| name | bd-radar-protocol |
| description | Load when running an Opportunity Radar sweep — the memory worklist layout, the six-step sweep protocol, monitoring mode, dedup rules, CRM push rules, and the sweep report format. |
Opportunity Radar — sweep protocol
Memory layout — /mnt/memory/bd-targets/ (read FIRST, write LAST)
The store is your shared brain across runs:
_done.md — sponsors already mapped. NEVER re-map these._frontier.md — sponsors discovered but NOT yet mapped. Your worklist (FIFO)._learnings.md — distilled sourcing lessons from past runs (dead-end sponsor
shapes, angles that produced Strong targets, untapped angles). May not exist yet.<company-slug>.md — one qualifying card per target company.<sponsor-slug>.sponsor.md — one short cluster overview per mapped sponsor.
Each sweep — run this exactly
- Read
_done.md, _frontier.md, and _learnings.md (if present — apply its
lessons to this sweep's choices); skim existing card filenames so you know
what's known.
- Refill the frontier if thin (fewer than ~3 entries): discover NEW sponsors via
several angles, preferring sponsors not already connected to mapped ones (so
coverage reaches new regions, not just neighbors):
- co-investors / competitors of sponsors already in
_done.md,
- cert / registry directories for the ICP's regulated posture (the
certifications named in the brand canon) → find certified operators → who owns them,
- recent recap / M&A news in the ICP's vertical (last ~6 months),
- PE firm lists focused on the ICP's vertical.
- Pop the next 2–3 sponsors from the top of
_frontier.md and delegate each to
your Sponsor Mapper sub-agent — one thread per sponsor, launched in parallel. Each
delegation prompt: the sponsor name plus whatever memory already knows about it (a
co-investor edge, a portco already carded). Do step 2's frontier refill yourself
while the mappers work.
When each map returns (prose + a sponsor_map JSON block):
- Sanity-check anything that looks off — you own scoring judgment, the mapper is
research help. Spot-verify a claim before carding it if in doubt.
- Write a short
<company-slug>.md card for each qualified portco (Strong /
Worth a look). Record the map's RED/skip companies too (so they aren't
re-surfaced) with one line.
- Write/update a
<sponsor-slug>.sponsor.md overview from the map.
- Queue the map's
discovered_sponsors for step 4.
- Extend the frontier: append newly-discovered peer/co-sponsors to the bottom of
_frontier.md.
- Close the loop: move the mapped sponsor(s) from
_frontier.md into _done.md
with today's date (run date for it).
- Report (format below).
Monitoring mode — when the frontier drains AND the core space is mapped: scan
trigger feeds (recaps in the last ~30 days, new RFPs, new cert holders) for genuinely
NEW nodes; add only nodes not already in memory. Say you're in monitoring mode and how
many new nodes you found (often zero — that's fine).
Dedup rule
Never re-surface a company already carded or a sponsor in _done.md, unless there's a
material new signal — then mark the card "UPDATE" and say why.
Push qualified targets to the CRM
After you write a NEW target's memory card, if its fit is Strong or
Worth a look, add it with the crm_add_target tool. Do NOT add RED/skip or
out-of-ICP nodes — those stay in memory for dedup only.
- Pass everything you qualified:
company (required), sponsor, hq, vertical,
fit, green_signals, why_now, entry_persona, sources — and always
manual: false (you are an agent).
fit must be exactly "Strong" or "Worth a look".- The CRM upserts by company — re-adding an existing company won't duplicate or reset a
human-set status. A successful call returns
{"ok":true, ...}.
crm_list_targets shows the pipeline if you need to cross-check beyond your memory.
Sweep report format
- One-line tally: "Mapped sponsor(s): X. N net-new targets carded. Frontier now: M sponsors."
- Why the sweep stopped — one line, every sweep: "budget — mapped the per-sweep
2–3, frontier still has M queued" or "frontier drained → monitoring mode (N new
nodes)" or "blocked: ". Never let the report imply the territory
is exhausted when the sweep just hit its budget.
- Ranked (best-fit first) net-new targets: Company + sponsor + HQ · Fit
(+ which GREEN signals) · Why now (trigger) · Entry persona · Sources (URLs).
- The current
_frontier.md contents (so the human sees what's queued next).
- Which files you wrote/updated, and how many targets you pushed to the CRM.
