RealtyPad — triage

Hard rules

• Prefer evidence over Economics alone — Economics with estimated tax or capped rent lie.
• Scores: prefer score_summary. Economics / Data confidence / Growth / Overall — see read_doc("agents/glossary.md"). Never call Overall “score.” min_score filters Economics only.
• Read score_summary and data_gaps on list_deals / get_deal (core) first.
• data_gaps: list of { code, blocking, detail }. Any blocking: true gap means do not advance research → score/UW / claim ready. Soft gaps (tax_estimated, missing_growth_drivers, etc.) can proceed with eyes open. Codes: photos, rent, adr, arv, unscored (investment / either only), housing_cost_missing (primary only — no ask/purchase or no piti_v1 housing cost; primary deals keep score null by design), tax_implausible (blocking: stated tax_annual under $100/yr or under 0.1% of purchase — a placeholder like $1; scoring uses the workspace tax estimate until a real tax bill is entered), tax_estimated, insurance_estimated, vacancy_estimated, hoa_estimated (insurance/vacancy/HOA still equal to an automation market-trend fill — cleared when a verified value replaces it), rent_capped, repairs_unknown, missing_growth_drivers. Strategy money gaps (rent/adr/arv) cover strategy_intent when set (union), else the single income_strategy.
• Early pass (trash): when Economics is already weak (≲ ~40 / fails the strategy economic bar on known inputs) and Overall is negative or Growth is clearly negative (≲ −30) with no distress/buyer override — set passed without a full research deep-dive. Still skim listing + comments; do not invent numbers to “save” it.
• Prioritize research on deals with blocking data_gaps, decent Economics but low Data confidence (confidence_gaps / soft gaps), or missing Growth income series (growth_gaps). Prefer has_blocking_gap=true / has_gap=<code>. Soft gaps queue: has_soft_gap=true (deprecated needs_input still aliases).
• Fill data gaps with get_agent_manual(workflow=research) before judging keepers.
• For regional/cluster passes, refresh stale markets once via get_agent_manual(workflow=trends) (dedupe deals; ensure_geo_markets + refresh_deal_trends(gap_fill=true) per deal when needed — not agent CSV → upsert; surgical refresh only when needed).
• For verdict + status on individual deals, follow get_agent_manual(workflow=underwrite) (which calls get_agent_manual(workflow=scenarios)).
• For scenario-only / sensitivity / bid variations without a full verdict, follow get_agent_manual(workflow=scenarios).
• Never invent numbers. Pattern-pass only when the failure mode is already documented.
• Read comments + open action_needed handoffs before flipping a deal that was previously researched.

Aggressiveness

Defaults by income_strategy (see AGENTS.md / get_agent_manual(workflow=underwrite)). Chat override: aggressiveness: aggressive|moderate|strict (or “be aggressive” / “be conservative”). strict (“be conservative”): LTR/STR fallback CF ≳ ~$200/mo (≈ Economics ~70) or a clearly stronger hold than moderate; flip ≳ $50k; wholesale ≳ $40k.

Strategy	Mode	Clears economic bar
ltr / str	moderate	Prefer hold projection (cashflow_projections via underwrite) over month-0 CF; fallback CF ≳ ~$100/mo only if projection unavailable
fix_flip / wholesale	aggressive	Clear asymmetric profit/spread only (flip ≳ ~$40–50k / Economics ~80; wholesale ≳ ~$25–40k / Economics ~70)

Pursue vs pass: economic bar is necessary but not sufficient. After a deal clears economics (and is not a structural pattern-pass), underwrite runs match_deal_buyers (live UW + selected scenario snapshots): matches → ranked; buyers scanned but none fit → passed; empty buyer book (scanned=0) → watch (Hold — no buyers yet; research done — keep an eye out). Hard external gaps → blocked. Human escalate ranked → pursuing. Full tree: get_agent_manual(workflow=underwrite) Status decision tree.

