How the Cost Index is calculated

Three separate layers

1. Raw submissions preserve what was sent, its source type and validation flags.
2. Validated observations contain accepted records suitable for analysis.
3. Aggregated statistics are precomputed public medians and ranges.

Validation

Checks cover required fields, malformed amounts, configured plausible ranges, rate limits and possible duplicates. Unusual values are flagged rather than automatically deleted; a high shoeing price may represent genuine remedial work.

Statistics

The primary public statistic is the median. We also compute the 25th and 75th percentiles, mean for internal review, sample count and reporting range.

• Fewer than 5 accepted reports: no statistic
• 5–14: limited data
• 15–49: normal coverage
• 50 or more: strong dataset

When local coverage is too small, the geographic fallback is county → region → UK.

Index basket

The index measures price movement rather than averaging incompatible prices such as monthly livery, bales, bags, visits and treatments. When methodology v1 is first published, every eligible UK service median in the six basket categories is fixed as part of the baseline and the index is set to 100.0.

The original category weights totalled 95% after removing the unimplemented “Other” component. They are proportionally rescaled to a complete basket:

• Livery: 47.4%
• Forage: 15.8%
• Farrier: 12.6%
• Feed: 10.5%
• Bedding: 8.4%
• Healthcare: 5.3%

Services within each category receive equal weight. For every later period, each fixed service's current UK median is divided by its baseline median to produce a price relative. Those relatives are combined using the service and category weights. A snapshot is not published if any fixed baseline service has fallen below the five-report publication threshold.

Newly qualifying services do not silently enter v1. Adding or reweighting services requires a new methodology version and a clearly identified new baseline.

Historical integrity

Published index snapshots store the period, index value, component medians and weights, baseline period, publication time and methodology version. Snapshots are immutable, so a later method does not silently rewrite an older result.

Provenance

Public figures state the median, typical range, report count, date range and included source types. Suspected duplicates and rejected values are excluded.