March was the month Re-entry went from "risk scores on PRs" to "a full operational intelligence layer." We shipped 30+ changes across risk scoring, the dashboard, integrations, and platform infrastructure. Here are the highlights.
Team Risk Intelligence Report
The biggest addition this month: a comprehensive risk intelligence report you can generate for any team.
It answers the question every engineering leader asks: "How accurate is our risk detection, and where are we most exposed?"
The report includes:
- Prediction accuracy — precision, recall, and false positive rate computed from resolved assessment outcomes
- Risk distribution — breakdown of low, medium, high, and critical assessments across all PRs and pushes
- Assessment coverage — total assessments, repositories monitored, and period activity
- Top risky files — the 10 files most correlated with incidents and reverts
- Hotspot directories — directories that repeatedly cause problems
- Risk trend — a smooth chart showing how your average risk score has changed day by day
- Intervention summary — blocked merges, required reviews, created tickets, and escalations
- Developer profiles — opt-in, GDPR-compliant view of per-contributor risk patterns (anonymous mode on by default)
The report is tier-gated: free users see calibration, outcomes, distribution, and coverage. Pro and enterprise users get the full picture including file heatmaps, trends, and intervention data.
Per-File Risk Heatmap
Every PR assessment now shows a file-level risk map. Files that have historically been involved in incidents or reverts are highlighted with intensity indicators.
This means you can instantly see: "This PR touches 2 files that have caused production incidents before." No digging through git blame needed — the institutional knowledge is surfaced automatically.
Repository Context Learning
The risk engine now builds context profiles for every monitored repository. It tracks which files have been reverted, which directories cluster around incidents, and which authors typically work in which areas.
This context is fed back into scoring, so assessments get more accurate over time. A change to a file that has been reverted three times in the past month will score higher than the same change to a stable file.
Bitbucket Integration
Re-entry now supports Bitbucket Cloud alongside GitHub. Connect your workspace, select repositories, and get the same PR risk monitoring, automated governance, and code review analysis.
The integration includes OAuth authentication, webhook delivery, and full parity with the GitHub feature set.
Smarter Risk Scoring
We fixed a category of false positives that had been bothering pilot users:
- Non-code files (.md, .json, .yaml, lock files) no longer trigger code injection warnings
- Lock-file-only PRs are capped at low risk
- Linear tickets are now detected alongside Jira — the engine is tracker-agnostic
- Risk explanations use plain English levels (Low/Medium/High/Critical) instead of misleading percentages
- Score deduplication prevents redundant assessment records when a PR hasn't changed between polls
Dashboard Improvements
The dashboard got a significant UX pass:
- Keyboard shortcuts — navigate signals with J/K, expand with Enter, focus the command bar with /
- Stacked sidebar — risk summary stays visible while chatting or reviewing PRs
- Filter collapse — filters hide behind a single button to reduce visual noise
- Score history — click any PR's score badge to see a timeline of how the risk score evolved
Other Changes
- Password reset flow (email-based, secure token)
- Redesigned risk driver display with severity indicators
- Wrong-branch PR detection (feature → main instead of develop)
- Auto-assign reviewers on high-risk PRs via policy actions
- Webhook scoping fix — only selected repos are analyzed
- GitHub disconnect now fully clears cached data
- Public changelog with search and filtering
What's Next
April is focused on AI-powered code comprehension (the risk engine will read and understand your actual code changes), self-hosted model support, and assessment feedback loops. More on that soon.