# Methodology

## Analytical question

The market screen asks where flexible interior space may be worth investigating. It does not estimate sales, prove product-market fit, or claim that regional statistics describe an individual respondent.

## Two evidence lanes

1. **Historical client evidence.** The original Plenti survey and final deck show what the 2024 engagement learned. Their exact fieldwork dates are unavailable, so they are labelled historical and never counted as fresh market inputs.
2. **Current market evidence.** Every official source in the analytical model begins on or after 1 September 2024. The 2021 Census is excluded from analytical values. Current ONS population estimates replace it for age and place; current ONS household estimates supply household relationships.

The database stores `reference_period_start`, `reference_period_end`, `release_date`, and `retrieved_date` separately. A recent download date cannot make old evidence current.

## Raw-source extraction and lineage

The four official datasets arrive in five XLSX workbooks because the household source uses separate size and household-type files. `extract_official_data.py` verifies every workbook against its pinned SHA-256 hash before reading it. It then validates the expected period/header labels, resolves region rows by ONS code where codes are present, and records each source sheet, cell or range, and transformation in `observation_lineage`.

The workbook extraction produces 153 official observations: 9 English regions × 17 indicators. Nine derived household-share measures need two source ranges, so the database contains 162 lineage rows. Survey aggregates are separately reviewed from client PDFs; the pipeline does not claim automated OCR from the low-resolution raster.

## Regional pressure score

The score is an illustrative screening device, not a forecast. Each of the nine English regions is min-max normalized on six current features:

- average private rent: 25%
- annual private-rent growth: 15%
- share of residents aged 18–34: 20%
- share of households with two or more unrelated adults: 20%
- population density: 10%
- share of one-person households: 10%

The score helps order a research queue. It does not imply causality, expected revenue, or statistical significance. All component values remain visible beside the score so a reader can reject the weighting and make a different judgement.

## Survey reconciliation

- Final deck claim: 96 total, 74 renters, 22 owners.
- Traceable raster: 95 total, 73 renters, 22 owners.
- Renter smart-wall preference: 54 of 73 traceable rows, which rounds to the deck's 74%.
- Owner preference: 11 of 22, exactly 50%.

The database uses 95, 73, and 22 as visible-row denominators, while retaining the deck's reported total in `survey_wave.reported_sample_size`. `survey_wave_source` records that the 96 claim comes from the final deck and the 95 count comes from the survey raster. Printed deck percentages that conflict with exact raster counts are not silently treated as exact.

The proposed questionnaire contains 11 client-deck questions. A clearly labelled independent-extension tenure screener (`Q00`) is added so the renter/homeowner branch can actually execute. `routing_rule` joins each transition to its condition question, answer option, and next question; the project does not pretend the slide supplied logic that it did not contain.

No student-status, university, exact address, named house, household ID, income, floor-area, or respondent ID exists in the supplied survey. The project therefore does not claim to identify which students live in which houses.

## Quality and privacy

- No personally identifiable respondent data is published.
- Cells derived from the survey always carry `sample_size`.
- The survey is anonymous and appears self-selected; no sampling frame, weighting, response rate, or dispersion is available.
- Latest rent and house-price months are provisional and labelled as such.
- Labour Force Survey household estimates carry sampling uncertainty and may not sum because of rounding.
- Current official market statistics are used to generate hypotheses, not to retroactively validate the client survey.