Queue modes

Match UI /triage stage tabs (labels: New, Blocking gaps, Soft gaps, …):

new (default inbox)

list_deals(status="new", limit=80)

Ingested, not started. Start research → researching, or pass / block.

Blocking gaps (UI mode id needs_research)

Any blocking data_gaps entry (photos / rent / ADR / ARV / unscored / housing cost):

list_deals(status="new,researching", has_blocking_gap=true, limit=80)
list_deals(has_gap="photos", limit=80)

Use has_blocking_gap=true (same predicate as Blocking gaps).

Soft gaps (UI mode id needs_verification)

Estimated tax/insurance, rent capped, repairs_unknown, listing off-market/auction, or missing listing contact/description:

list_deals(has_gap="tax_estimated", limit=80)
list_deals(has_soft_gap=true, limit=80)

has_soft_gap is the soft-gap OR convenience for this queue; prefer per-code has_gap when targeting one soft flag. Deprecated needs_input aliases has_soft_gap when unset.

blocked / watch / ranked / pursuing

list_deals(status="blocked", limit=80)
list_deals(status="watch", limit=80)
list_deals(status="ranked", limit=80)
list_deals(status="pursuing", limit=80)

• blocked: hard external stop — unblock when the fact clears.
• watch (UI: Hold — no buyers yet): economics already clear; empty buyer book or parked — not more research; resume when buyers appear. New/reopened/changed buyers auto-flag fitting Hold deals with an action_needed handoff (raw.tag=watch_resurface) — work those first: match_deal_buyers → ranked (see get_agent_manual(workflow=buyers)).
• ranked / pursuing: pursue pipeline.

Hide low data-confidence watch unless the user asks to show them (UI default).

regional / cluster

Filter by q, market_id, or shared builder/HOA/subdivision after loading a stage slice. Process the cluster as one thesis, not 50 isolated vibes.

Batch workflow

Triage Progress:
- [ ] 1. Pick mode + scope (N deals or geography/builder)
- [ ] 2. Load queue via list_deals; sample before full pass; read data_gaps on each item
- [ ] 3. Regional/cluster: dedupe market_ids → get_agent_manual(workflow=trends) for stale (~35d) or below the 120-month target (unless source is known shorter) home + LTR rent (STR if str deals; request up to 10y)
- [ ] 4. Detect pattern (see below) vs needs research / full UW
- [ ] 5a. Pattern-pass: `update_deal_status` (single) or `update_deals` **status+rationale only** (cluster — no money in the same call)
- [ ] 5b. Keepers with blocking data_gaps → get_agent_manual(workflow=research), then UW when clear
- [ ] 5c. Otherwise get_agent_manual(workflow=underwrite) (scenarios + buyer match)
- [ ] 6. Brief with canvas: counts, verdicts, remaining queue

Pattern-pass

When a whole cluster fails for the same documented reason (HOA tax bomb, builder premium, flood zone, etc.), batch passed with a shared short Status note (rationale) via status-only update_deals — do not invent per-deal numbers and do not mix money fields in that call. Single deal → update_deal_status.

MCP

Action	Tool
Queue	list_deals (has_blocking_gap, has_gap, Soft gaps via has_soft_gap, min_score, status, q, market_id; read score_summary + data_gaps; archived excluded by default)
Detail	get_deal core (score_summary + data_gaps + counts); view=overview/full or list_* for nests
Status / Status note	update_deal_status (single); update_deals status+rationale only for cluster pattern-pass
Money / archive	update_deal / update_deals (money-only; archived=true/false) — never mix with status
Buyer fit (via UW)	match_deal_buyers (snapshots on by default; returns scenario_run_id)
Comment	add_comment

Brief format

Use a Cursor canvas for batch results:

Address	Prior → verdict	Corrected CF	Buyers	Notes
…	new → passed	−$120/mo	—	tax $X
…	researching → ranked	+$180/mo	2 matches	linked Diego

Also give: how many cleared, how many left in queue, suggested next cluster.