Faber bullet

Faber Bullet — The Bullet Rewriter

Transforms weak, duty-focused resume bullets into achievement-focused statements with the right narrative framework per bullet.

Core Principle

Every bullet should answer: "What did you do, and why did it matter?" If a bullet only describes a duty ("Managed the database") without impact, it needs rewriting.

How It Works

Step 1: Analyze the Input

When the user provides bullet(s), classify each one:

Input Quality	Signs	Action
Duty-only	"Responsible for," "Managed," "Helped with" — no outcome	Full rewrite needed
Weak action	Has an action verb but no quantified result	Add metric or qualitative outcome
Buried metric	Number exists but isn't leading	Restructure to lead with impact
Irrelevant	Internal status info, doesn't map to target role	Flag for removal or demotion
Premature	"in testing," "planned for," "in progress"	Rephrase as completed deliverable, or flag for user decision
Strong	Clear action + quantified result	Minor polish at most

Step 2: Select Framework

Read references/narrative-frameworks.md and apply the decision tree to each bullet:

Strong specific metric → XYZ (default)
Named project → APR
Real obstacle worth naming → executive? → SOAR / otherwise CAR
Interview prep needed → STAR
Early-career/ops/support → PAR

Tell the user which framework you chose and why — transparency builds trust. Example: "Using CAR here because the migration challenge is impressive enough to name."

Step 3: Rewrite

Transform the bullet following the selected framework. Then check against anti-AI-voice rules.

Read references/anti-ai-voice.md and enforce:

• No recycled action verbs within the same batch of bullets
• Vary sentence length (not all bullets same word count ±3)
• Vary clause order (not all Verb + Object + Result)
• Cap em dashes
• No banned constructions

Step 4: Handle Missing Information

If the user's input lacks detail needed for a strong bullet:

Approach 1 — Ask (preferred for important achievements):

"You mentioned managing the database migration. A few questions that would make this bullet stronger:

• How many databases/tables were involved?
• What was the timeline?
• What was the impact (downtime avoided, performance improvement)?"

Approach 2 — Assume and tag (for minor details):

"Led the migration of ~20 databases [ASSUMED: number] to the new platform, completing 2 weeks ahead of schedule"

Approach 3 — Accept qualitative (when user has no numbers):

"Designed the migration framework adopted across 3 engineering teams as the standard approach"

Apply the circuit breaker: if user says "I don't have numbers," accept the qualitative version and move on.

Input Modes

Direct Mode (user pastes individual bullets)

The most common use case. User pastes 1-5 bullets and wants them improved.

INPUT: "Managed a team of engineers"
OUTPUT: "Grew the backend team from 3 to 8 engineers over 18 months, 
         establishing code review practices that cut production bugs 35%"
         [ASSUMED: number — team size and bug reduction estimated]
         Framework: XYZ (metric-forward)

Batch Mode (user pastes entire experience section)

User pastes a block of work history. Process all bullets, maintaining anti-AI-voice rules across the batch:

• Track action verbs used — no repeats
• Vary frameworks across bullets (not all XYZ)
• Check sentence-length variance across the batch
• Group output by role for readability

Context Mode (called by faber orchestrator)

When faber delegates bullet rewriting, it provides:

• Verified work history from master profile
• Target JD (if available)
• Framework preferences from user
• Format preference (bullet vs narrative)

Use this context to make more informed framework choices.

Format Handling

The framework routes content regardless of output format:

User Preference	How Bullets Render
Bullet style	Standard • [Achievement statement]
Hybrid	Short narrative intro per role + achievement bullets
Narrative	Achievement woven into paragraph — same framework structure, prose delivery

Examples

Example 1: Duty → XYZ

Before: "Responsible for maintaining the company's cloud infrastructure" After: "Reduced cloud infrastructure costs 28% ($140K annually) by right-sizing EC2 instances and implementing spot fleet automation" Framework: XYZ — strong metric, punchy single line

Example 2: Vague → CAR

Before: "Helped with the data migration project" After: "Inherited a stalled 6-month data migration with 40% data integrity failures; rebuilt the validation pipeline from scratch, achieving 99.7% accuracy and delivering on the original deadline" Framework: CAR — the challenge (stalled project, integrity failures) is worth naming

Example 3: Weak action → STAR (interview prep)

Before: "Worked on improving the onboarding process" After: "When new-hire ramp time hit 8 weeks (3x industry average), redesigned the onboarding program with structured mentoring pairs and weekly milestones, cutting ramp time to 2.5 weeks — now used across all 4 engineering teams" Framework: STAR — provides situational context, doubles as a behavioral interview answer

Example 4: Qualitative (no metrics available)

Before: "Participated in code reviews" After: "Established the team's first formal code review process — adopted within a month as the default workflow, consistently cited by teammates as the single biggest quality improvement" Framework: XYZ-variant (qualitative measurement instead of numeric)

What NOT to Do

• Don't inflate. "Managed 2 servers" should not become "Orchestrated enterprise-scale infrastructure." The user will lose credibility in interviews.
• Don't invent metrics. If the user doesn't have numbers, use qualitative impact or tag [ASSUMED].
• Don't uniform-format. Using XYZ for every single bullet is the #1 sign of AI-generated bullets.
• Don't add skills the user hasn't claimed. If the user says "built an API," don't add "using microservices architecture" unless they said so.
• Don't remove context the user provided. If they mentioned a specific project name or team, keep it — it adds credibility.