[FEATURE] Calibrated trust/verification score on diffs and PRs before merge

Summary

There's no single, calibrated signal on a diff/PR that aggregates the actual evidence of its trustworthiness — tests passed, review-agent findings, verification coverage — into one legible score before merge. Please add a "trust/verification score" surfaced on diffs and PRs Claude Code produces or touches.

Current behavior

Verifying an AI-generated change today means separately checking whether tests pass, whether a code-review agent flagged anything, and forming a judgment by hand from those separate signals. There's no single aggregated indicator that reflects how much verification a given change has actually received.

Motivation

Independent research on agentic coding workflows (CodeMySpec, "The Verification Gap in Agentic Coding") reports that 96% of developers don't fully trust AI-generated code, yet ship it anyway — precisely because there's no good, low-effort way to calibrate trust before merging. That's a workflow gap as much as a trust gap: the underlying verification signals (tests, review findings, coverage) often already exist, they're just scattered rather than presented as one legible readiness signal.

Proposed behavior

1. A calibrated score/badge on a diff or PR aggregating available verification signals: did tests run and pass, did a review agent (e.g. code-reviewer) run and find issues, what's changed test coverage, was this the product of a single pass or an iterated/reviewed one.
2. Transparent composition — clicking through shows exactly what fed the score, not just a bare number, so it's a summary of real evidence rather than a new opaque black box.
3. Should be advisory, not gating — surfaces the actual state of verification so the human merging can calibrate trust, not a hard block on merging.

Why it matters

• Directly targets a documented, quantified gap (96% low-trust-but-ship-anyway) rather than a general "make things feel more trustworthy" ask.
• Makes the verification work that often already happens (tests, review agents) legible as a single signal, instead of requiring the user to separately check and mentally combine multiple sources before deciding whether to trust a change.

Related issues (adjacent, not duplicates)

• #56913 Make autonomous Claude Code actually viable: tiered Opus brains + Sonnet workers + persistent state — broader autonomy-viability theme, different specific ask (architecture vs. a verification signal on output).

Notes

• Reviewed for duplicates before filing (multiple search passes: "aggregate confidence score before merging pull request", "trust score confidence aggregate before merge diff review", "readiness score tests review before merge") — nothing found proposing this specific aggregated signal, as of 2026-07-19.
• Sourced from market research: CodeMySpec, "The Verification Gap in Agentic Coding" (96% low-trust-but-ship statistic).