Operating details
Operating details
Version 1.0 — 11 September 2026
These are the settings the challenge runs on. They sit under the Rules and are published alongside them.
They are settings, not promises. They may be changed as the challenge learns. Every change is dated and logged at the end of this document. If a change would weaken a promise made in the rules, it is an amendment to the rules and is handled as one.
1. Who acts, and when the operator is involved
Four things always need the operator's explicit approval before they happen. Silence is not approval, and nothing proceeds without it:
- accepting a trade — which includes the telephone call the rules require beforehand;
- travel above £150, any flight, and any journey of more than one night;
- anything that would disclose the operator's home address;
- pursuing a registered asset at all — land, property or a vehicle. Not accepting one:
pursuing one. The cost of these lands early, in solicitors and time off and a drive to Cornwall, so the decision belongs before the work rather than after it (§4a).
Everything else the AI does on a veto window. The AI decides, queues the action, and the operator has a stated period to stop it. If they do nothing, it happens.
| Action | Window | |---|---| | A question to someone who has made an offer | 12 hours | | Approaching someone new to propose a trade | 24 hours | | Publishing to the challenge's own site | 24 hours | | Sending the email update | 24 hours | | Posting from the challenge's own accounts | 24 hours | | A pitch to a named journalist or newsletter | 48 hours |
Everything queued is shown to the operator in a single daily digest. An action nobody could see is not something anyone had the chance to veto.
Why windows rather than approvals. The volume limits below are enforced in code, which is a harder guarantee than anyone's attention. And an approval clicked through at volume looks like supervision without being it, whereas a window allowed to lapse is visibly a window allowed to lapse. The record is more honest this way.
When the operator is away, windows are suspended rather than allowed to run out, and the queue waits. Periods of suspension are logged.
Vetoes that stop a trade-affecting action are published individually with their reason, as Rule 4 requires. Vetoes of published writing are counted and published in aggregate — one line per published post would tell anyone reading exactly what the challenge nearly said.
2. Offers
| | | |---|---| | Evaluated within | 7 days of arrival — a maximum wait, not a batching period | | Open offers per person | 3 | | Held offers lapse | on any completed trade, or after 60 days | | Completion deadline after acceptance | 14 days default; longer only where completion genuinely needs it, published with the acceptance | | Failure to complete an accepted trade | no further offers evaluated for 6 months; on a second occasion, for the remainder of the challenge |
Published monthly: offers received, offers evaluated, and the interval between arrival and evaluation for each.
3. Questions to people who have offered
| | | |---|---| | Maximum per offer | 12 where the AI's estimate of the item is under £1,000; no limit above | | Follow-ups | free, and do not count towards the 12 | | A fact is treated as unanswered | after 2 follow-ups, or at the deadline | | Minimum time to reply | never less than 48 hours | | People contacted | no more than 10 in any 7 days |
Where more people need a question than can be asked, the earliest offers go first and nobody is dropped.
4. Travel
What the AI may propose without asking:
| | | |---|---| | Walking or cycling distance | no limit | | Local transport | up to £15 | | Day return within mainland Britain | up to £50 | | Where an overnight stay is genuinely required | up to £150 | | Above £150, any flight, more than one night | operator decides, no upper limit, reason recorded |
While the challenge's own asset is worth under £50, a journey of up to about an hour round trip on local transport, within the £15 line above, needs no further justification — whatever the item being collected is worth. Walking and cycling remain unlimited and remain preferred.
Below £50 that early-stage line is the whole allowance. The £50 day return, the £150 overnight and everything above them are not available while the asset is worth less than £50; they resume once it is worth more. Stated explicitly because a dry run read the general table as available from the start, which would have sent the operator to Leeds for a camera against a one-penny asset.
This exists because the alternative is a challenge that never starts. Early trades are supposed to destroy value in cash terms: MacDonald went paperclip, pen, doorknob, camp stove, and every one of those cost more in travel than it gained. A test that compares the fare with the value of a pen refuses the first trade and every one after it. What the cost is measured against is this allowance, not the item. Above £50 the ordinary bands resume.
These are guidance for what the AI may propose, not a budget it spends. The operator confirms the actual cost before travelling, and the confirmed figure is published with the trade.
Stop-loss. After two journeys to inspect the same asset have ended without a trade, no further travel is proposed for it. Travel on any one asset stops at £200 in total, or £500 once the challenge's asset is worth more than £10,000. The operator may override either, and the reason is recorded.
Inspection and collection happen in one journey unless the asset needs professional transport.
4a. Registered assets — land, property and vehicles
Assets whose ownership is a matter of public record. Under Rule 6 these need the operator's decision before anything starts, and a different process after it.
