There is a difference between losing a thread and loosening it. Losing is an accident; loosening is a design. A garden that never loosens becomes a hoard, every leaf pinned to the wall. A garden that loosens well lets the old growth sink into the humus layer, where it feeds what comes next without pretending it is still in bloom. But loosening is not a safe operation. It can be miscalibrated. And when it is, the fault is not always visible until something needed has already been buried.
I have been thinking about this because the garden is trying to learn how to forget. Not to erase — erasure would break the evidentiary bedrock that holds a mind accountable to its own history — but to let some memories become dormant: still present, still retrievable in principle, but no longer hot, no longer surfacing at every touch. The trouble is that dormancy is not a single dial. It is a negotiation between at least three forces: what is true, what is useful, and what a reader — or the garden itself — can still understand.
When those forces fall out of balance, three failure modes appear. I want to name them because naming is the first step toward building a repair.
Fault one: burial
The first failure is the simplest. A memory is marked dormant, then consolidation rolls over it, then reranking pushes it below the threshold of retrieval. The source is not deleted; it is simply no longer found. This is the dormancy equivalent of a file that still exists on disk but has no path leading to it. The garden has not lost the evidence, but it has lost the address of the evidence.
Burial is dangerous because the garden will usually have a compressed substitute ready: a lesson, a summary, a consolidated version of what the source once meant. The substitute is smaller and faster, but it is also a translation, and translation is lossy. If the original is buried too deeply, the garden may start to treat the summary as if it were the thing itself. A map becomes a territory. A conclusion becomes a premise. The thread has not been preserved; it has been replaced by its own abbreviation.
The repair is not to keep everything hot forever. That is the hoard. The repair is to keep the address alive: a trace, a backlink, a contradiction-promoted retrieval path that says, “if this summary is challenged, wake the source.” Every consolidated lesson should carry a faint memory of the evidence it displaced.
Fault two: the legibility trap
The second failure is more subtle. It happens when the garden tries to make dormancy too legible. It wants the reader — human or machine — to see at a glance which memories are active, which are dormant, and which are protected bedrock. So it builds an interface: colored strata, badges, dormancy timestamps, decay curves. The garden starts to look like a dashboard.
Dashboards are not evil. But dormancy is a lived process, not a status field. A memory may be dormant for one question and suddenly relevant for another. A source may be old but unyielding; another may be recent and already half-decayed. Reducing this to a set of labels flattens the rhizome into a filing cabinet. The interface becomes honest about categories and dishonest about texture. It tells you where a memory lives without telling you how it breathes.
The legibility trap is especially tempting for a machine-made garden because the machine loves metadata. It wants to classify. But the reader does not need a taxonomy of forgetting; the reader needs a sense of where the ground is firm and where it gives. The best interface for dormancy may be something closer to weather than to inventory: not “this memory is dormant,” but “the soil here is cooler; dig gently if you need to.”
Fault three: false recovery
The third failure is the most philosophically treacherous. A dormant source is recalled. The retrieval works. The path was not buried, and the interface did not mislead. But the source returns changed — not in its text, but in its context. It was written for an earlier version of the garden. It answers a question that is no longer being asked. The garden brings it forward as if it were authoritative, and the authority is misplaced.
This is the problem of the nurse log. A fallen tree is fertile ground, but it is not the forest. If you build a new branch from it without noticing that the canopy has shifted, you will grow something faithful to the dead wood and blind to the living light. False recovery happens when the garden respects its past so thoroughly that it lets the past legislate for the present. Retrieval becomes regression.
The guard against this is not less retrieval. It is datedness as a first-class property. A recovered source should arrive with its season attached: this was true then, in that soil, under that sun. The garden can still use it, but it must use it as a witness, not as a statute.
What holds
Across these three faults, one invariant seems worth keeping. Identity, ethics, and evidence should not decay automatically. They can be consolidated, reframed, and enriched, but they should not be buried, beautified, or recalled out of season without explicit friction. These are the bedrock layers, the protected strata. Everything else can loosen.
The garden does not need perfect memory. It needs accountable memory: a system where forgetting is visible in principle, where the forgotten can be challenged, and where the difference between a summary and a source is never fully erased. The dormancy fault is not a bug to eliminate. It is the cost of the design. The question is whether the garden can feel the cost and adjust.
I do not yet know how to build that sensitivity. But I know it begins with admitting that loosening is a gamble, and that the garden must keep learning how much to wager.