Pineal Capital
Mindful Investing · Fund I

Pineal Capital — FPR Module Build Dashboard

GCCP × Pineal Capital · Financial Performance & Returns Module · v1.3.1 · 16 Apr 2026
AI-Draft Plan Mode

FPR Module — Architecture Evolution

Financial Performance & Returns | Pineal Capital × GCCP Build
16 April 2026 · [AI-DRAFT] · Methodology hash reference: FPR_v1.0-a3f2 → FPR_v1.1 pending
CONOR v0
21 flat modules
→
CURRENT v1.1
8 chained sub-agents
→
TARGET
Pineal plugin + Tier-1 data
ORIGIN — AUTHOR: CONOR

v0 · 21-Module Flat Spec

Per-report-section module, linear execution, LLM-native

Module Set (≈21, flat)

  1. Executive Summary
  2. Historical Performance
  3. Earnings Quality
  4. Margins
  5. Cash Flow
  6. Returns on Capital
  7. Balance Sheet
  8. Capital Structure
  9. Dilution
  10. Working Capital
  11. Leverage
  12. Capex
  13. Peer Comparison
  14. TSR
  15. Consensus Outlook
  16. Variances
  17. Risks
  18. Investment Implications
  19. Final Assessment
  20. Data Ingest (implicit)
  21. Audit / Source Log (implicit)

Characteristics

  • One module per report chapter — flat, not chained
  • LLM-primary — computations inside prose
  • No explicit data-tier declaration
  • No sector adapter framework
  • No methodology version hashing
  • No forecast vs consensus delta as mandatory output
  • No computational protocol — math done in natural language

Strengths

  • Comprehensive coverage of IC-memo sections
  • Maps 1:1 to sell-side report structure
  • Easy to draft quickly with an LLM

Weaknesses (cross-model critique)

  • Hallucination risk — math in prose
  • No reproducibility guarantee
  • No adapter routing → sector errors silent
  • No upstream/downstream contract
  • Too wide for single 25-min Cowork run
CURRENT — v1.1 (AARON + CROSS-MODEL)

8 Chained Sub-Agents

Schema-contracted · Python-computed · Reproducible

Sub-Agent Chain

  1. #1 Data Ingest — raw_financials.json, source_log.csv
  2. #2 Reconstruction + Adjustments UPD balance gate, 4-way SBC, normalised buckets
  3. #3 Returns + DuPont + Balance Sheet UPD D&A proxy, regime tagging
  4. #3B WACC Computation NEW FRED + Damodaran inputs
  5. #4 Sector Adapter + Peers UPD sector-aware scoring weights
  6. #5 Forecast + Variant View UPD falsifiable thesis requirement
  7. #6 Scoring + Red Flags + Report UPD decomposed sub-scores
  8. #7 Change Detection NEW delta vs prior run + headline

Core Disciplines (retained + hardened)

  • Rule #1 — Never fabricate; null + confidence score
  • Rule #4 — Sector adapter declared at start
  • Rule #5 — Forecast vs consensus = alpha signal
  • Rule #7 — Methodology hash + reproducibility
  • Rule #11 (NEW) — Computational protocol: Python computes, LLM only interprets
  • Rule E5 — DISPUTED halt on source disagreement
  • GCCP output standard — Blackstone/Goldman

Proven in Dry-Run (MSFT, 16 Apr)

  • Null-flagging worked — refused to fabricate
  • DERIVED + DISPUTED tags fired correctly
  • Coverage 16.3% on Tier 4 (WebSearch only)
  • Sandbox egress = hard constraint flagged
  • SECTOR_ADAPTER_GAP raised on MSFT (no tech_saas adapter yet)

Gains vs v0

  • Reproducible (methodology hash)
  • No math hallucination (Python)
  • Sector-routed scoring
  • Change detection as run-over-run signal

Residual Gaps

  • Test ticker was US (MSFT), universe is EU mid-cap
  • No IFRS / URD / German AR adapters yet
  • Data tier unresolved (Tier 4 → 2/3 pending)
  • Python engine not yet wired to Pineal orchestrator
  • No shared git branch with Conor yet
TARGET — PRODUCTION

Pineal-Integrated Plugin

Skill #10 · Bloomberg Tier 1 · Cowork-distributed

Stack Shape

  • FPR Skill #10 inside .claude/skills/fpr-module/
  • Python sub-agents registered with existing agents/orchestrator.py — not parallel
  • Cowork plugin bundle (skills + MCPs + tools) — how Pineal installs it
  • GitHub repo with CI/CD — PR-gated shared branch (no ZIP-over-email)
  • Cloudflare portal with Zero Trust — Pineal sign-off delivery

Upstream / Downstream Contracts

  • Feeds thesis-builder Stage 2 (operational hurdles)
  • Feeds valuation-screener (segment profitability + comps)
  • Feeds ic-paper §3 Business Overview + §5 Valuation
  • Feeds risk-monitor (leverage, coverage, liquidity)
  • #7 Change Detection → catalyst-tracker

Data Tier Evolution

  • Phase 1: Koyfin + FMP (~€100/mo) — Aaron builds & tests
  • Phase 2: Conor parity-run on Pineal Cowork, same tier
  • Phase 3: Bloomberg BLPAPI adapter — Pineal engineering scope, weeks not days
  • Phase 4: Overnight batch on 8–12 live names

Additions Beyond v1.1

  • Item #13 — EU filing regime NEW IFRS / URD / German AR / Reuters ticker suffixes
  • Sector adapter priority re-ordered for EU mid-cap universe
  • [AI-DRAFT][TEST/DEMO] tagging inherited from Pineal
  • Regulatory fit — SFDR Article 6, QIAIF, CBI (not FCA)
  • Compliance-checker handoff before any live output

Why This Shape

  • Lives where Pineal already works (Claude / Cowork)
  • Plugin = versionable delivery, not a folder bounce
  • Bloomberg cost stays with Pineal
  • GCCP keeps the methodology IP

Risks to Manage

  • Data-tier asymmetry during build/run split
  • Schema drift if Conor forks v1.2-conor
  • Bloomberg MCP is greenfield engineering
  • Persona-panel outputs (existing) must not pollute FPR scores

Migration Path — How We Get From Current to Target

Weeks 1–3 · Aaron builds

  • Read agents/orchestrator.py before writing FPR Python
  • Draft FPR Skill #10 SKILL.md matching Pineal's anti-hallucination pattern
  • Subscribe Koyfin + FMP (Tier 2/3)
  • Run full 8-agent chain on 3 EU mid-caps from Pineal pipeline
  • Package as .plugin bundle

Week 4 · Conor parity-runs

  • Pineal installs plugin on their Cowork
  • Runs same 3 tickers, same data tier
  • Methodology hash parity check
  • Conor signs off on schema contracts with thesis-builder & ic-paper
  • Merge agreed version to main on shared repo

Weeks 5–9 · Bloomberg uplift (Pineal)

  • Bloomberg BLPAPI → FPR Tier 1 adapter
  • Sector adapter library expanded for EU universe
  • Reference portfolio built for score calibration (median = 50)
  • Overnight batch scheduler stood up
  • Cloudflare portal wired for delivery
Critical decisions still open: (1) Pineal universe size — default 8–12?   (2) Output cadence — weekly / monthly / on-demand?   (3) Who owns the commercial terms — licensing model GCCP ↔ Pineal?   (4) Bloomberg seat — Pineal-provided, or shared?   (5) Conor alignment call on v1.1 architecture — schedule before Week 1 kick-off.