main backend's utils.performance_monitor family.
• shared/scripts/performance_dashboard.py, a different script.
• CHANGELOG history, and the generated api.ts, which only matches the names as words.

Verification

• No importer is left for the 7. After the change, git grep -wE over the retired names (bare name and path form) turns up only the README, the phase-5 constant, and the boundary items above. performance_monitor.py's own imports (:12-30) include none of the 7.
• Only the intended files differ from base. performance_monitor.py, quick-deploy-verification.sh and README_ASYNC_VALIDATION.md are identical to base.
• Hooks and linters. Pre-commit passed on every commit, including the Python file size guard, Detect Hardcoded Values (Strict), Function Length Check, the Python canonical check, flake8 F821 over the whole repo, and Detect Mass Deletions. black, isort and flake8 are clean over the changed Python files.
• CI is the evidence for the SLM and repo suites. Nothing was run locally.

Model Used

Claude Opus 5 (claude-opus-5)

Single-issue rationale

The owner split off #16282's three remaining modules as #16305, #16306 and #16307. They're different scopes (an SLM feature, the reconciler, and Company OS), and each lands in its own PR. What's left of #16282 is retiring the 7 modules whose function is already live elsewhere, and no other open issue shares that scope.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Advanced

Run ID: 25efd458-a641-49e4-be83-f44a6032ee12

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[mrveiss]

Review at c0c6478097: changes requested. Seven of the eight retirements hold, but performance_monitor.py can't go yet. It's a dependency of all three modules the owner ruled kept (09:05Z on #16282).

#	Severity	Where	Finding	Fix
1	High	autobot-slm-backend/monitoring/performance_monitor.py (deleted)	All three kept modules import from it by its bare name: • business_intelligence_dashboard.py:35: from performance_monitor import ALERT_THRESHOLDS (#16307) • performance_benchmark.py:27: from performance_monitor import VMS (#16305) • performance_optimizer.py:19: from performance_monitor import PerformanceMonitor (#16306) No live code imports those three today, so nothing breaks at runtime yet. But after this PR none of them can import, and the owner's ruling is that none of them is retired: they get wired in. So performance_monitor isn't "imported by no live code" in the sense the deletion rule means. It's a dependency of work the owner chose to keep. test_bi_dashboard.py stubs a fake performance_monitor module (:44, :89), so that test keeps passing and hides the break.	Either keep performance_monitor.py in this PR (retire seven), and retire it in whichever of #16305–#16307 lands last. Or move ALERT_THRESHOLDS, VMS and PerformanceMonitor somewhere the three kept modules import from, and point them at it here. Either way, record the dependency on #16305–#16307 (a note or a blocked_by), so wiring them in doesn't start from a missing import.

Verified, and it holds for the other seven (advanced_apm_system, ai_performance_analytics, claude_api_monitor, comprehensive_monitoring_controller, metrics_adapter, monitor_control, performance_dashboard):

• No code file references them in import or path form, and neither start_monitoring.sh nor monitoring_config.yaml is referenced outside docs. The three claude_api_monitor mentions left are in test_phase5_cleanup.py, which asserts that file is absent (:185). phase_validation_system.py:756 points at a different, older path, scripts/performance_dashboard.py.
• Every live equivalent the body cites exists on base d5a0c6dcb at its cited line. I spot-checked all 12: middleware/api_request_counter.py:35, api_requests.py:39 record_request, api/performance.py:610 and :743 and :833, api/npu.py:457, performance.py:295 update_npu_metrics, prometheus_metrics.py:428 record_claude_api_request, claude_api.py:17 ClaudeAPIMetricsRecorder, main.py:688, api/monitoring.py:695 and :354.
• Baselines only shrink, and only for deleted files: .flake8 −1, hardcoded-values −1, and both ratchet files −5. Nothing is added.
• Tests: test_claude_api_monitor.py goes with its module. The two SLM tests that change do so in comments only, and test_phase2_migration.py moves off ClaudeAPIMonitor onto the metrics manager.
• quick-deploy-verification.sh: its two checks of logs/performance_monitor.log go with the log's only writer, and the body flags it.

