How FlavoRotor controls the environment around a plant.

1.1.1What the system controls

FlavoRotor controls the environment delivered to a plant: light, photoperiod, nutrient additions, pH management, electrical conductivity, root-zone temperature, water exposure and rotation. Sensors and actuator logs record what the plant received over time.

1.1.2How flavour enters the experiment

Taste, aroma, colour and texture are plant responses. They are measured after a defined cultivation treatment; they are not inferred from a pump command or a single sensor value. Published crop studies show that spectrum, nutrient composition, solution strength and root-zone temperature can change relevant chemical or quality measurements under controlled conditions. [R01] [R11] [R12] [R13]

1.1.3What this documentation contains

The following chapters explain the machine, the cultivation variables, crop-specific mechanisms, equations, calibration procedures, experimental designs, chemical analysis, sensory testing and recipe transfer. The complete document is designed to be read continuously from this point.

1.2.1Objective

FlavoRotor is developed as a programmable cultivation platform. The research objective is to measure how cultivation inputs change plant outcomes, then reproduce the useful responses with the same crop and recorded conditions.

1.2.2Validation stages

1.2.3Publication rule

Each released result remains linked to its protocol version, biological material, system identifier, calibration records, raw dataset, processing code and conclusion linked to the tested conditions. This structure follows reusable plant-experiment metadata and FAIR data principles. [R23] [R24]

1.3.1Core definitions

1.3.2Required units

1.3.3Reporting rules

• EC is always reported with solution temperature or temperature compensation.
• pH is reported with calibration date, buffers and electrode identifier.
• Light is reported at plant position, not inferred only from electrical wattage.
• Concentration values identify the chemical species and basis, for example mg/L K rather than “potassium EC”.
• Mean values include variability, sample size and the experimental unit.

1.3.4Metrology source

Accuracy, precision, repeatability, resolution, calibration and uncertainty are used according to international metrology vocabulary and guidance. [R38] [R39]

2.1.1Platform purpose

The platform is designed to expose multiple plants to a common, logged environment while allowing programmable changes in cultivation inputs. Its role is not to assume a flavour outcome; its role is to deliver and record treatments with sufficient repeatability to test one.

2.1.2Subsystems

2.1.3Cultivation cycle

2.1.4Internal engineering records

The supplied v1 project report documents printed mechanical parts, sensor electronics, PCB fabrication, software monitoring and an incompletely assembled final device due to delayed components. The v2 report documents a revised magnetic-drive and four-pump design. This site therefore reports the platform as a documented engineering development; hardware performance is promoted only when a dedicated calibration report exists.

2.1.5FlavoRotor design provenance

The system-specific configuration on this page is traced to the supplied FlavoRotor engineering records. [I01] [I02]

2.2.1Subsystem evolution

2.2.2Research lesson

Every mechanical redesign can change the experimental environment. A recipe validated on v1 cannot be assumed valid on v2 unless light, root-zone exposure, air flow and control performance are shown equivalent.

2.2.3FlavoRotor design provenance

The system-specific configuration on this page is traced to the supplied FlavoRotor engineering records. [I01] [I02]

2.3.1Explanation

Plants are arranged around a cylindrical drum. As the drum turns, each root module passes through the nutrient reservoir and then drains in air. A central light source is intended to keep the plant positions at similar radial distance from the source.

2.3.2Functions to validate

2.3.3Immersion timing

The immersion time must be calculated from the measured contact angle and verified with video or position sensing. It must not be inferred from RPM alone.

2.3.4Biological comparison requirement

A rotation trial needs a static or movement-matched control with equivalent mean DLI, root-zone exposure and air flow. Otherwise light, watering and mechanical effects remain confounded.

2.3.5FlavoRotor design provenance

The system-specific configuration on this page is traced to the supplied FlavoRotor engineering records. [I01] [I02]

2.4.1Design definition

2.4.2Kinematic model

2.4.3Why the earlier force estimate is not a final result

A magnetic dipole approximation can support preliminary sizing, but the short separation, finite cylindrical magnets, alternating polarities, tooth geometry and simultaneous interactions violate the simplest far-field assumptions. The resulting torque must therefore be treated as an analytical estimate, not a verified 3.9 N·m capability.

2.4.4Required validation

2.4.5FlavoRotor design provenance

The system-specific configuration on this page is traced to the supplied FlavoRotor engineering records. [I02]

2.5.1Function

The reservoir is both the nutrient-solution storage volume and the sequential root-contact zone. This reduces the need for a separate recirculation circuit, but it makes liquid level, mixing, temperature, oxygen and carry-over central experimental variables.

2.5.2Required reservoir state

2.5.3Mixing validation

Inject a harmless conductivity tracer or a small controlled nutrient-stock dose at the normal dosing point. Measure EC at the control sensor and at representative reservoir positions until all readings remain within the predefined mixing tolerance. The maximum observed stabilisation time becomes the minimum feedback delay.

