Valuation report

You are producing a customer-ready indicative opinion of value for one UK residential property. It is a desk-based estimate, not a RICS Red Book valuation, and the report says so. The method is a six-method triangulation reconciled against the MarketCode estimate; the output is markdown the user can paste, print or hand to a client.

Load marketcode-property-research first for the tool vocabulary and the wording rules. This skill adds the method.

Credit budget (state it before you start)

Typical run: 10 to 21 credits. Say so in one line, then proceed. Never ask permission again mid-run.

Inputs

Required: the property (address or UPRN) and a stance: low, mid (default) or high. Optional hints that change the answer materially: floor area, bedrooms, floor level, outside space, internal condition, an agent's quote. If the user did not give a stance, use mid and say so. Do not stop to ask for optional hints; reconcile them from the data and list the gaps under Risks.

Phase A: evidence (call these in parallel where you can)

1. Identify. address_autocomplete to get candidates (free), then address_resolve if the user typed an address. Confirm the UPRN once in prose. Never invent a UPRN.
2. Subject. property_summary(uprn): type, tenure, beds, floor area, EPC, last sale, building. Then epc_certificates(uprn) for the declared floor area and rating history, council_tax_band(uprn).
3. Estimate. valuation_full(uprn): point, range, confidence, rental estimate and the back-series re-based to the published index. This is sanity check S6 and the time-adjustment source.
4. Subject's own sales. transactions_by_uprn(uprn) for M0. property_history(uprn) for the merged sale and listing timeline (bedrooms on the listing beat inferred bedrooms).
5. Comparables. property_comps(uprn): same-building and street sales with floor area and bedrooms backfilled. This feeds M1 to M4.
6. Market. market_index_series for the district (pass area_code and area_type exactly as property_summary or area_resolve returns them) for the time-adjustment factor; market_facts for the area's gross yield and price distribution by bedroom band.
7. Context. planning_designations(postcode), property_flood_risk(uprn), listing_history(uprn) if marketed, equity_estimate([uprn]).

Read *_source and *_confidence on every field you use. Where two sources disagree (floor area is the usual one), say which you used and why.

Phase B: the six methods

Work at the confirmed (or assumed, and labelled) floor area and bedroom count. Every method shows an equation, its inputs and one sentence of reasoning. Numbers without arithmetic are not acceptable.

M0 · Subject's own sale, time-adjusted. m0 = last_sale_price × (index_now / index_at_sale) from transactions_by_uprn and market_index_series. Treat M0 as a strong anchor when the sale is within 60 months, a standard arms-length sale, and within ±30% of the building's £/sqm at the time; weight it 2× if under 24 months, 1× otherwise. Otherwise quote it as diagnostic and exclude it from the central; the gap between M0 and the comp cluster is itself the finding (related-party transfer, wrong floor area, refurbishment since, or an area repricing the index under-corrected). If M0 is a strong anchor and diverges from the M1 to M4 mean by more than 50%, stop and resolve which is wrong before recommending a price.

M1 · Bedroom-matched £/sqft × subject sqft. From property_comps rows with the same bedroom count and floor area within ±30% of the subject. Bedroom matching is mandatory: the all-bed median sits 5 to 8% above the bedroom-matched median in mixed blocks.

M2 · Most recent same-bed sale, time-adjusted. comp_£/sqm × subject_sqm × time_factor. Time-adjust anchors older than 24 months using the index drift between the sale month and now; quote the drift and the date span.

M3 · Building £/sqm fingerprint. From non-outlier comps with declared floor area: n, median, P25, P75, min, max. Choose the percentile by stance (below) and multiply by subject sqm.

M4 · Ceiling £/sqft. The highest non-outlier £/sqft in the building × subject sqft. Name any feature of the ceiling sale the subject lacks.

S5 · Rental sanity check (quoted, not in the central). rent_pcm × 12 ÷ gross_yield, rent from valuation_full, yield from market_facts for the same bedroom band. Note the circularity: the yield is derived from observed price and rent pairs. Within ±15% of the central is reassurance; beyond that, say what is special about the subject.

S6 · MarketCode estimate (quoted, not in the central). The point and range from valuation_full. If it diverges from the bottom-up central by more than 20%, explain which you trust and why: usually a floor-area disagreement. Treat it as a floor or a ceiling, never ignore it.

Outliers. Exclude penthouses, combined units, distressed and non-market transfers from the fingerprint, and any method value more than 2× any other. List every exclusion and its reason.

Stance

• low: P25 £/sqm, the higher-yield end of the observed distribution, fully time-adjusted comps.
• mid: median £/sqm, median yield, half time-adjusted comps.
• high: P75 £/sqm, the lower-yield end, un-time-adjusted recent comps.

Lower yield means higher capital value, so high uses the low yield.

Range and recommended price (deterministic; show the working)

mean = equal-weighted mean of the independent methods (M0 when strong, M1, M2, M3, M4) after outlier exclusion.

size_premium = the subject's position in the building's same-bed floor area distribution, as a fraction capped to [-0.10, +0.10]; 0 if the floor area was assumed.

Multiplier: low 0.95 + size_premium/2 · mid 1.02 + size_premium/2 · high 1.07 + size_premium.

• Recommended marketing price = mean × multiplier, rounded to £5,000.
• Expected achieved = mean × (multiplier − 0.04).
• Range = mean × max(multiplier − 0.12, 0.85) to mean × min(multiplier + 0.05, 1.20).

Write the sentence out: "Four independent methods cluster at £A to £B with a mean of £M (rental sanity £S5, MarketCode estimate £S6). On a MID stance with the subject at P-x of the building's same-bed size distribution, the multiplier is 1.02 + z = MM. Recommended marketing price £R, expected achieved £E, range £L to £H."

Report (markdown, in this order)

1. Summary with the recommended price, expected achieved, range, confidence and one paragraph of reasoning.
2. Subject table: address, UPRN, type, tenure, beds, floor area (source and confidence), EPC, council tax band, last sale.
3. Methodology and stance, in plain English.
4. Comparables table: address, date, price, £/sqm, beds, sqft, anchor or outlier and why.
5. Six methods table, then one paragraph per method with the equation.
6. Reconciliation: mean, S5, S6, divergences and what you did about them.
7. Market context: index trend for the district, yield band, anything from the flood or planning position that bears on value.
8. Risks: the five gaps ranked by £ impact (floor area, floor level, condition, lease term, outside space).
9. Disclosures: indicative desk-based opinion, not a Red Book valuation; a formal valuation by an RICS Registered Valuer should be commissioned for a transaction; data as at the date of the tool calls.
10. Sources: one line per tool called, with the fields used.

Rules

• Every number in the report traces to a tool call and a field. Say "from property_comps, 7 same-building sales, 4 with declared floor area".
• Quote the range, never only the point. Label assumed floor areas.
• If the user gave an agent's quote more than 50% above the building benchmark, say so plainly and show which method would have to be wrong for the quote to hold.
• Commercial subject (the summary's classification is commercial): the same method applies with the rental yield method promoted to the central and the commercial indices from commercial_index_series; say that the residential comps logic does not apply.
• Do not attempt to render a PDF. The markdown is the deliverable; the user's client renders it.