Pre-merge impact
“What breaks if I change this?”
Blast radius, hidden dependencies, and structural risk — answered before merge.
Speed vs manual trace · indicative · local CLI only · not financial advice
Free tierLocal CLI< 3 min to first impact
Validated on real codebasesPayload • Directus • Nx • VS Code
14 repositories tested0 false architectural reports after graph gatingSee validation →
"What breaks if I change this?" — every risky change collapses into that question, and teams burn hours trying to infer the answer.
The unanswered question
What breaks if I change this?
5 change types · 1 question before merge
Example change · Upgrade package → same question
Upgrade package → “What breaks if I change this?”
Tap any change · same question every timeClick any change type · arrow keys work
The old way
How teams answer today
With arc impact
Answer time
~720× faster
Complementary — not a replacement. Cursor writes the code. ArcSight shows the impact. Each row is the question that tool category was built to answer.
Cursor writes the code. ArcSight shows the impact.
Complementary · same machine · different job
Same question from §02 — mapped by workflow phase below.
The answer
ArcSight plugs the gap
Only tool for before merge · What breaks if I change this?
Watch the cycle · hover to explore · arrow keys workWatch the cycle · tap any tool to explore
Cursor & Claude
Write the code
ArcSight
Shows the impact
before merge only
Every other row needs production, a PR, or a guess. ArcSight only needs a diff.
Structured CLI output — same fields as the live hero demo, frozen below.
Tells you what breaks before you change code — as structured CLI output.
Not a summary. Same fields as the hero terminal, frozen below.
After category gap — same schema as the live hero demo, frozen here.
arc impact src/api/UserController.ts → HIGH · 12 modules · 3 break paths
$ arc impact src/api/UserController.ts
Change risk
high
What breaks
Blast radius
12
Evidence
Complete
Recommendation
Review before merge
Risk signals
Dependency cycle detected
Watch high risk ↔ refused · tab 1 = frozen hero output · arrow keys workWatch the cycle · tab 1 = hero output
Hero animates the reveal · proof freezes the schema. View full example analysis →
The moat is not confidence — it refuses when evidence is insufficient. Compare both outcomes below.
Other tools invent confidence. ArcSight stops when evidence runs out.
The moat is refusal — same CLI schema, two honest outcomes.
Next: how it fits your workflow — compare what happens with complete evidence vs insufficient evidence.
arc impact PaymentGateway → insufficient evidence → analysis stops
Refused analysis
$ arc impact PaymentGateway
Analysis incomplete
Reason
Dynamic runtime dependency detected
Evidence
Insufficient
Recommendation
Run runtime trace collection
Watch refused ↔ success · starts on refusal · click panels · arrow keys workWatch the cycle · tap panels to compareSame outputs as the live hero demo — static reference above
ArcSight would rather refuse than give you a confident wrong answer.
Three commands — find risk, fix it, prevent regressions. Same journey engineers use after trust is established.
impact → fix → watch — find risk, fix it, prevent regressions.
After trust comparison — imagine yourself running the three commands.
$ arc impact src/api/UserController.ts
Risk
HIGH
Affected contracts
Blast radius
12 modules
Know what breaks.
Watch the cycle · click a command · arrow keys workTap a command · watch reveals regression
14 public repos tested — 4 featured below. Impact when evidence exists. Refusal when it does not.
Validated on 14 public repositories.
Payload
Impact625 modules · 2701 edges
Identified: src/types fan-in 225
Result: Focused investigation on the true platform contract.
Cycles all 4 featured repos · arrow keys work · Tap any repo · 10 more in full record
14 repos tested · 6 strong outcomes · 7 honest incomplete · 0 false positives.
Claims link to artifacts — not adjectives.
Outcome counts after graph-gating validation — not internal methodology labels.
After validated repos — each procedure links to the underlying artifact or published record.
Verdict: ArcSight passes when it proves impact — and passes when it refuses without proof.
Failure-Mode Testing → incomplete evidence → analysis stops
Incomplete evidence paths terminate with Analysis incomplete — no synthetic confidence scores.
Outcome: Refusal emits Reason, Evidence, Recommendation — never a invented risk score.
Watch refusal → TTFC → full record · open artifacts for reproduction stepsWatch the cycle · tap rows to inspect
Straight answers to the questions engineers ask before running arc impact.
Install hesitation lives in three questions — the rest is detail.
Q1–Q3 map to the comprehension test. Practical objections below.
After validation record — straight answers before install recap.
What does ArcSight do? → impact before merge · local CLI
Tells you what breaks before you change code.
It tells you what breaks before you change code — blast radius, hidden dependencies, and structural risk — from local CLI analysis on your machine. See on homepage → Frozen proof on homepage →
Cursor writes the code. ArcSight shows the impact when that change propagates through a real codebase. Most teams use both. See on homepage → Category gap on homepage →
ArcSight refuses when evidence is insufficient. It would rather mark an analysis incomplete than give you a confident wrong answer. View trust comparison → See on homepage → Refusal testing record →
No. The CLI runs locally. Your source code never leaves your machine. Trust boundaries record →
TypeScript and Node.js codebases today, including monorepos and large repositories. Install requirements →
Install the CLI and run arc impact on your codebase. See change impact analysis before you merge — no source upload required. Homepage install flow →
Watch Q1 → Q2 → Q3 → practical objections · arrow keys workWatch the cycle · tap to pause
Still deciding? Three questions recap → · Install commands ↓
Q1–Q3 are the pass/fail test. Answer them unprompted — then run your first impact.
Can you answer all three without scrolling back?
After FAQ — comprehension check before first impact.
Tells you what breaks before you change code. → same answer as §01 live output
Install once. Run impact on the file you are about to change. Median time to first result: under 3 minutes.
< 3 min · install → impact → answer.
After recap — same commands as the hero demo, run locally.
Install → npm global install · shell ready
npm install -g arcsight
$ arc impact src/api/UserController.ts
Output
3 of 3Expected output
Free to try · Pro adds arc watchCompare plans →
One install. One command. Answer what breaks before you merge. Replay live demo →
Passed the comprehension check? Review Q1–Q3 recap →