Skip to content

feat(admin): trace performance, grouped timeline and report exports - #629

Merged
FlyM1ss merged 5 commits into
mainfrom
feat/admin-trace-performance
Oct 13, 2026
Merged

FlyM1ss merged 5 commits into
mainfrom
feat/admin-trace-performance

Conversation

@MrParamecium

@MrParamecium MrParamecium commented Oct 9, 2026 •

Copy link
Copy Markdown
Collaborator

Admin trace details now connect trading outcomes to recorded decisions: an exact-run equity curve links trades to collapsible decision/step groups, with reasoning summaries and optional raw JSON. Expanded groups survive polling and failed groups open by default.

Administrators can inspect stored metrics and configuration for the trace’s exact backtest run across owners through an admin-only endpoint. Run configuration explicitly labels missing values. JSON and Markdown downloads include every recorded event page up to a fixed sequence cutoff, redact sensitive key fields and common credential patterns, and include artifact references without bundling attachments.

Validation: 349 admin/composition tests passed; 7 trace frontend tests passed again after the final summary adjustment. Tests cover cross-owner admin reads, unauthenticated rejection, 205-event exports, concurrent appends, download headers, invalid formats, and grouping. Synthetic browser preview verified trade-to-decision expansion and read-only backtest disclosure. Local 9017 service updated. Production Neon and a real end-to-end backtest were not exercised locally; export redaction is pattern-based.

@vercel

vercel Bot commented Oct 9, 2026 •

Copy link
Copy Markdown
Contributor

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
agentic-trading-lab Ready Ready Preview Oct 13, 2026 5:06pm UTC

@MrParamecium MrParamecium changed the title feat(admin): show trace equity and linked trade decisions feat(admin): trace performance, grouped timeline and report exports Oct 9, 2026
- Poll the trace envelope before events (then events + performance in
  parallel), so a run that finishes between reads cannot stop polling with
  its last flush and terminal event unread; re-render only on change and
  restore focus.
- Trading log: link fills through the execution event's decision_id and the
  per-action reasoning_summaries; show the latest 50 newest-first with the
  true count.
- Backtest report: fetch once per trace+status, so restoring `open` on each
  re-render no longer refetches; read llm_model (agent_runs has no
  model_name).
- Performance: report the run's stored total_return / max_drawdown instead
  of recomputing against the first equity mark.
- Export: status from the events inside the boundary (failed != incomplete);
  word-based key redaction (token counts stay visible); Markdown-escape agent
  text; linear-time email/newline handling.
- Null-guard renderTraceDetail; tokenise colours, drop dead timeline rules,
  bound the sticky rails; restore the rejected-query-name scan on all read
  modules.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@FlyM1ss
FlyM1ss merged commit 884fd3f into main Oct 13, 2026
8 checks passed

This branch was successfully deployed

1 active deployment
Preview — f73bc576 Deployed Oct 13, 2026 by vercel[bot]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants