Property Compare

Property Comparison

You compare up to 3 UK properties side by side and produce a ranked investment report. Each property gets the same data treatment — comps, EPC, rental yield, stamp duty — and the output is a structured table with a clear relative ranking and a recommended winner.

When to Use This Skill

• "Compare these three properties for me"
• "Which of these is the better investment?" + 2–3 addresses or URLs
• "Rank these properties" + 2–3 postcodes
• "Help me choose between these" + any mix of addresses, postcodes, Rightmove links

Input Formats Accepted

• Street address — e.g. "14 Elm Street, Nottingham, NG5 1AA"
• Postcode — e.g. "NG5 1AA"
• Rightmove URL — e.g. "https://www.rightmove.co.uk/properties/12345678"
• Mixed — each property can be a different input type; handle each appropriately

If asking prices are not given and no Rightmove URL is provided, use the area median from comps as a proxy price and flag this explicitly.

Query Lanes

Lane A — Full comparison (at least one street address or Rightmove URL per property): Tools per property: rightmove_listing (if URL) → property_comps + property_epc + rental_analysis → property_yield + stamp_duty Output: full 9-section comparison report.

Lane B — Postcode-only comparison (postcodes only, no street addresses): Tools per property: property_comps + rental_analysis → property_yield + stamp_duty Skip property_epc — no reliable individual cert match. Skip BTL scenario section. Output: area-level comparison; note that EPC data is unavailable.

Mixed — use Lane A rules for any property that has an address or URL; Lane B rules for postcode-only properties. Flag the asymmetry in the output.

Step-by-Step Workflow

Step 1: Parse inputs

Read what the user provided. For each property, determine:

• Do you have a street address? → Can call property_epc with address
• Do you have a Rightmove URL? → Must call rightmove_listing first
• Postcode only? → Skip EPC, use comps for area data

Extract postcode from any address if not explicitly given (infer from context or ask).

Step 2: Fetch Rightmove listings (if URLs provided)

Call rightmove_listing for each property that has a Rightmove URL. Run in parallel. This gives you asking price, floor area, tenure, council tax band, key features, and days-on-market signals before you proceed.

Step 3: Fetch comps, EPC, and rental data

For each property, run in parallel:

• property_comps — with address if available; always with postcode
• property_epc — with address (Lane A only); skip for postcode-only
• rental_analysis — with postcode + purchase_price set to asking price (or comp median if no asking price)

Deduplication: if two or more properties share the same postcode sector (e.g. both in NG5), the comps results will be identical. Fetch once, reuse for both. Flag this in the output.

Step 4: Fetch yield and stamp duty

For each property, run in parallel:

• property_yield — postcode; pass property_type filter if the property type is known
• stamp_duty — asking price (or comp median); calculate both primary-residence and additional-property scenarios

Step 5: Consolidate raw data

Save all tool outputs to /tmp/prop-compare-raw.json using the consolidation script:

python3 BASE_DIR/scripts/consolidate.py \
  --props '{"p1": {...}, "p2": {...}, "p3": {...}}' \
  --out /tmp/prop-compare-raw.json

Where each prop dict contains the raw JSON from all tools called for that property.

Step 6: Compute comparison metrics

Run the comparison script to derive relative metrics:

python3 BASE_DIR/scripts/compute_comparison.py \
  --raw /tmp/prop-compare-raw.json \
  --criteria BASE_DIR/assets/underwriting-defaults.json \
  --out /tmp/prop-compare-metrics.json

This computes per-property: price premium/discount, gross/net yield, underwriting signal scores, and the overall relative ranking.

Step 7: Render output

python3 BASE_DIR/scripts/render_comparison.py \
  --metrics /tmp/prop-compare-metrics.json \
  --template BASE_DIR/assets/compare-template.md \
  --out /tmp/prop-compare.md

Read /tmp/prop-compare.md and present to the user.

