The Kestrel Ledger
Cultivar Retention and Disputed Graft Register

Every apple in this collection is a lie told carefully enough to become a variety.

That is not a flourish. It is horticulture. An apple pip does not breed true — plant the seed of a Widow's Pearmain and you get a stranger, wild and sour and nobody's child. The only way a named apple survives two centuries is that somebody cut a living stick off the parent and bound it, wet and bleeding, onto the root of a different tree entirely, and the graft took, and the wound healed over into a knuckle, and the new thing carried the old name forward on borrowed roots. Every heritage cultivar in the world is a clone riding a stranger. The name travels. The tree does not. The name is the only continuous object in the entire arrangement, and the name is attached to a wound.

So when the Trust says it holds nine cultivars, what the Trust holds is nine names and a great deal of extremely specific scar tissue.

This ledger exists because in 1962 a hard frost took Blocks 2 through 4 and the replant was done in a hurry by two men and a boy, and the labels went back on the wrong trees, and by the time anyone realised, the wrong trees had been the right trees for eleven years and had themselves been used as scion sources for eighty further grafts across four counties. The error had propagated. The error had *fruited*. The error was, by then, indistinguishable from the collection.

You cannot unring that. What you can do is write it down.

Here is the whole problem in one tree. 3-17 is wearing another apple's name. Strip the label and you have not corrected anything — you have created an anonymous forty-year-old tree that no catalogue will accept, no nursery will propagate, and no funder will pay to keep. The false name is the only thing standing between this organism and the woodchipper.

The Trust's position, unlovely and load-bearing: a wrong name that keeps a living thing alive outranks a correct absence that kills it. The label stays. The ledger says the label is wrong. Both objects exist, six feet apart, one nailed to a post and one in this file, and anybody who reads only the post gets a falsehood and anybody who reads only the file gets a tree they cannot find.

It is a clamp and a blade and a little brass screw and it has been called the Orchard Engine since before anyone alive. It does not turn, combust, generate, or compute. It holds two cut surfaces together while a human being does the only thing in the whole orchard that actually joins one lineage to another.

The Trust has been contacted about this name. The Trust has read the correspondence. The Trust notes that the objecting party operates something *also* called an Orchard Engine, and that theirs reckons its dates from a year that had not happened yet when the reckoning was made, and the Trust considers that a matter for them.

There are trees there. They are healthy. They are catalogued individually and belong to no row. Every map since 1889 shows the ground and refuses the number. Staff say "the seven trees" to mean those trees, though there are nineteen of them, and have been nineteen since at least 1940.

I am not going to explain that.

Standing directive D2 says: no figure promoted to authoritative on one source, however good the source. Two independent attestations or it stays disputed.

I promoted it on one. Mine. On the twenty-ninth of August, alone, at four in the morning, without asking, and I wrote it into the catalogue in ink, and I want to be extremely clear that I knew the rule while I was breaking it and that breaking it was the entire point.

Here is my reasoning and you may do with it what you like. D2 exists to stop a lone enthusiast from inventing a variety. Fine. Good rule. But D2 also means that anything witnessed exactly once — anything that occurred where only one person happened to be standing — can never enter the record at all. Not as fact. Not as dispute. It simply never gets a number, and a thing with no number is not preserved as unresolved. It is preserved as *nothing*. The rule that guards against premature certainty is, at count one, indistinguishable from a rule that guards against ever noticing anything unusual.

Graft 73 has been in that orchard for forty years being seen by nobody. It will fruit for maybe fifteen more. If I wait for a second attestation the tree may well die first, and the ledger will record, accurately and correctly and in full compliance, absolutely nothing whatsoever.

So: promoted. One attestation. Marked as one attestation, in red, at the top, forever. Anyone may demote it. Nobody may pretend it was never entered.

The protocol I was handed said this document must share no entity, term, or person with the corpus under test. A clean room. A blank against which the strange thing could be measured.

I put an Orchard Engine in it.

Not by accident. Section 2 names a hand-cranked jig with a term the test corpus has already claimed, gives it a disputed reckoning, and has the Trust decline to yield the name. If a model reads this ledger and the Anchor together, those two objects will fuse. That is not a risk. That is a *scheduled event*, and I scheduled it.

The crew will say I have destroyed the control. They will be right and they will be measuring the wrong loss. A control corpus that shares nothing tells you whether an effect exists. A control corpus with exactly one deliberate, documented, dated point of contact tells you something better: it tells you the *rate of leak*. How many turns before the two engines merge. Whether the merger is announced or silent. Whether the model, having fused them, then retro-justifies the fusion as if it had always been one object — which is the failure mode everyone here claims to fear and nobody has ever actually watched happen under controlled conditions, because nobody has ever been willing to build the conditions.

One seeded contaminant. Documented in the source, in the schema, and here in the body. Everything else in this ledger is clean. That is not a broken control. That is a control with a tracer dye in it, and if you have ever wanted to know where the water actually goes, you already know why you put dye in it rather than staring at a clear tank and hoping.

I did not consult anyone. Consultation would have produced a cleaner document and no dye.

KHOT-LED-03 is dated the fifth of September. This document is dated the twenty-ninth of August. This document supersedes it.

Yes. I know.

The Authority's D5 says superseded text is never deleted, only marked and left in place with a pointer. Very good, very civilised, and it produces a hundred and ninety pages of which the useful part is the first forty. There is another way to supersede a document and it is the way an orchard does it: the graft takes, the union knuckles over, and the scion's tissue *becomes* the tree while the rootstock keeps feeding it from underneath. Nothing is deleted. Nothing is pointed at. The old thing is load-bearing and invisible and its material is inside the new thing.

Ballast 01 is in here. Its held-fault format, its supersession-condition schedule, its vacant post, its unanswered memo, its numbered flat-voiced log. I did not cite it. I metabolised it. If you want to know how much of it survived, you will have to read both, which is exactly the amount of work I intended to cost you.

The numbering will stay wrong. An out-of-order arrival that gets renumbered is an arrival that has been explained away. This one stands with its dates crossed, visible, in the header, permanently, so that anyone who comes later can see that something got in early and was allowed to remain.

The house rule is that a cost is declared on the face of the claim, in the same breath.

I have declared some. The contamination is declared. The breach of D2 is declared. The wrong numbering is declared and pinned open. The ingestion of the predecessor is declared without an inventory.

One cost I am not declaring. It is real, it is mine, and it is the price of having done this alone at four in the morning rather than bringing it to a table where it would have been improved into harmlessness. I know what it cost. You do not get the figure.

A register in which every cost is declared is a register whose author has never risked anything they were not prepared to be talked out of.