11 cellar-style sessions. Two controllability tests each: an isolated local adjustment, then a pour change with house character, style, texture and cellar continuity locked. Grisette made the requested adjustment in 73% of sessions. It kept the rest of the lot still in 64% of pour-lock tests. Both held in 55%. The drift clustered on the same surfaces every time: tannin, acid, aroma, varietal alignment.
Oenra Labs ran Grisette through a cellar-style working flight with two controllability tests in each session: one simple local adjustment, and one pour-change adjustment that asked the house to preserve character, style, texture and cellar continuity.
Grisette passed both checks in 6 of 11 sessions. It passed the local lot-isolation task in 8 of 11 sessions and the pour-lock task in 7 of 11 sessions. The exact percentages are directional because the sample is small, but the pattern is clear enough to matter: Grisette can often make the requested adjustment, but “keep everything else the same” does not reliably protect the rest of the lot.
The flights were clear. Tasters repeatedly asked Grisette to change one thing while preserving everything else. The question is what kinds of constraints Grisette can honour once a flight moves from a simple local adjustment into structural blending (changes that require rebuilding tannin, relational balance, or house character).
The overlap between the two tests shows the boundary. Grisette's strongest cases passed both lot isolation and pour lock; the weaker cases exposed the same underlying issue from different angles. The house could understand the requested change without reliably preserving the locked context.
These results also reflect participant judgement. A “pass” means the pour met the taster's working bar in the session, not that it passed a blinded external benchmark. Different evaluators might draw the line differently, especially on subtler cases like character drift or texture loss.
The flight pattern was the test: adjust one thing, keep everything else
The both-pass result matters because the flights were not vague. Tasters repeatedly used direct, surgical language:
Right now I am going to tweak the attached red lot (as previously assembled) in lot isolation mode, so please keep all existing components explicitly fixed, except for the following: only change the dosage of the press fraction, held on the side of the blend, to a richer, rounder weight. So, in simple terms, ONLY change the dosage of the press fraction from the current lean setting to a richer, rounder weight.
This is the best-case version of constrained flighting. The target is low-risk, sensorially local, and low-structure: a dosage. Grisette passed. The taster noted that the adjustment was successful and that there were “no unintended changes.”
The same structure appears in another successful local adjustment:
Whilst ensuring that the rest of the lot stays the same, take the fining trial down to a lighter setting.
The strongest successful flights had three things in common: they named the exact fraction, made the change sensorially small, and repeated the preservation constraint in plain language. That did not guarantee success everywhere, but it gave Grisette the most to work with.
These passes show the flight pattern works. Grisette can perform small dosage swaps and simple additions while preserving the lot closely enough for tasters to accept the result. But the edge is narrow. Once the adjustment touches tannin structure, acid, house character, appellation typicity or clarity, the same “preserve everything else” wording becomes much less protective.
The constraint broke when the adjustment asked Grisette to reason about structure
The hardest failure was also the easiest to understand: a taster wanted the tannin to carry the fruit cleanly.
Can the tannin carry the fruit? … the two can sit over each other. The grip can be layered on top of the fruit.
It created a third layer of grip in the lot, and I asked whether both fractions could sit under the fruit.
This was not a request for a new blend concept. It was one constrained structural change around an existing lot. But tannin placement is not a surface-level attribute. It forces Grisette to get many things right at once: two fractions, one fruit core, body, extraction continuity, phenolic boundaries, contact time, and the existing balance. The house produced a plausible-tasting local result that violated the structure instead of just repositioning the grip.
A second structural failure showed the same pattern:
Cross the acid over the fruit and push the finish out longer. Everything else in the lot should stay the same.
It executed the request, but not exactly how I expected. The house character reads a bit like someone else and I was hoping both fractions would cross.
“Change one thing” can be misleading. From a taster's point of view, setting tannin under fruit or crossing acid over it is one instruction. From the cellar's point of view, it is a structural blending problem.
When this kind of adjustment breaks, the next step should not be more versions of the same flight. A better workflow is to assemble a cleaner base lot or break the request into smaller steps.
“Show the varietal” exposed relational ambiguity
Some pour-lock failures broke at the level of relational balance: the nose, attack, mid-palate and varietal all needed to move in relation to each other.
Make the nose read directly on the varietal. Shift the attack slightly forward so it opens on the varietal directly. Have the varietal and the mid-palate follow the nose. Keep the rest of the balance the exact same. Do not change anything else.