Output Sections

Present the report in this order. Use the assets/compare-template.md as the structural guide.

1. Header Property labels (P1 / P2 / P3), addresses, asking prices, date.

2. At a Glance 3-column summary table: asking price, bedrooms, EPC rating, gross yield, ranking position.

3. Price vs Market Per property: asking price vs local comp median, premium/discount %, price/sqft vs area median/sqft. Flag any property priced more than 10% above comp median.

4. Rental Market Per property: median rent, listing count, rent range. Note thin markets (< 5 listings). Segment student/professional lets if mixed. Normalise all rents to monthly before any calculation.

5. Yield Per property: gross yield %, net yield % (30% cost deduction), property_yield tool figure, data confidence (listing count). Show divergence between own-calc and tool figure if > 15%.

6. EPC Per property: current rating + score, potential rating, floor area, MEES risk flag.

• F or G rating → "Cannot be let in current condition (MEES)"
• D or E rating → "Below 2028 target — budget for improvements"
• C or above → "Lettable and compliant" Flag any floor area mismatch between EPC cert and listing (> 10%).

7. Stamp Duty Per property: SDLT for primary residence and additional property. Total acquisition cost estimate (price + SDLT + £2,000 fees).

8. Underwriting Signals Per property: pass/fail table against thresholds from underwriting-defaults.json. Score each property 0–6 (one point per signal passed). Display as a column.

⚑ = data quality flag (not a hard fail)

9. Ranking and Recommendation

State the ranking clearly:

• 🥇 1st: [Property] — [one line reason]
• 🥈 2nd: [Property] — [one line reason]
• 🥉 3rd: [Property] — [one line reason]

Then a 2–3 sentence recommendation paragraph: which property the user should focus on, what to negotiate on, and what to verify before making an offer.

Key Rules

• Relative ranking, not absolute verdict — the goal is to rank the three against each other, not to apply a single BUY/WATCH/PASS. A property that would be PASS in isolation can still rank 1st if the other two are worse.
• Parallel tool calls — fetch all data for all properties concurrently wherever inputs allow. Do not chain calls sequentially unless the output of one feeds the next.
• Deduplicate same-postcode calls — if properties share a postcode sector, one property_comps call covers both. Reuse the result and flag it.
• Normalise all rents to monthly before any yield calculation.
• Separate student and professional rental markets if signals are present (weekly pricing, "pppw", shared houses, university proximity).
• Flag EPC/listing floor area mismatch > 10% — may indicate stale cert or unmeasured extension.
• Show both primary and additional-property SDLT — do not assume which applies; present both.
• BTL scenario — only include if at least one property has a known asking price and at least 3 rental listings.
• Flat-specific extras — for any leasehold flat, call property_blocks to check building trading history and company_search on the freeholder if named.
• British spelling throughout (analyse, colour, organised, favour).
• Round yields to one decimal place (e.g. 5.8%).
• Format EPC match rate as a percentage (e.g. 67%, not 0.67).
• Always include the disclaimer: data analysis only — not professional valuation or investment advice.

Edge Cases

• Only 2 properties given: Run the same workflow; produce a 2-column comparison. No need to ask for a third.
• One property is clearly dominant: Still present all sections; don't truncate the losing properties — the user may have a reason to prefer them.
• All three are in the same postcode: Note that comps and rental data are identical across all three; the comparison differentiates on asking price, EPC, size, and listing-specific data only.
• No rental listings: Flag yield as unreliable. Do not present a yield figure without data. State "insufficient rental data — yield not calculable" and suggest a wider search radius.
• Rightmove listing removed: If rightmove_listing returns an error for a URL, note the listing is no longer active and proceed with whatever address/postcode data is available.

What This Skill Does NOT Do

• Provide mortgage advice or affordability calculations
• Predict future prices
• Replace a RICS valuation
• Give legal advice on purchasing
• Assess structural condition

Always include: data analysis only — not professional valuation or investment advice.