P1Project
Restart churn
The same project is restarted repeatedly instead of continuing from a useful handoff.
→ Improve handoffs or project instructions.
Enterprise AI analysis
Find where AI work gets stuck.
Turn captured sessions into explainable workflow patterns. See where context is wasted, work is repeated, policy drifts, or model cost outruns the task—then open the evidence behind every finding.
signal.evidenceexample · open
P2 · PROJECT
Groundhog files
12
Sessions
4
Contributors
7
Repeated files
src/auth/session.tsread in 8 sessions
docs/architecture.mdread in 6 sessions
src/lib/api.tsread by 3 contributors
Likely next action
Add durable project context so the next contributor does not reconstruct the same map from raw files.
From totals to diagnosis
High usage is a number. A Signal tells you what to inspect.
Signals look for specific, repeated patterns in captured work. They connect the finding to the relevant sessions, files, models, projects, or teams so the next action is grounded in evidence.
What
A named workflow pattern—not a generic usage spike.
Where
The user, project, team, or organization where evidence concentrates.
Why
Linked sessions and inputs that explain why the finding appeared.
Pattern catalog
Ten signals for recurring AI workflow waste.
Each pattern is intentionally narrow. The goal is to explain what kind of friction is present and point to the evidence—not label every high-token session as waste.
P2Project
Groundhog files
The same files are re-read across sessions or contributors to recover context again.
→ Add durable project context or a file brief.
P3Habit
Exploration without landing
A session spends heavily on searching and reading but produces few or no edits.
→ Tighten task briefs and entry points.
P4Habit
Session idle gaps
The next user prompt repeatedly arrives after the configured wait threshold.
→ Review interruption and handoff patterns.
P5Habit
Session bloat
Sessions continue past the member's normal completion point while context keeps growing.
→ Define handoff or session-splitting guidance.
P6Organization
Compression policy drift
Supported hooks remain off or observe-only when the organization requires enforcement.
→ Check policy sync and stale installations.
P7Project
Project onboarding cost
New contributors use materially more tokens than veterans on comparable project work.
→ Build or improve a project context pack.
P8Team
Cross-team duplicated work
Separate teams repeat substantially similar work without finding the earlier result.
→ Make the prior work or artifact discoverable.
P9Habit
Context-weight burn
A session carries much more cached context through each tool call than its normal baseline.
→ Split, hand off, or delegate heavy sessions.
P10Organization
Model–task mismatch
A costly model is used where comparable work is routinely completed with a lighter model.
→ Set targeted model-selection guidance.
Scope the response
Fix the system at the level where the pattern lives.
A higher-scope Signal does not mean everyone caused it. It means the repeated evidence is broad enough that fixing one session or one user may not solve the underlying problem.
- 01Habit
Change how a session is run.
Improve prompts, entry points, handoffs, session boundaries, or interruption patterns.
- 02Project
Fix context once for the codebase.
Add durable project maps, file briefs, onboarding context, or clearer local instructions.
- 03Team
Make good work reusable.
Share conventions, surface prior artifacts, and remove repeated discovery across contributors.
- 04Organization
Repair policy or selection guidance.
Correct installation drift and focus model policy where evidence shows repeated leverage.
Evidence, not a verdict
Reduce the search space. Keep a human in the decision.
Signals are investigation leads—not employee-performance scores, automatic policy violations, or guarantees that a suggested change is correct.
01
Capture
Normalize session metadata, tool calls, project attribution, model usage, and policy state.
02
Evaluate
Measure specific patterns over completed sessions and compare with relevant historical baselines.
03
Surface
Group findings by user, team, project, or organization with projected opportunity or open count.
04
Investigate
Open linked evidence before acknowledging, resolving, dismissing, or routing a change.
P2
The same files are read across multiple sessions.
Inspect whether the project needs a map, a current README, or one durable file explanation.
P3
Search and reading dominate while edits stay low.
Give the next session a file, symptom, failing command, or suspected module as its entry point.
P10
Comparable work succeeds with a lighter model.
Write targeted model guidance instead of blocking premium models globally.
Deterministic anomaly analysis
Catch unusual activity without a black-box classification.
Nightly detectors explain what matched, preserve occurrence history, and link to sessions when evidence is available. Every detector includes a guardrail for how the finding should—and should not—be interpreted.
AD-001
Usage baseline deviation
Daily session or token volume materially above that member's own rolling history.
Guardrail: Compares each member with themselves, not with an organization-wide threshold.
AD-002
New technology fingerprint
A first-seen combination of dependencies, imports, manifests, tools, or project activity.
Guardrail: Novelty is a review lead, not evidence of a policy or access violation.
AD-003
Runaway or looping session
Repeated failures, identical retries, failed build loops, abnormal duration, or high tool volume.
Guardrail: Indicates possible operational waste, not malicious behavior.
Product note · July 2026
RCLM Signals: Patterns for Finding AI Workflow Waste
Read why Signals uses specific patterns, how the evidence is calculated, what each pattern suggests inspecting next, and where today's telemetry still limits the result.
Enterprise session analysis
Give every finding an evidence trail.
Connect recurring workflow friction and deterministic anomalies to the sessions your admins need to review—across teams, projects, models, and providers.