Changelog - 2026-03-10_02

Competitive Landscape Research & Evaluation System (Initial Implementation)

Two new agents that discover, evaluate, and classify competitors using Perplexity Sonar Pro. The system splits competitive analysis into two phases with opposing cognitive biases: the researcher uses completion bias to find as many candidates as possible, while the evaluator uses skepticism bias to challenge whether each candidate is actually a competitor.

Design spec: context-v/Introducing-a-Competitive-Landscape-Research-and-Evaluation-System.md

Architecture

Two-Agent Design

Source Priority

Both agents follow the same hierarchy:

1. Dataroom — highest trust, pre-seeded into candidates

2. Deck — competitive context from the pitch deck

3. User-provided — known_competitors and search_variants from company JSON

4. Web research — Perplexity Sonar Pro queries

Four-Criteria Evaluation

Classification Rules

• direct_competitor: High on 3-4 criteria

• indirect_competitor: High on 2 criteria

• adjacent: High on 1 criterion

• not_a_competitor: High on 0 — removed from results

Changes

New File: src/agents/competitive_landscape_researcher.py

~480 lines. Discovers candidate competitors through multi-query web research.

• _extract_market_category(): Pulls product/market context from deck analysis and research state

• _get_deck_competitors(): Extracts competitor mentions from pitch deck competitive_landscape field

• _get_dataroom_competitors(): Extracts structured competitor data from dataroom analysis

• _generate_queries(): Uses Perplexity to generate 5 targeted search queries, incorporating user-provided search_variants; falls back to template queries on failure

• _search_for_competitors(): Executes a single query, extracts structured competitor data (name, description, website, funding, online presence, etc.)

• _deduplicate_candidates(): Merges candidates from multiple sources by name, combining source queries and URLs

• _save_research_artifact(): Saves 1-competitive-research.md (human-readable) and 1-competitive-research.json (structured)

• competitive_landscape_researcher(): Main agent function — orchestrates all phases

New File: src/agents/competitive_landscape_evaluator.py

~560 lines. Evaluates and classifies candidates, runs gap analysis.

• _evaluate_candidates(): Sends all candidates to Perplexity for skeptical evaluation against four criteria; returns evaluations + identified gaps + gap queries

• _search_for_gaps(): Runs up to 2 additional queries to find competitors missed in initial research; filters out already-known companies

• _merge_evaluation_with_candidate(): Combines evaluation results (classification, reasoning, boolean criteria) with original candidate data (funding, website, online presence)

• _save_evaluation_artifact(): Saves 1-competitive-evaluation.md with tables (direct, indirect, adjacent, removed) and 1-competitive-evaluation.json

• competitive_landscape_evaluator(): Main agent function — orchestrates evaluation + gap analysis, determines confidence level (high/medium/low based on direct competitor count)

Modified: src/state.py

New TypedDicts for the competitive landscape pipeline:

• CompetitorCandidate: Raw candidate with name, description, website, funding details, online presence dict, source tracking, and origin flags (from_deck, from_dataroom)

• CompetitiveLandscapeResearch: Researcher output — candidates list, queries executed, totals

• EvaluatedCompetitor: Post-evaluation competitor with classification, reasoning, four boolean criteria, and complete investor list

• CompetitiveLandscapeEvaluation: Evaluator output — evaluated list, direct/indirect name lists, removed list, gap additions, summary, confidence

New MemoState fields:

• search_variants: Optional[List[str]] — user-provided search terms for competitive research

• known_competitors: Optional[List[str]] — user-provided known competitor names

• competitive_candidates: Optional[CompetitiveLandscapeResearch] — researcher output

• competitive_landscape: Optional[CompetitiveLandscapeEvaluation] — evaluator output

Updated create_initial_state() to accept and initialize all new fields.

Modified: src/workflow.py

• Imported both competitive agents

• Added competitive_researcher and competitive_evaluator nodes

• Updated edge sequence:

  Before: section_research → cite
  After:  section_research → competitive_researcher → competitive_evaluator → cite

• Updated generate_memo() to accept search_variants and known_competitors parameters

Modified: src/main.py

• Loads search_variants and known_competitors from company JSON config

• Prints them during startup if present

• Passes them through to generate_memo()

Artifacts Produced

Each run creates two new artifact files in the output directory:

Company JSON Config

New optional fields in data/{Company}.json:

{
  "search_variants": ["enzyme supplements", "metabolic health startups"],
  "known_competitors": ["Company A", "Company B"]
}

• search_variants: Incorporated into Perplexity query generation for more targeted results

• known_competitors: Pre-seeded into candidate list with source_queries: ["user_provided"]

Pipeline Position

Dataroom → Deck → Research → Section Research → Competitive Researcher →
Competitive Evaluator → [CITE on 1-research/] → [GATE 1] → Writer → ...

Both agents run in the research phase, after section research generates per-section research files, and before citation enrichment. The writer receives the evaluated competitive landscape data in state and can use it directly.

Status

Initial implementation — not yet tested with a live run. All imports verified, workflow builds with 28 nodes. Awaiting end-to-end test before promoting to feat(agent).