applications

name	applications
version	1.1
description	Track job and volunteer applications with CV snapshots, status updates, conversion metrics, and reply predictions
triggers	["откликнулась","applied to","sent CV","status update","conversion","prediction","probability"]

Applications Tracking

Manage the full lifecycle of job and volunteer applications: snapshot CVs, track statuses, calculate conversion, and predict reply probability.

---

Core Principle

CVs in role folders evolve. Applications must be frozen.

Every time the user sends a CV, we snapshot the exact text into the application folder. This is not a copy — it's a historical record of what was actually sent.

---

Stage Definitions

The pipeline tracks two things: conversion funnel (how many ever reached each stage) and current status (where each app is right now).

Stages

Stage	What happened	Who's involved
Sent	CV sent, no response yet. Stays Sent until company replies or user decides to close.	—
Replied	Company responded in any way (positive, negative, forwarded to team). Any human response counts. A rejection = Replied + Rejected.	Recruiter / HR / Contact person / Referral conduit
Screening	First live conversation happened. Purpose: basic fit check, "do we want to continue talking?". Can be with recruiter OR a team member — the key is it's an initial evaluation, not a decision-maker conversation.	Recruiter / Initial contact / Team member
Interview	Substantive evaluation with the decision maker or hiring manager. Purpose: assess skills and fit in depth. If the person can make or strongly influence the hire decision = interview.	Hiring manager / MD / Decision maker
Offer	Formal or informal offer made	—
Accepted	User accepted the offer	—
Rejected	Final rejection at any stage (by CV, after screening, after interview). Can coexist with Replied.	—

How to tell Screening from Interview

Ask: can this person make the decision?

Situation	Stage
Call with recruiter / HR about your background	Screening
Call with team member who says "I'll report back to [decision maker]"	Screening
Exploratory call where contact says "let me check with [MD/lead]"	Screening
Meeting with hiring manager who makes the final call	Interview
Meeting with MD / founder / director who decides	Interview
Task / case study given after screening	Interview (part of it)

Pipeline Table = Conversion Funnel

The Pipeline table in applications_index.md tracks how many applications ever reached each stage (cumulative), not just current status. This is what gives meaningful conversion rates.

Current status is tracked in the Active Applications table per app.

Example: An app that went Sent → Replied → Screening → Rejected counts in ALL stages it passed through:

• Sent: +1
• Replied: +1
• Screening: +1
• Rejected: +1

Example: An app rejected by CV without conversation only counts:

• Sent: +1
• Rejected: +1

---

Directory Structure

applications/
├── README.md                              ← what this directory is
└── my_data/
    ├── README.md                          ← how to add applications
    ├── applications_index.md              ← master tracking file (all apps)
    └── app_[company_slug]_[YYYY-MM-DD]/   ← one folder per application
        ├── cv_sent.md                     ← exact CV text sent (with diff notes)
        ├── cover_email.md                 ← (optional) email/cover letter text
        └── [user files]                   ← (optional) PDFs the user saved here

Naming Convention

• Folder: app_[company_slug]_[date_sent] (all lowercase)
• Examples: app_sana_engagement_2026-04-30, app_eea_volunteer_2026-05-01
• Slug by company name, not role type

Files Per Application

Required:

• cv_sent.md — the exact CV text. Always includes:
  • Header: Sent date, Channel, CV source, whether user-modified
  • Full CV text as sent
  • "AI Notes — Diff" section at the bottom (what changed from role version)

Optional:

• cover_email.md — cover email or letter text
• notes.md — response tracking, follow-up timeline
• User-placed files (PDFs, screenshots) — don't touch these

---

applications_index.md Format

Text-based with summary table for active applications, detailed entries below.

# Applications Index

Tracking all job and volunteer applications. Each entry links to a snapshot of the exact CV sent.

