Back to the discipline

Principle

Warrant: Derivation

Access to the record is conditioned on contributing to it.

Nobody free-rides on trust generated by data they declined to supply. The condition is written into the relationship at the start, where it is a term, rather than raised later, where it is a favour.

01 · Why the question is structural

A record is worth querying because of what other parties put into it.

What makes a record worth consulting is not the skill of whoever keeps it. It is the breadth of what has been fed in. A record assembled out of one party's own history tells that party what it already knew. A record assembled out of many tells each of them something none of them was in a position to observe.

That is the entire source of the value, and it settles what kind of question access is. Who may draw on the record is not a commercial question about tiers and entitlements. It is a structural question about what keeps the thing worth drawing on at all.

Stated the other way round: every query answered against the record spends something somebody else supplied. If queries can be spent by parties who supply nothing, spending and supply come apart, and the record thins in the way a commons thins. Not through anyone behaving badly. Through arithmetic.

So access is governed by a rule about contribution, not by a rule about payment. What is paid, and by whom, is a separate decision, settled separately at who pays.

02 · The condition, as a term

Warrant: Foundation

Agreed at the start, where it is a term, never asked for afterwards, where it is a favour.

The condition is short enough to state in one line. A party that draws on the record contributes real outcomes back to it. Not summaries. Not the cases that reflect well. Not an extract prepared for the purpose. The outcomes as they occurred, on the deployments the record was queried about.

Where that condition sits in the relationship decides whether it survives. Placed at the start, it is a term of participation: priced into the arrangement, agreed between parties whose interests are still symmetrical because nothing has happened yet. Placed afterwards, once the record has been useful and a bad outcome has arrived, it is a request, and a request arriving at that moment asks a counterparty to volunteer the one thing they would most prefer to keep. The wording does not change. The position in time changes everything.

It goes in the contract, in plain language, alongside the other two decisions this firm makes before an engagement rather than during a dispute: who bears the cost of a wrong answer, and how liability is allocated when one arrives.

  1. 01 Access is conditional, not purchased The right to put questions to the record is a term of the relationship, not a product bought separately from the obligation attached to it.
  2. 02 What is contributed is outcomes What actually happened on the deployment the record was consulted about, in the form it happened, at the point it is known.
  3. 03 Unflattering outcomes are in scope A contribution filtered down to the successes is not a contribution. It is advertising, and the record has no use for it.
  4. 04 The term is fixed before outcomes exist Written at the start, while both parties are still deciding, and not renegotiated once results begin to arrive.

03 · The outcomes nobody volunteers

The crews that performed, and the ones that did not.

A record of successes is a brochure. What makes a record load bearing is the presence of the deployments that went badly: the crew cleared for work it then struggled with, the qualification that did not survive contact with the plant, the estimate that proved optimistic under conditions nobody had written down. Those cases carry more information per observation than any success, because they are the ones where the prior was wrong, and a body of evidence that cannot see where it was wrong has no mechanism for becoming right. They are also the cases nobody volunteers. Which is exactly why supplying them is a term rather than an appeal to good citizenship.

A record containing only what its contributors were content to send is a record of what they were content to send.

04 · A shared record is not a rating

One is an opinion sold to whoever will buy it. The other collapses the moment it can be taken from without being added to.

The two get confused constantly, and the confusion matters here because they fail in opposite directions. A rating is an opinion about a subject, formed by one party and sold to another. Its integrity problem sits on the payment side: a rating paid for by the party being rated has its incentive running the wrong way, which is the argument at who pays. Its supply side is unremarkable, because the party forming the opinion needs no cooperation from the subject in order to keep publishing.

A shared record has the opposite profile. Nobody has to buy a flattering answer to break it. It is enough that the parties who could feed it decide, individually and quite reasonably, to consume rather than contribute. The failure is depletion, not flattery, and depletion has no villain in it.

A rating

An opinion about a subject, formed by one party and sold to another. It is worth exactly as much as the independence of the party forming it, which is why the payment relationship decides its value.

Bayswater is not one, and the reasoning is at seven classifications, refused.

A shared record

An accumulating body of checked outcomes that many parties draw on and many parties supply. Its worth is set by what has been put into it, so its failure mode is exhaustion rather than corruption.

Fixed here, and used in this sense throughout the site.

Free-riding

Drawing on trust generated by observations somebody else supplied, while withholding your own. Rational at every individual step, and fatal to the record in aggregate.

The condition in section 02 exists to make it structurally unavailable rather than merely discouraged.

05 · Withholding runs one way

The parties with the worst outcomes have the strongest reason to withhold them.

If withholding were random it would cost the record precision and nothing more. It is not random. The reason to withhold is strongest exactly where the outcome was worst, because that is where it is most embarrassing, most likely to be contested, and most likely to be read as an admission by somebody with an interest in reading it that way.

The consequence is not a record with gaps in it. It is a record with a bias whose direction is known in advance. It is most confident precisely where it has been least tested, and least accurate precisely where inaccuracy costs the most: the deployments that go wrong, the conditions nobody anticipated, the failures where somebody is hurt or a programme is ruined. Wrong in the middle of its range, a record is a nuisance. Wrong at its edges it is dangerous, and the edges are the first thing selective contribution removes.

Which is why the condition cannot be softened into an aspiration or a best-efforts clause. A best-efforts obligation to contribute is honoured most faithfully by the parties with least to hide, and that reproduces the same bias one degree removed, wearing better manners.

Three kinds of outcome, the reason each is withheld, and what the record says as a result
The outcomeReason to withhold itWhat the record then says
The deployment that went wellNone. It reflects well and it is easy to sendConfident, and correct
The deployment that went badlyContested, uncomfortable, and possibly evidenceSilent, where it was needed most
The near miss with no consequenceNothing formally happened, so nothing is reportedBlind to the conditions that precede a failure
The pattern is not dishonesty. It is a predictable ranking of what is comfortable to send, and the ranking is inverted from what the record needs.

06 · Accuracy as a shared liability

Accuracy stops being this firm's problem alone.

The condition does one further thing, and it is the reason this is a principle rather than a policy. A record kept by one party, fed by one party and defended by one party makes accuracy that party's private burden. Every counterparty is then free to treat it as a supplier's claim: to argue with it when it is inconvenient, to lean on it when it is not, and to feel no obligation at all when it turns out wrong.

Reciprocity changes the relation. A party that has contributed outcomes holds a stake in the record being right, because its own observations are inside it and its own future questions are answered out of it. Accuracy becomes a shared liability rather than a vendor's promise, and the argument changes character with it: no longer whether the record can be trusted, but what is missing from it and who is going to supply that.

This is the only arrangement under which a shared record survives contact with the parties it describes. Every other version asks a counterparty to accept a judgement it had no hand in producing, which it will do for as long as the judgement is convenient and stop doing on the first day it is not.

Two obligations therefore sit on this firm from the next engagement forward: revisiting its own claims against what happened, set out at closing the loop, and publishing a stall in verified coverage rather than smoothing it, set out at the plateau principle. The counterparty supplies the outcomes without which neither can be met. Whether every counterparty accepts that is a live question, kept where the live questions are kept, at open questions.

The record is worth querying because of what other parties put into it.

Write to Bayswater