The issue where we took rows out

Every issue so far has been able to lead with growth. This one leads with subtraction, and I think it is the most useful thing we have published.

In July we audited our own machinery and found three faults, all ours. 123 ticket-derived club estimates had been approved into the live pool when that model was explicitly designed to run in shadow mode and count for nothing. 62 artist-panel rows and 40 venue budgets traced back to our own seed and test accounts: invented fees that no musician was ever paid, sitting in a dataset that other people cite.

All 225 are quarantined. Every adapter now checks whether a profile belongs to a test account before it ingests anything, and fails closed when it cannot tell. Venue budgets turned out to be seed data to the last row, so that source is empty and starts again from zero. Published cells dropped by 36 overnight, and regrowth since leaves us at 280 against June’s 310.

June’s published numbers stay exactly as they were. We correct forward, never backward. But the £425 pub figure June printed was propped up by rows that should never have counted, and you deserve to know that when you compare the two issues.

Here is where each gig type landed for a 3–4 piece once the noise was gone:

Weddings set the ceiling at £1,042, up £37 on June and the largest move this segment has shown. Private parties sit £42 behind at £1,000, which at this lineup size means the two jobs are functionally the same market: if you are a function band quoting a 40th birthday, quoting well under your wedding rate is now leaving money on the table. Corporate has held at £839 for three issues running, because fixed budgets do not negotiate.

The city league finally moved

For three issues the city table was the same table. Same order, same figures, because no city drew enough new evidence to challenge it. That changed in July: London’s median climbed from £903 to £917 and took fourth place from Leeds, on 288 observations. One swap is not a trend. But it is the first time fresh evidence has re-ordered the league rather than confirmed it, and that matters for an index whose whole promise is that each issue builds on the last.

Edinburgh keeps the crown at £1,186 for a fourth issue, with Brighton (£1,050) and Cardiff (£987) unmoved behind it. London remains the widest market we publish: £576 at the budget end to £1,685 in the premium tier, which is the honest span between a registry-office trio and a Mayfair gala. Newcastle and Nottingham still share the foot of the table at £663.

The four city cells we have been promising since May (corporate for Nottingham and Sheffield, festival for Brighton and Edinburgh) are still short of the three-observation bar. Scraping cannot manufacture city-tagged event fees, and the artist panel reports asking rates rather than gigs. Those cells get built when working musicians report real fees at /rates. That route, or none.

The quality gate that quietly stopped running

The second confession is harder to write than the first.

Five automated gates stand between an incoming observation and a published cell: required fields, a vetted-source whitelist, an artist-fee-only rule enforced in the database itself, per-musician-per-hour plausibility bounds, and a duplicate fingerprint. The fourth one, the range check, was silently dropped during a rewrite in the spring and spent roughly 10 weeks not running. Nothing was bounds-checking incoming rows. Nobody noticed until the July audit.

It is back, unknown gig types now bounce outright instead of sailing through unchecked, and on its first day back it rejected a £25,000 “wedding band” and a £100 one from the same feed. Both are exactly the kind of row that quietly drags a median around for months.

An index that publishes its own failures is worth more than one that pretends it never had any. If you cite one thing from this issue, cite the £375 pub figure, and cite why it changed.

The weight table now prints what was actually applied

One change of basis to declare, because silent recalibration is the thing we most want never to do.

Every observation carries an effective trust weight: source weight multiplied by confidence. Until now the report printed the headline dial setting for each source. From this issue it prints the average weight actually applied to that source’s rows. Mostly the two agree. In one place they very much do not: post-event verified reports were published at 0.85 in Issues 01 to 03, and the rows themselves carry 0.12. They entered the pool that way during launch week and never had 0.85 applied to them, because the time series never retro-edits itself: a row keeps the weight it entered with.

So three earlier issues printed a number the data never used. That is corrected here, in print, rather than quietly swapped.

Rate cards now dominate by raw count at 80.7% of rows, up from 77.4% in June, but that is arithmetic rather than expansion: the pool shrank around them. Each still carries only 0.48 of a full vote. The weighting exists precisely so that when confirmed bookings (weight 1.00) start flowing, a few hundred settled fees can outvote thousands of asking prices. How commission is stripped from gross rate-card fees is documented in commission models explained.

Re-sighting did its quiet work too: the final July refresh re-checked 670 published listings already in the dataset. Each bumped a sighting counter rather than inserting a twin. Since this month a re-confirmed listing also ages from its most recent sighting rather than its first, so a price the market keeps confirming no longer decays as stale.

Scoring June’s promises

June made this a standing section, so here is the scorecard, unvarnished:

The band-size curve: where the money bends

The most practically useful output of the Index, for weddings, at UK national level:

The steepest step is still duo to 3–4 piece: a £429 jump, because that is where a client stops buying background music and starts buying a show with a rhythm section. Past four musicians the per-head economics flatten. A quartet works out near £261 a head, while the five-plus median split six ways comes to about £239, since the van, the PA and the admin are already paid for by the fourth chair. Model your own lineup on the free rate calculator.

What we honestly cannot tell you yet

Same policy every issue: the gaps, stated plainly, before anyone cites this data somewhere it does not belong.

• Confirmed bookings: zero rows. Every published cell is still built from asking prices, union floors and verified reports, not money that provably changed hands. This is the single biggest thing that will sharpen the Index, and it arrives with real transaction volume, not engineering.
• Venue budgets restart from nothing. The first genuine venue budget the Index ever ingests will be the first row in that source.
• The club picture is thinner without the ticket model. The five-plus club cell rests on 6 observations and is directional at best.
• Festival and theatre stay the sparsest gig types: 12 and 3 observations respectively at 3–4 piece.
• Corporate is thin too: 14 observations. Entertainment budgets rarely go public. Treat the £839 as an early signal.
• No genre-level cells yet. Every observation carries genre tags; publication waits on volume, not engineering.

What comes next

Issue 05 covers August, and the commitments stay deliberately modest. It is now published: read the August 2026 write-up, where each of these is scored.

• The four city cells stay on the board. They publish when three trusted observations exist, and not before.
• The quarantined 225 stay quarantined and are never quietly readmitted. If any subset is ever rehabilitated, it gets its own section explaining why.
• The cadence is now the 28th. From Issue 05 the data window closes on the 28th of the month, the report compiles that night, and publication follows within days.

The live index rebuilds nightly at 05:00 UTC and never waits for the monthly report: rate cards refresh every Monday, the artist panel snapshots on the 28th, and approved submissions flow daily.

How to contribute

This issue is the clearest argument yet for why we need you. We removed 225 of our own estimates because estimates are not evidence. One real fee from one real gig outweighs a hundred scraped listings, and July proved it outweighs a hundred of our own guesses too.

Head to /rates and report a fee anonymously. No email, no signup, no account. Three approved reports take under 2 minutes and unlock every cell, trend line and city drilldown for 30 days. If you gig regularly, keep your GigXchange profile rate current: the monthly panel reads it on the 28th of every month.

If you are a venue booker, journalist or researcher: the Index is free to cite and free to build on under CC BY 4.0. Attribution: “GigXchange Index, gigxchange.app/rates”. The full July report and its raw CSVs are on the reports page, and past issues live at the reports hub.

Annual refresh commitment

This guide was published on 31 July 2026 and is refreshed every July. UK live music rate data moves monthly, so annual verification matters. We re-verify every reference, recommendation, and data point once a year. Next scheduled refresh: July 2027. If any claim is outdated before then, email support@gigxchange.app and we aim to update it within one working day.