***

## Pipeline

**Total sent: N** · First: YYYY-MM-DD · Last: YYYY-MM-DD

| Stage     | Count | Rate                          |
| --------- | ----- | ----------------------------- |
| Sent      | N     | —                             |
| Replied   | N     | X%                            |
| Screening | N     | X%                            |
| Interview | N     | X%                            |
| Offer     | N     | X%                            |
| Accepted  | N     | —                             |
| Rejected  | N     | Company A (reason) · Company B |

**Funnel:** Sent → Replied → Screening → Interview → Offer → Accepted / Rejected / Ghosted

***

## Active Applications

| Company | Role | Channel | Prediction | Status | Updated |
|---------|------|---------|------------|--------|---------|
| **Company** | Role Title | Channel | 🔥🔥🔥 / 🔥🔥 XX-XX% | ✅/🟡 | YYYY-MM-DD |

**N active** · **N replied** (X%) · **N pending**

***

## Channel Performance

| Channel | Sent | Replied | Rate | Applications |
|---------|------|---------|------|-------------|
| **Personal referral** | N | N | X% | Company ✅ · Company ❌ |
| **Cold email** | N | N | X% | Company ✅ |
| **Company website** | N | N | X% | Company ✅ · Company ❌ |
| **Referral link** | N | N | X% | Company ❌ |

**Insight:** [Pattern observed across channels]

***

## All Applications

*Newest on top.*

### N. Company Name — Role Title
* **Date:** YYYY-MM-DD · **Channel:** Channel · **Status:** ✅/🟡/❌ Status
* **CV:** role_[name]/cv.md (snapshot/version)
* **CL:** [Brief description or "None"]
* **Prediction:** 🔥🔥 XX-XX% [→ ✅ accurate / → ❌ overestimated]
* **Notes:** Brief context
* **Folder:** `app_[company]_[date]/`

---

Workflow

Step 1: User says they applied

With a target role (has JD):

1. Check if role folder exists in target_roles/my_data/role_[name]/
2. Copy cv.md from role folder → app_[company]_[date]/cv_sent.md
3. Add header with: Sent date, Channel, CV source, "Snapshot — awaiting final text"
4. Add entry to applications_index.md (newest on top, increment number)
5. Add Reply prediction and Prediction rationale (see rules below)
6. Update conversion funnel counts
7. Update target_roles/my_data/roles_index.md:
   • Update Market Signals counts (total applied, replied, pending, etc.)
   • Add new row to Role Type Fit Matrix
   • Update Positioning Insights if this application provides new signal
   • Update Strategy section if needed (e.g., new role type to pursue/test)
   • Update "Last Updated" timestamp

Without a target role (volunteer, networking, spontaneous):

1. Create custom CV directly in app_[company]_[date]/cv_sent.md
2. Add header and entry to index
3. Add Reply prediction and Prediction rationale (see rules below)

Step 2: User sends final text

1. Compare user's text with the snapshot in cv_sent.md
2. Update cv_sent.md with the actual final text
3. Add "AI Notes — Diff" section documenting:
   • What the user changed from the role version (or generated version)
   • New details not in story files (flag these!)
   • Formatting changes
   • Sections added/removed
4. Update the index entry if needed

Step 3: Status updates

1. User reports a response or status change
2. Update the application's Status and Last update in index
3. When outcome is known (Replied / Rejected / etc.), note in rationale whether prediction was accurate
4. Update Active Applications Summary table (add/remove rows, update counts, adjust predictions)
5. Recalculate conversion funnel
6. Update target_roles/my_data/roles_index.md:
   • Update Market Signals counts (replied/rejected/pending)
   • Update Role Type Fit Matrix row (Result column)
   • Update Positioning Insights if this outcome provides new evidence
   • Update Market Signals → What's working / What's not working if patterns emerge
   • Update Strategy section if this outcome changes approach
   • Update "Last Updated" timestamp
7. If useful, create notes.md in the app folder with timeline

Step 4: Periodic review

When asked (or proactively):

• Count applications by status
• Calculate conversion rates
• Identify patterns (which CV versions get responses?)
• Review prediction accuracy — compare predictions to actual outcomes, identify calibration errors