Refs #16282 is right either way, since three modules remain.

[mrveiss]

Addendum to the finding above: this PR's CI will go green even though the three kept modules can't import, so a green run is not evidence the deletion is safe.

• performance_benchmark.py (slm(monitoring): wire performance_benchmark in as a fleet benchmark, and fix its import of a module that does not exist #16305) and performance_optimizer.py (slm(reconciler): fold performance_optimizer's tuning logic into the reconciler, then remove the module #16306) have no tests. Only the ratchet baseline mentions them.
• business_intelligence_dashboard.py (company-os: fold business_intelligence_dashboard's cost and ROI analysis into Company OS, and system-wide stats into the SLM #16307) has two tests, and both install a fake performance_monitor in sys.modules before importing it: test_bi_dashboard.py:89–91 and test_bi_dashboard_10778.py:83–85. So they pass whether or not the real module exists.
• The import-hermeticity sweep reaches api/ and autobot_shared/, not monitoring/, so it won't flag them either.

Nothing in CI can see the break. It has to be settled in the diff (keep the module, or move the three symbols), and the tests for #16305–#16307 should import the real dependency rather than a stub.

[mrveiss]

Delta re-review since my changes-requested review at c0c6478097: approved at 20c6a53818. It merges on a settled green at that SHA. It now retires 7 modules, and performance_monitor.py stays until the last of #16305–#16307 lands.

Check	Evidence at 20c6a5381
performance_monitor.py is restored unchanged	Its blob is byte-identical to base d5a0c6dcb, so the three kept modules' imports (ALERT_THRESHOLDS, VMS, PerformanceMonitor) resolve again. quick-deploy-verification.sh is also byte-identical to base, so commit 2 nets out to nothing.
Keeping it doesn't keep a retired dependency	performance_monitor.py imports none of the seven retired modules, and it doesn't read the deleted monitoring_config.yaml or start_monitoring.sh. Its only file access is appending to a metrics file (:583).
Deleted set	Exactly the seven modules (advanced_apm_system, ai_performance_analytics, claude_api_monitor, comprehensive_monitoring_controller, metrics_adapter, monitor_control, performance_dashboard), plus monitoring_config.yaml, start_monitoring.sh and test_claude_api_monitor.py.
Baselines only shrink, and only for retired files	.flake8 and the hardcoded-values baseline drop performance_dashboard.py. Both ratchet files drop advanced_apm_system.py, ai_performance_analytics.py, monitor_control.py and performance_dashboard.py. Nothing is added, and performance_monitor's entries stay.
The dependency is recorded	#16282 (09:19Z): #16308 retires 7 of the 11, the rest stay, and the issue closes after its sub-issues rather than on this PR. The README records that whichever of #16305–#16307 lands last removes performance_monitor.py.

Everything from my first review of the seven stands: no code references, and all 12 cited live equivalents verified at their lines on base. Refs #16282 is right. When #16305–#16307 come for review, their tests should import the real performance_monitor, not the sys.modules stub test_bi_dashboard*.py installs, so the dependency is actually exercised.

[github-actions]

✅ SSOT Configuration Compliance: Passing

🎉 No new hardcoded values of either class — ssot and other both block.

Known backlog in pipeline-scripts/hardcoded_values_baseline.txt is suppressed and tracked in #14371.

[mrveiss]

Content review: approve.

Verified independently against base ([] no manual link, body is Refs-only): grepped for live importers of all 7 retired modules across autobot-slm-backend — none found outside the retired set itself (metrics_adapter.py imports claude_api_monitor, but metrics_adapter is also being retired in this same PR, so that's not an external dependency). Spot-checked two of the table's line-number claims against base: prometheus_metrics.py:428 record_claude_api_request and api/monitoring.py:354 /metrics/fleet both match exactly.

Good boundary discipline: performance_monitor stays because all 3 modules the owner chose to wire in (#16305/16306/16307) still import it by bare name — retiring it here would break work in flight elsewhere. The README documents which sub-issue removes it last, so that decision isn't lost.

closingIssuesReferences=[] matches body (Refs #16282 only). CI green, no missing required contexts. Ready.