Changelog - 2025-11-27 (#2)

Scorecard Agent & CLI Implementation

Overview

Implemented the Scorecard Agent and CLI tool to automatically generate Hypernova Capital's 12-dimension scorecards for emerging manager fund commitment memos. This completes the scorecard system introduced in changelog 2025-11-27_01.

New Feature: Scorecard Agent

Purpose

Generate structured scorecards for fund commitment memos using Hypernova's 12-dimension emerging manager evaluation framework. The agent produces three horizontal markdown tables (one per dimension group) with scores, percentile labels, and rationales grounded in memo artifacts.

Implementation

Created src/agents/scorecard_agent.py (~210 lines):

def scorecard_agent(state: MemoState) -> Dict[str, Any]:
    """Generate an emerging manager scorecard for fund commitment memos."""

def _load_scorecard_template() -> Dict[str, Any]:
    """Load the YAML scorecard template."""

def _load_section_snippets(output_dir: Path) -> Dict[str, str]:
    """Load key memo sections as additional context for scoring."""

def _build_scorecard_prompt(...) -> str:
    """Build the LLM prompt with context and scoring instructions."""

Gating Conditions

The scorecard only generates when:

• investment_type == "fund"

• outline_name == "lpcommit-emerging-manager"

Skips gracefully for direct investments or other outline types.

Context Sources

The agent loads multiple context sources for scoring:

1. state.json - Company metadata, memo mode, descriptions

2. 1-research.json - Web research data

3. 2-sections/ - Key section files:

   • 02-gp-background--credibility.md

   • 03-fund-strategy--thesis.md

   • 04-portfolio-construction.md

   • 05-track-record--analysis.md

New Feature: CLI Scorecard Generator

Purpose

Standalone CLI tool to generate scorecards for existing fund memos without regenerating the full pipeline.

Implementation

Created cli/generate_scorecard.py (~120 lines):

# Usage
python cli/generate_scorecard.py "FundName"
python cli/generate_scorecard.py "FundName" --version v0.0.5
python cli/generate_scorecard.py output/FundName-v0.0.1

Features

• Resolves artifact directory from company name or direct path

• Validates investment type and outline before proceeding

• Uses VersionManager for version resolution

• Rich console output with progress indicators

New Feature: Scorecard YAML Template

Purpose

Define the 12-dimension evaluation framework with scoring rubrics, evaluation guidance, and evidence sources in a structured YAML format.

Implementation

Created templates/scorecards/lp-commits_emerging-managers/hypernova-scorecard.yaml (~460 lines):

Structure

metadata:
  scorecard_id: "hypernova-emerging-manager-v1"
  name: "Emerging Manager Evaluation Framework"
  version: "1.0.0"
  firm: "Hypernova Capital"
  applicable_types: ["fund"]
  applicable_modes: ["consider", "justify"]

scoring:
  scale:
    min: 1
    max: 5
    labels:
      1: "Bottom quartile"
      5: "Exceptional"

  percentile_mapping:
    5: "Top 2-5%"
    4: "Top 10-25%"
    3: "Top 50%"
    2: "Bottom 50%"
    1: "Bottom 25%"

dimension_groups:
  - group_id: "founder_dynamics"
    name: "Founder & Deal Dynamics"
    dimensions: [empathy, theory_of_market, ecosystem_imprint, hustle]

  - group_id: "execution_style"
    name: "Execution & Investment Style"
    dimensions: [pull_force, pounce, integrity, grip]

  - group_id: "portfolio_management"
    name: "Portfolio Management & Discipline"
    dimensions: [judgement, discipline, positioning, prudence]

dimensions:
  empathy:
    name: "Empathy"
    short_description: "Insight-driven empathy with founders..."
    full_description: |
      The ongoing privilege to invest in competitive early-stage deals...
    evaluation_guidance:
      questions: [...]
      evidence_sources: [...]
      red_flags: [...]
    scoring_rubric:
      5: "Exceptional founder relationships..."
      1: "Transactional approach..."

Mode-Aware Scoring

The scorecard agent handles both memo modes with distinct instructions:

Consider Mode

• Objective assessment for potential LP commitment

• Acceptable to highlight missing data as risk

• Must ground judgments in provided context

• Recommendation options: PASS / CONSIDER / COMMIT

Justify Mode

• Assumes LP has already committed

• Uses existing memo sections as primary source of truth

• When data is limited, defaults to 3/5 with inference rationale

• Avoids generic "no information available" language

• Recommendation: COMMIT (with supporting rationale)

Scorecard Output Format

Three Horizontal Tables (4 dimensions each)

Group 1: Founder & Deal Dynamics

Group 2: Execution & Investment Style

Group 3: Portfolio Management & Discipline

Summary Section

• Overall assessment paragraph

• Bullet list: Standout strengths (scores of 5)

• Bullet list: Areas of concern (scores of 1-2)

• Final recommendation: PASS / CONSIDER / COMMIT with rationale

Files Changed Summary

New Files (3)

Usage

# Activate virtual environment
source .venv/bin/activate

# Generate scorecard by company name (uses latest version)
python cli/generate_scorecard.py "Avalanche"

# Generate scorecard for specific version
python cli/generate_scorecard.py "Avalanche" --version v0.0.3

# Generate scorecard by direct path
python cli/generate_scorecard.py output/Avalanche-v0.0.3

Example Output

╭──────────────────────────────────────────────────────────╮
│ Generating Scorecard for: Avalanche                      │
│ Artifact directory: output/Avalanche-v0.0.3              │
╰──────────────────────────────────────────────────────────╯
✓ Loaded state.json
Scorecard generated for Avalanche using Hypernova emerging manager framework
Saved to output/Avalanche-v0.0.3/scorecard.md
Scorecard file: output/Avalanche-v0.0.3/scorecard.md

Integration with Existing System

Artifact Trail

Scorecard output saved to existing artifact directory:

output/FundName-v0.0.x/
├── 0-deck-analysis.json + .md
├── 1-research.json + .md
├── 2-sections/*.md
├── 3-validation.json + .md
├── 4-final-draft.md
├── state.json
└── scorecard.md          # NEW

Agent Registration

Added to src/agents/__init__.py:

from .scorecard_agent import scorecard_agent

Checklist Complete

From changelog 2025-11-27_01 "Next Steps":

• Implement Scorecard Agent to auto-score dimensions from memo content

• Add scorecard table insertion to memo sections (future)

• Create direct investment scorecard template (future)

Contributors

Implementation by Claude Code (Anthropic) with user guidance.