---

Diff Notes Guidelines

The "AI Notes — Diff" section in cv_sent.md is critical for learning. Track:

What to document:

• Contact info changes (placeholder → real)
• Sections added/removed (Certifications, Education, etc.)
• Bullet changes (removed X bullets, added Y)
• Date fixes (e.g., "Nov 2020" → "Oct 2020")
• Company name format changes
• New details not in story files (flag with ⚠️ — may need to be added to stories)

What NOT to document:

• Minor formatting (bullet style * vs -)
• Trivial whitespace

Flag new facts: If user added details not in story files, note them explicitly:

"VWFS: 'voted Best Employee 2018' — NOT in story files. Consider adding to story_02."

This helps decide whether to update source stories later.

---

Conversion Funnel

Recalculate when any status changes:

Response rate = Replied / Sent
Screening rate = Screening / Sent
Interview rate = Interview / Sent
Offer rate = Offer / Sent
Close rate = Accepted / Offer

Also track by channel (referral vs website vs email) and by CV type (role-based vs custom).

---

Reply Prediction Rules

Every application must have a Reply prediction and Prediction rationale.

Rating Scale

🔥🔥🔥	🔥🔥	🔥
High (60-70%+)	Moderate (30-50%)	Low (<25%)

Add percentage ranges to all predictions (e.g., "🔥🔥 Moderate (40-50%)").

Rating Factors

Consider when predicting:

Factor	Weight	Examples
Channel	High	Referral (+20-30%) vs Website/Cold (+0-10%)
Fit	High	Direct domain match (+20%) vs Stretch (-10%)
Competition	Medium	Well-known brand = more applicants = lower odds
Title match	Medium	Exact title (+10%) vs Transferable skills (0%)
Experience level	Medium	Meets years requirement (+10%) vs Below (−15%)
Gaps	High	Critical gap (e.g., required certification missing) = −20%

Channel multipliers:

• Personal referral (someone recommends you directly): +25-30%
• Referral link (employee gives you application link): +15-20%
• They ask for CV (company solicits): +20-25%
• Cold email that gets response: +10-15%
• Website application: baseline (0%)
• LinkedIn Easy Apply: often filtered = −5%

Examples:

Scenario	Prediction
Referral + perfect fit	🔥🔥🔥 High (60-70%)
Referral + moderate fit	🔥🔥 Moderate (40-50%)
Website + perfect fit	🔥🔥 Moderate (25-35%)
Website + stretch fit	🔥 Low (10-20%)
Cold email + domain perfect	🔥🔥 Moderate (35-45%)

Rationale Format

Always include:

• ✅ Strengths: What makes this application strong
• ❌ Weaknesses: What reduces chances
• 🎯 Key factors: Why this rating

Example:

* **Reply prediction:** 🔥🔥 Moderate (30-40%)
* **Prediction rationale:** Referral (link, not personal) — CV reaches human, not ATS.
  Cross-functional coordination + team growth + process improvement — good match.
  Cons: no developer tools experience, career gap since 2024, "7+ years PM" counts legal work.
  High competition (JetBrains brand).

For Retroactive Predictions (rejected applications)

When adding predictions to already-rejected applications:

• Add — *retroactive* to indicate this wasn't made before outcome
• Note whether rejection was predictable or unexpected
• Example: 🔥 Low (10-15%) — *retroactive*

Calibration Over Time

Track prediction accuracy:

• When outcome is known, compare to prediction
• If consistently over-optimistic: lower baseline ratings
• If consistently under-optimistic: raise baseline ratings
• Channel-specific patterns may emerge (e.g., referrals work better than expected)

Review quarterly to identify patterns and adjust prediction logic.

---

Integration With Other Skills

• job-analysis: Creates role folders with JDs → these generate CVs
• starr-achievements: Source of truth for CV content → diff notes may reveal missing story details
• company-profiles: Provide context for cover letters

Key relationship: target_roles generates CVs → applications snapshots them. The two directories must not be confused.