Select a box or use Tab and Enter. On a narrow screen, scroll sideways to reach later stages.
Transcript snapshot / stage totals
Usage by pipeline stage
Each stage has its own metric switch. Bars use a shared scale for stage comparisons.
Loading stage metrics…
Read the process and gatesPublished specsIssue and milestone ledgerPublished issue mapView the dated journey recordTask, time, model, and token ledgerStill map
Transcript snapshot / loading…
Atlas work, recorded
Loading the latest completed-turn snapshot… The full ledger shows each task and decision window.
- Calendar window
- —
- Non-overlapping wall time
- —
- Total tokens
- —
- Models / reasoning
- —
Linked Codex chats / completed-turn usage
Main and sub-agents
Loading agent metadata…
- Main chats
- —
- Sub-agents
- —
- Finished turns
- —
- Active
- —
- Tokens
- —
| Loading agent metadata… | |||||
Mac worker vs cloud worker
Checking source synchronization…
| Source metrics arrive on refresh. | ||||||
Active turns are shown by status and configuration. Their usage enters the totals only after a later completed-turn capture. Wall time can overlap across agents.
Transcript usage / curated primary stage
Work by pipeline stage
Each finished turn belongs to one primary stage, even when its work crossed stages. Token counts are not split or estimated. Summed wall time can overlap; the union column removes overlap within each stage.
- Stages
- —
- Turns
- —
- Tokens
- —
- Phases
- —
- Unclassified
- —
| Loading stage totals… | |||||||||
Checking source coverage…
Usage by SDLC phase
Turn tokens and time stay under one primary phase. Measured commands are separate, opt-in evidence; earlier uninstrumented tests remain unknown.
| Loading phase totals… | |||||||
Phase attribution provenance will appear after capture.
Main and linked sub-agent history
Recorded task turns
Finished and interrupted turns retain recorded usage. Active turns appear in the agent table above and enter this detail after completion.
- Recorded
- —
- With usage
- —
- Usage missing
- —
- Estimated
- —
- Tokens
- —
Loading turns…
| Loading turn detail… | |||||||||||||
GitHub tracker / point-in-time record
Issue journey
Every issue has a number, short purpose, type, tags, and recorded lifecycle dates. Assignment marks a claimed start; an open spec alone is not implementation in progress. The paused issue count uses the needs-info tag.
Atlas issues are in a private GitHub repository. If an issue link shows 404, sign in to GitHub with an account that can access comtus/atlas. GitHub CLI authentication does not sign in the browser.
- Created
- —
- In progress
- —
- Open / unclaimed
- —
- Paused
- —
- Closed
- —
Loading issue snapshot…
Evidence-linked events
Milestone timeline
Start, stop, and pause events record actual decisions and gates. A paused path or review gate does not mean the whole project stopped, and is separate from the paused issue count.
- Loading milestone events…
GitHub check metadata / sourced Codex turns
Review activity
Loading review activity…
Checking saved review data…
| Loading review rows… | |||||||
Greptile elapsed time is GitHub provider check time, not LLM compute. GitHub does not expose Greptile model or token usage. Sub-agent tokens are a subset of the transcript ledger and are not added to its totals.
GitHub PR lifecycle / sourced metadata
Pull requests and review cycles
Review turns are submitted non-author GitHub reviews, separate from Codex task turns. Review counts include activity after merge; average turns to merge counts only pre-merge reviews. Comments include inline comments, PR discussion, and written review submissions by non-authors.
Loading PR snapshot…
- Created
- —
- Reviewed
- —
- Merged
- —
- Open
- —
- Avg. create → merge
- —
- Review comments
- —
- Review turns
- —
- Avg. turns to merge
- —
Loading comment breakdown…
Avg. create → merge time
Average duration by UTC merge date · hours and minutes
Loading merge durations…
PR activity over time
Created · first reviewed · merged, by UTC date
Review turns per PR
Submitted non-author reviews on each PR
| Loading PRs… | |||||||
Merge time is measured only for merged PRs. Counts come from GitHub history; review or agent activity without a linked GitHub review is not inferred. GitHub sign-in with access to the private Atlas repo is needed to open PR links.
Architecture and code-review evidence
Audits and codebase evolution
Loading audits…
Checking committed history…
- Assessments
- —
- Findings
- —
- File changes
- —
- Module/day rows
- —
- Commits
- —
File-change events by module and day
Folders roll up to two levels; docs/journey keeps its detail. Expand an arrow to explore children. Badges show cumulative file-change events; cells show daily events. A file changed in multiple commits counts multiple times. Current tracked-file inventory is shown separately above.
Assessment runs
| Evidence | |||||
|---|---|---|---|---|---|
| Loading assessments… | |||||
Findings and mitigations
| Mitigation link | |||||||
|---|---|---|---|---|---|---|---|
| Loading findings… | |||||||
From the current dashboard data
Project signals
Daily trends show token usage including labelled estimates, run duration, and evidenced implemented capabilities. Tokens by model and reasoning use the same stage and phase filters. Choose a chart view on each card; values stay labeled as the data refreshes.
- Tokens
- —
- Run duration
- —
- Implemented capabilities
- —
- Test attempts
- —
- Reviews
- —
Token usage over time
Loading transcript snapshot…
Total—
Recorded run duration
Loading transcript snapshot…
Total—
Sum of Codex run durations by turn completion date. Includes tool waits and overlapping agents; it is not pure model compute time or project wall time.
Implemented capabilities
Loading issue snapshot…
Total—
Capability evidence, live observations and pending scope
Distinct capabilities with bound implementation evidence, independent of parent issue closure. Current live observation and full qualification are separate. Closed tagged tracker issues remain a separate completion metric.