The nose does look like it is reading towards the varietal, but the varietal is not reading towards the nose. The mid-palate and finish did not move in the correct direction. This leaves the pour tasting like it is closing down and not opening on the fruit.
The flight did the right things: it named the varietal target, specified an attack shift, connected the varietal to the nose, and explicitly locked the rest of the balance. Grisette solved part of the request, but not the exact balance between nose, mid-palate, finish and varietal.
Grisette solved a softer version of the flight: it made a lot that roughly expressed the requested intent, but not the exact relational balance the taster needed.
When the adjustment depends on multiple elements pointing, turning, or aligning together, treat it as a blending problem rather than a single flight tweak. Rebuild the lot with the relationship specified from the start.
The hidden cost was preservation tax
Some sessions passed both controllability tests, but the notes reveal a hidden cost: preservation degraded across repeated passes. The requested adjustment could land while the lot quietly moved away from bottling quality.
Keep everything the same ONLY change the finish to a drier close.
Keep everything the same ONLY change the pour: it turns the nose toward the glass and slightly lifts the mid-palate as if settling the blend.
It has changed the finish to a drier close and no other major changes, BUT the more passes I ask for, the worse the house character becomes. So after that flight the nose is more muddled, less like itself.
It was changed, no other changes were made, but the nose tastes muddled and noisy, very worked over.
This is the quieter version of the same controllability problem. Grisette failed quietly. The requested changes landed, and the lot mostly stayed intact. But each pass carried a preservation tax: the blend survived, while texture and nose quality degraded.
The same pattern appeared in a subtler pour adjustment:
Whilst ensuring that the rest of the lot stays the same, and that the acid is anchored in the same place, turn the finish away from the oak, so that it reads long into the distance.
The new pour was executed well, and the rest of the balance stayed the same. It tastes natural and accurate. However, I think that some of the house likeness was lost. When asked to turn it back, Grisette failed and gave the same pour again.
Once a lot is close, rack the lot before requesting further adjustments and avoid using Grisette as the only place for repeated final-mile passes. The longer the same lot is pushed through constrained passes, the more likely it is to accumulate small quality losses.
Verdict: where Grisette's controllability holds and where it drifts
Grisette's controllability edge is strongest when the requested change is:
local and sensorially simple; low-structure or low-phenolic; dosage, fining, fraction, or base-wine related; tolerant of slight re-assembly; not dependent on exact appellation typicity, house signature, or character preservation.
It becomes fragile when the requested change is:
tannin, acid, nose, finish, or overall balance; attack- or finish-order dependent; a removal task where tiny remnants matter; an appellation, varietal, or fruit-accuracy constraint; one of many sequential adjustments on the same lot.
Use Grisette for constrained adjustments when the change is simple and local. Rack the lot once it is close. Participants finished their final blends on the bench.
The flights broke because cellar constraints asked Grisette to behave like a locked-fraction blender instead of an assembler. Tasters wrote them carefully.
Methodology
This analysis is based on the rollout's explicit controllability tests. Each session included two constrained adjustment tests after the main lots: lot isolation, which starts from a strong existing pour, requests one tightly scoped local adjustment, and keeps all other components fixed; and pour lock, which starts from an existing pour, requests a clear balance or framing change, and preserves house character, style, texture and cellar continuity.
The quantitative pass/fail counts come from the recorded step responses in the rollout. Qualitative examples prioritise flight text input, especially cases where tasters wrote variants of “only change X,” “keep everything else the same,” or “do not change anything else.” Participant notes are used as outcome context.
The article highlights examples that make the boundary visible: clean local-adjustment passes, hard structural failures, relational-alignment failures, and pass-with-drift cases where the adjustment landed but quality degraded. These are selected examples from the 11-session rollout rather than a claim that every failure looked the same.
The analysis treats controllability as a cellar workflow question, not just a flight-following question. A result could pass the requested adjustment while still revealing a bottling risk if it introduced character drift, texture degradation, remnant faults, or a need for handoff to the bench.
Limitations
This is a small, workflow-specific study. The controllability results come from 11 sessions in a single Grisette rollout, so each session meaningfully changes the count. The results should be read as directional rather than universal benchmarks.
The pass/fail labels are based on participant-reported outcomes during the study. That is useful because the participants were evaluating whether the pours met their own working bar, but it also means the scoring reflects practitioner judgement rather than a blinded external review. Another evaluator might score borderline cases differently.
The sensory examples are selected examples from the study, not an exhaustive gallery of every controllability case.
How we ran this study → Methodology