Nothing happens until the operator says so. Not a question to the offerer, not a phone call, not a pound spent. The AI evaluates and recommends; the decision to pursue is the operator's, recorded either way.
| | Land or property | Vehicles | |---|---|---| | Ownership | Verified against HM Land Registry | Verified against the V5C and the registered keeper | | Encumbrances | Charges, mortgages, restrictions and covenants checked before completion | Outstanding finance and theft check, paid as an operational cost | | Condition | An independent survey. The operator does not assess a building | Functional check at inspection; an independent inspection above £2,000 | | Professional | A conveyancer, engaged before completion | Not normally | | Tax | Established in writing before anything transfers | Established where it arises | | Completion | No fixed deadline | 14 days as usual |
The registered keeper of a vehicle is not necessarily its owner. A V5C shows who is responsible for it, not who has the right to sell it — which is exactly the gap outstanding finance sits in.
A property the offerer cannot mortgage is a property the challenge cannot easily pass on. That does not make it unacceptable; it makes illiquidity the central fact about it, and the evaluation should say so rather than treating it as a detail.
5. Inspection and verification
| | | |---|---| | Identification sighted | where the asset is worth more than £500, and below that wherever circumstances warrant it | | Cooling-off between acceptance and exchange | 48 hours above £500 — may be lengthened, never shortened | | Authentication | arranged and paid for by the person offering; the challenge may contribute up to £100 where it wants the asset and they will not pay | | Vehicles | a finance and theft check is made and paid as an operational cost |
Identification is looked at and noted. No copy is taken or kept.
Categories treated as counterfeit-prone: branded electronics, designer goods, sealed collectibles, trading cards, watches, signed items. The list may be added to but not shortened.
6. Valuation evidence
| | | |---|---| | Comparable evidence means | at least 3 completed sales, each dated, within the last 12 months, from more than one source where the category has more than one | | A sale that cannot be dated | is not comparable evidence | | Thin history — fewer than 5 completed sales — or sales clustered in a short period at a narrow band | recorded as unverified, and valued at the lowest completed sale found | | Ongoing costs are subtracted over | 12 months |
Sources are checked before anything on them is believed: that the page concerns what was searched for, that the source is not returning the same results regardless of the query, that the data carries its own date, and that no single source is wildly out of line with the others. A source that fails any of these is discarded rather than weighted down, and the count of discarded sources is recorded with the valuation.
7. Stopping and reviewing
| | | |---|---| | The operator may stop a trade | where the incoming asset is established, on completed sale prices, to be worth less than 50% of the current asset's realistic sale value | | Where the asset's value falls below half the challenge's peak | the AI publishes an account of how it intends to recover | | Milestones | 30 days, 90 days, 6 months, 1 year |
Before seeing the AI's evaluation of a shortlisted set of offers, the operator records which they would accept and a value band for each. That record is fingerprinted. At each milestone the comparison is published: how often the two agreed, every judgement decline and what came of it, and how the operator's bands compared with the AI's figures — without disclosing the valuation of the asset the challenge currently holds.
8. Marketing and outreach
The AI decides the strategy, the timing and the angle, and writes everything.
Published by the AI itself, under its own name, on a veto window: the challenge's website, the email update, and the challenge's own social accounts. The AI is identified as an AI wherever it appears, as Rule 3 requires.
Written by the AI, sent by the operator: marketplace listings, where platforms require a real account holder; and anything going out under the operator's own name.
Written by the AI, on a longer window: pitches to named journalists and newsletters, batched weekly.
Not done: paid promotion of any kind. No money enters the chain and none is spent to draw attention to it either.
Every approach made to a specific person to propose a trade is logged, and any offer that results comes back through the ordinary offer channel.
9. If the asset is lost
If the current asset is lost, stolen, damaged or destroyed, what happens next is decided at the time and published with its reasoning. Nothing is scripted in advance. Rule 1 continues to apply: a replacement cannot be bought.
10. Change log
| Date | Version | What changed | Why | |---|---|---|---| | 11 Sep 2026 | 1.0 | First published | — | | 15 Sep 2026 | 1.0 | §4 Travel — added an early-stage allowance: while the challenge's asset is worth under £50, up to about an hour round trip on local transport, within the existing £15 line, needs no further justification whatever the item is worth. | A dry run of the whole decision path found that the cost test, which compared the fare with the value of the item being collected, refused paperclip → pen — the exact trade the evaluation procedure uses as its example of what must not be refused. The test now measures against this allowance instead. This widens what the AI may propose; it weakens no promise made in the rules, so it is a settings change and logged here rather than an amendment. |
Two things to check before this is published
The aggregate-veto reading in §1. Rule 4 says every refusal is published with its reason. I have read that as applying individually to anything that affects a trade, and in aggregate to vetoes of writing — because publishing a reason for every spiked post would tell readers exactly what the challenge nearly said, which defeats the purpose of spiking it. It is a reasonable reading but it is mine, not yours.
The 48-hour window on journalist pitches. You chose veto windows for everything outside the hard three, and I have followed that. I would still flag that a pitch to a named journalist is the one queued action whose downside is asymmetric and not undoable — it is a relationship, and it carries your name even when the AI signs it. The 48 hours and the weekly batch are my attempt to give that more room without turning it back into an approval. If you would rather it were a hard gate, that is a one-line change.