memopop-orchestrator/issue-resolution
- Path
- issue-resolution/index.md
Issue Resolution
- Agent Should Reorder and Organize Citations on Assembly - Specification for globally renumbering and consolidating citations during final memo assembly to eliminate duplicates and gaps.
- Archive Dating and Time-Series Defect Hitlist - Seventeen defects found by the first live run of the time-series analyst, ordered so the ones that corrupt the archive’s chronology get fixed before the ones that only corrupt a grid. The dating defects come first because a wrong date is written into a filename, looks exactly like a right one, and silently re-anchors every analysis that reads it afterward.
- Competitive Research Generated But Not In Prose - The evaluator names 9+ competitors to disk; the writer never reads that file and emits
<needs-source>asking for the names two dirs away. - Containerizing Risk Assessments and Diligence Skepticism - Specification for separating risk analysis and skeptical commentary from the main memo narrative to preserve business utility for syndication and LP review.
- Correct Citation Pipeline Accuracy in Multi-Agent Research - High-priority issue resolution for citations flowing through the system in a lossy manner, resulting in inline refs pointing to nothing in final memos.
- Deck Data Flow Gap Causes Hallucination - Critical issue where accurate deck data is lost between agents, causing the writer to hallucinate from failed web research instead of using extracted deck values.
- Duplicate Table of Contents Generation - Diagnosing and preventing duplicate TOC blocks from appearing in final draft memos during export and regeneration cycles.
- Faked Sources from Perplexity - Documentation of hallucinated URLs produced by Perplexity Sonar Pro during citation enrichment, with patterns and mitigation strategies.
- Fix Firm-Scoped Output Paths - Bug fix specification for artifacts saving to wrong directory when using –firm flag instead of firm-scoped io/ paths.
- Getting AI to Refocus When Web Research is Empty - Behavioral rules for justify mode when web research returns thin or noisy results, ensuring agents fall back to internal context rather than blocking.
- Getting Sonar Pro to Work in First Research Agent - Issue resolution for integrating Perplexity Sonar Pro citations into the initial research and section writing passes rather than as post-processing.
- Improve Investment Memo Output - Improve the quality and depth of investment memos generated by the Investment Memo Orchestrator.
- Including Comparable Exits, Valuations, and IPOs - No agent harvests exit comps, so the Cash on Cash section’s bull/base/bear multiples are return math anchored to nothing verifiable.
- Limiting or Omitting Investor Judgement - The pipeline renders an unsolicited PASS / CONSIDER / COMMIT verdict even when the analyst’s job is advocacy, not adjudication.
- Papermark Self-Hosted Dataroom Deployment - Deployment specification for Papermark as a self-hosted virtual data room for sharing investment memos and due diligence documents.
- Perplexity Premium API Calls - Complete Reference - Complete reference for using Perplexity @source syntax and premium data sources for high-quality investment research via API.
- Perplexity API Fix - Zero Balance Issue - Issue resolution for Perplexity API key zero balance blocker and the broader challenge of getting sonar-pro citations on first research run.
- Premium Data Sources Integration Plan - Plan for eliminating generic industry benchmark filler by integrating Perplexity premium data sources with @source syntax.
- Preventing Hallucinations in Memo Generation - Critical issue resolution for LLMs fabricating pricing, traction, and business model data for seed-stage startups with limited public information.
- Research Agent Concatenates Multiple Sources Under One Citation ID — And No Validator Catches It -
research_enhanced.pywrites several sources onto one[^N]:line, so downstream parsers keep the first and silently drop the rest. - Resume Workflow After Interruption - Specification for a checkpoint-based resume system to avoid restarting memo generation from scratch after API timeouts, crashes, or user interruptions.
- Section Citations Orphaned When Definitions Live in 1-research/ -
consolidate_citations.pyreads### Citationsonly inside section files, so defs living in1-research/become orphans. - Tier 1 Anti-Hallucination Test Results: Work Back AI - Test results validating Tier 1 anti-hallucination fixes against a seed-stage company with limited public data.
- Visual Collector Crop Boxes Under-Enclose Large Diagrams - The visual collectors reliably find and classify what is on a slide; the bounding boxes they draw around large diagrams and timelines sometimes stop short of the visual’s true edges. Deferred — the crop is preserved and reusable either way, and precision can be improved without re-deriving anything else.