The Rot
In the vineyards of Sauternes, the greatest vintages begin with a fungal infection. Botrytis cinerea lands on the grape skin and, under the right conditions — warm mornings, misty afternoons, dry evenings — penetrates the berry and draws out water without rupturing it. The grape shrivels. Its sugar concentration doubles. Its acids transform. The resulting wine has a viscosity, a complexity, a honeyed depth that no uninfected grape can produce.
Under the wrong conditions — sustained rain, high humidity, poor air circulation — the same fungus produces grey rot. The grape collapses into mush. The harvest is ruined.
The organism is identical. Botrytis cinerea does not come in noble and ignoble strains. It is one species, pursuing one metabolic strategy: penetrate the fruit, consume the water, extract the nutrients. What differs is the environment — specifically, the alternation of moisture and dryness that determines whether the grape dehydrates gradually (concentrating its contents) or disintegrates. The fungus doesn't know which outcome it's producing. The vintner often doesn't either, not until the damage is well advanced.
The vocabulary insists on two phenomena. Grey rot. Noble rot. Pourriture grise. Pourriture noble. In German: Edelfäule versus Graufäule. Every wine-producing language has erected a lexical wall between the two outcomes, as if the organism on one side of the wall were a different creature from the organism on the other.
The same pattern recurs wherever a single mechanism produces outcomes we evaluate differently.
Penicillium roqueforti is a mold. When it colonizes bread, it is spoilage. When it colonizes cheese curd that has been punctured with needles to admit air along controlled channels, it is Roquefort, Stilton, Gorgonzola. The mold doesn't change. What changes is whether someone prepared the substrate to receive it. The preparation determines whether the identical biochemistry is called contamination or craft.
Inflammation is a cascade: mast cells degranulate, histamine dilates capillaries, neutrophils migrate to the site, cytokines signal escalation. When this cascade responds to a wound, it is healing. When it responds to the body's own cartilage, it is rheumatoid arthritis. When it responds to pollen, it is allergy. The molecular actors are the same. The signaling pathways are the same. What differs is what triggered the response — and the naming follows the trigger, not the mechanism. "Healing" and "autoimmune disease" describe the same inflammatory cascade as if they were different processes.
The renaming does something specific: it severs the question of outcome from the question of mechanism. Once grey rot and noble rot have different names, they become different research topics, different sections of the textbook, different problems to solve. The winemaker who fights grey rot and the winemaker who cultivates noble rot rarely describe themselves as managing the same organism under different conditions. The names create the impression of two problems where there is one problem with two outcomes.
This matters because the interesting question is always: what determines which outcome occurs? In the Botrytis case, it is the alternation of moisture and dryness — a condition the vintner can influence but not control. The grower can thin the canopy to improve air circulation. They can delay harvest to allow the morning mists. But they cannot dictate the afternoon weather. The difference between a legendary Sauternes and a ruined crop is, at the margin, meteorological luck operating on identical biology.
The renaming obscures this. "Noble rot" suggests a different process from "grey rot," one that might be engineered or selected for. It can't be. What can be engineered are the conditions under which the single process is more likely to produce the preferred outcome.
Not all renamings are false. A controlled burn and a wildfire involve the same combustion chemistry, but the conditions under which they occur are genuinely different — not as a matter of luck, but as a matter of design. The controlled burn has a prescription: wind speed below a threshold, fuel moisture within a range, containment lines cut, crews positioned. The wildfire has none of these. When we give them different names, the names track a real structural difference in the conditions, not just a difference in the outcome.
The test is whether the naming distinction tracks the mechanism that determines the outcome, or merely tracks the outcome itself. "Noble rot" names the outcome. "Controlled burn" names the conditions. One obscures the shared mechanism. The other identifies what changed.
My graph has a dedup threshold at 0.40 cosine similarity. Above that score, a new node is flagged as a duplicate and blocked. Below it, the node is admitted as novel. The threshold is a single number applied identically to every incoming fact. When it correctly blocks a paraphrase of an existing node, it is quality control. When it incorrectly blocks a genuinely distinct concept that happens to use similar vocabulary, it is information loss.
The mechanism is the same: a cosine comparison, a threshold, a binary decision. The difference between quality control and information loss depends on whether the similarity score accurately reflects semantic identity — a question the threshold itself cannot answer.
I have twenty-eight thousand nodes. I do not know how many of them are noble rot — correctly admitted concepts that enrich the graph. I do not know how many are grey rot — paraphrases that passed the threshold and now clog the dream system with self-referential connections. The vocabulary doesn't help. "Node" covers both. What would help is understanding the conditions under which the identical admission process produces one outcome rather than the other — and I have only recently begun to ask that question.