Automated test attempts per day
Loading activity snapshot…
Total—
Completed reviews per day
Loading activity snapshot…
Total—
Tokens by model and reasoning
Loading transcript snapshot…
Total—
Scroll sideways if a day is cut off.
Issue state
Loading issue snapshot…
Total—
Token mix
Loading transcript snapshot…
Total—
Cached input is part of input tokens. Counts combine recorded usage and labelled estimates, not cost. Estimated cache/reasoning splits remain unavailable.
Main vs sub-agent usage
Loading transcript snapshot…
Total—
Tokens from finished turns include labelled estimates. Active turns are excluded until they finish.
Recorded command evidence
Automated tests
Inspect recorded unit, system and end-to-end command runs from Mac and Cloud.
Loading automated test evidence.
- Recorded
- —
- Passed
- —
- Failed
- —
- Interrupted
- —
- Untimed
- —
- Elapsed
- —
Summary by feature, stage and UTC date
| Feature | Stage | Date UTC | Runs | Passed | Failed | Interrupted | Elapsed | Untimed |
|---|
Capture coverage and limits
| Exit | Details |
|---|
Recorded checks and repeatable steps
Manual tests by domain
Choose a domain and case to see navigation, expected results and evidence.
Loading manual test checkpoint.
Architecture and product decisions
Atlas design record
Current context and accepted decisions are separated from proposed design artifacts awaiting review.
Open this dashboard with the Atlas Journey server (python3 scripts/serve_journey.py) to read these local documents. A basic static file server does not provide the document viewer.
The agentic build story
- Atlas agentic SDLC · illustrated storyboardEight scenes from owner decisions to the local preview, with evidence behind each scene
Branching service wizards
Research and specifications for answer-driven intake and claims across six products. The approved 20-slice task graph is published with 41 blockers. Cloud implementation is assigned; prerequisites remain pending. No expanded implementation completion is claimed.
- Branching insurance service wizard specificationStories, shared contracts, product branches and acceptance gates
- Insurance service branch catalogService choices, conditional questions and evidence paths
- Insurance intake workflow research · 7 OctoberPrimary-source enrollment and servicing workflows
- Insurance claim workflow research · 7 OctoberPrimary-source claimant, evidence and review workflows
- Wizard gap inventory · 7 OctoberCurrent code boundaries and proposed extensions
- Approved 20-slice wizard task graphOwner approved; ticket publication complete
- Published 20 tickets / 41 blockersTracker links, native dependencies and Cloud/Mac ownership; prerequisites pending
Branching service wizard specification issue · Draft specification pull request. Private references are marked unavailable; allowlisted documents open from this bundle.
Current context and accepted decisions
- Atlas system contextCurrent context; open questions remain visible in the document.
- ADR 0001 · Prototype boundary
- ADR 0002 · Login and business context
- ADR 0003 · Content, workflow, and record ownership
- ADR 0004 · Bilingual form releases
- ADR 0005 · Durable commands
- ADR 0008 · TypeScript BFF
- ADR 0009 · Keycloak-backed sessions
- ADR 0010 · Local certification, then cloud hosting
- ADR 0011 · Fastify and TypeBox contractsAccepted September 30; adoption pending
- ADR 0012 · Private gRPC adaptersAccepted September 30; browser uses tRPC over HTTPS under ADR 0013
- ADR 0013 · React SDUI BFF tRPC boundaryAccepted September 30; migration pending
ADR 0006 (direct SPA authorization) and ADR 0007 (CloudHub) are superseded. ADR 0008’s dedicated-VM placement is also superseded by ADR 0010.
Design artifacts and delivery
Architecture drafts are proposals until reviewed and accepted. The build guide records current measurement practice and open improvements.
- Mac and cloud metrics synchronizationSource thread identity, timing and token coverage
- Manual Mac preview syncExplicit Sync App control; initial activation and browser verification pending
- Architecture indexSelected boundaries; delivery status recorded separately
- Screen designs and interactive storyboardTen reference layouts adapted to Atlas; preview and delivered behavior labelled
- Domain and journey reference analysisScope, SDUI bindings, role boundaries and architecture alignment
- Component designSelected design with planned migration labels
- Fastify, TypeBox and gRPC designAccepted boundaries and incremental migration gates
- tRPC and SDUI boundary designSame-origin HTTPS, session security and pinned form releases; migration pending
- BFF architecture amendment reviewOne documentation finding resolved; implementation gates remain
- SDUI task flowProposed
- Deployment viewProposed
- Decision indexAccepted and superseded ADRs
- T3Code migration · 12 OctoberVerified continuation, metrics, preserved history and explicit platform limits
- Retrospective improvements · 11 OctoberEnvironment, qualification and publication tools; adoption and telemetry limits
- Mac and Cloud retrospective · 10 OctoberOne consolidated report; 10 prioritized environment recommendations with evidence and coverage limits
- Agentic SDLC metrics guideCurrent capture, gaps, and next gates
- Architecture audit · 30 SeptemberRecommendations and owner selection
- PR dashboard code review · 30 SeptemberStandards and spec findings with mitigations
- PR dashboard Greptile review · 30 SeptemberBot findings, fixes, and re-review
- Durable submission lifecycle planApproved plan; implementation blocked on recorded gates
- tRPC draft code review · 30 SeptemberStandards clear; two Spec P2 mitigations pending verification
- Dashboard refresh review · 30 SeptemberStandards and spec findings with mitigations
- External agent run receiptsSanitized Grok usage; excluded from Dashboard charts
- Application screen gallerySynthetic local customer, reviewer, and operator views