deal-flow

<!-- autoresearch: variation B — sharper output via leverage scoring, valuation context, and banned-phrase quality gates --> Read `memory/MEMORY.md` for context (tracked verticals, active theses). Read the last 14 days of `memory/logs/` and extract company names from prior `### deal-flow` entries to build the dedup set. ## Goal Produce a weekly funding-round digest that tells the reader **what the capital reveals**, not just what closed. A flat list of rounds is noise; a ranked, contextualized read is signal. ## Steps ### 1. Gather candidates from tiered sources Run searches in parallel. Use the **current month + year** in queries. Aim for ≥25 candidate deals before filtering. If a source fails, record `source=fail` for the footer; do not abort. **Source fetch rule (applies to every WebFetch below):** if the fetch returns 404, 403, is rate-limited, or times out, log `DEAL_FLOW_SOURCE_MISS: <url> (<reason>)` and continue with remaining sources. Never abort the run on a single source miss — the tiered design is specifically so that any one source can fail. **Tier 1 — broad aggregators (always run):** - WebSearch: `"biggest funding rounds" week ${month} ${year} site:news.crunchbase.com` - WebSearch: `funding round ${month} ${year} site:techcrunch.com` - WebSearch: `"Series A" OR "Series B" OR "seed" raised ${month} ${year} startup` - WebFetch `https://news.crunchbase.com/sections/venture/` — extract this week's roundup if present (apply source-miss rule above) **Tier 2 ...