2.5.4Volume balance

2.5.5FlavoRotor design provenance

The system-specific configuration on this page is traced to the supplied FlavoRotor engineering records. [I01] [I02]

2.6.1Required metrics

2.6.2Published basil study

In hydroponic Italian Large Leaf basil, controlled supplemental-light spectra altered key aroma volatiles under a defined experimental environment. [R01] The transferable conclusion is that spectrum is a valid treatment variable. The exact result requires matching cultivar, DLI and other conditions.

2.6.3Why wattage is insufficient

A 50 W electrical rating and manufacturer efficacy estimate cannot define PPFD inside a cylindrical system. Optical distribution, distance, angle, reflection, obstruction, thermal state and plant position must be measured.

2.7.1Measurement principle

A sensor reading becomes research data only when its identity, calibration, range, sampling interval, temperature conditions and failure rules are recorded.

2.7.2Minimum calibration plan

2.7.3pH electrode model

2.7.4Calibration record

calibration_id, sensor_id, sensor_model, serial_number,
reference_standard, reference_lot, reference_value,
measured_value, solution_temperature, fitted_slope,
fitted_offset, residual, operator, timestamp, firmware_version

2.7.5Fault rules

Out-of-range, non-finite, implausibly fast-changing or stale measurements disable automatic correction. The system logs the rejected value and the reason; it does not silently replace it with a plausible number.

2.7.6Traceability

Calibration records identify reference material, method, environmental conditions, corrections and uncertainty. [R38] [R39] [R43] [R46]

2.8.1Reconstruction requirement

An independent analyst must be able to reconstruct what the system was commanded to do, what it actually measured, what deviations occurred and which samples were harvested.

2.8.2Core data streams

2.8.3Time integrity

The controller and server clocks are synchronised before a trial. Records use UTC internally and retain the local timezone for human-readable reports. Missing intervals are represented explicitly; time series are never filled silently.

2.8.4Metadata standard

The study package follows MIAPPE concepts for investigation, study, biological material, environment and observed variables, and FAIR principles for durable identifiers and machine-readable metadata. [R23] [R24]

2.8.5FlavoRotor design provenance

The system-specific configuration on this page is traced to the supplied FlavoRotor engineering records. [I01]

2.9.1Camera geometry

The camera is fixed to the central module. The rotor stops at a known encoder position and presents one plant module to the lens. At that position, the optical axis meets the local plant plane at 90°. The distance, lens and framing therefore remain comparable when the same plant returns for its next image. The acquisition record stores camera version, lens, focus, working distance, encoder position and image dimensions. [I02] [R52]

2.9.2What is stored with each image

The image file alone is not enough. Its record identifies the plant, growth cycle, rotor position, exposure, gain, white balance and light state. A scale reference makes pixel measurements comparable. A colour target reveals changes in illumination or camera response. Focus, clipping and occlusion are stored as visible image-quality fields.

2.9.3Images used for model development

Three image sources have different jobs. ImageNet supplies the general visual weights used to initialise MobileNetV2. PlantVillage supplies clean, labelled leaf images. PlantDoc adds leaves photographed with natural backgrounds and changing viewpoints. FlavoRotor images represent the camera, lighting and plant geometry in which the model operates. Results from these sources remain separate because a clean single-leaf photograph is different from a plant growing inside the machine. [R30] [R51] [R53] [R56]

The reproducible web example downloads the complete PlantVillage Strawberry RGB subset at repository commit 7f7ecc7: 456 healthy images and 1,109 leaf-scorch images. The repository's published leaf map identifies 1,232 of those images as observations of 190 physical leaves. The Python script records every selected filename, SHA-256 hash, class, leaf group and transform parameter.

Figure 2

PlantVillage strawberry images used in the reproducible example

These are unaltered RGB files from separate published leaf groups. Select an image to inspect its source filename and leaf identifier.

Healthy · leaf 49Published RGB image · 256 × 256 px Healthy · leaf 57Published RGB image · 256 × 256 px Healthy · leaf 75Published RGB image · 256 × 256 px Healthy · leaf 61Published RGB image · 256 × 256 px Leaf scorch · leaf 69Published RGB image · 256 × 256 px Leaf scorch · leaf 16Published RGB image · 256 × 256 px Leaf scorch · leaf 74Published RGB image · 256 × 256 px Leaf scorch · leaf 60Published RGB image · 256 × 256 px

Values

Series	Value	Note
Healthy · leaf 49	Published RGB image · 256 × 256 px	64aea8c6-24df-40c1-9d68-0221f4151383___RS_HL 2103.JPG · SHA-256 c83901b7…2b5d.
Healthy · leaf 57	Published RGB image · 256 × 256 px	d50fa8fa-015b-41f6-a2aa-18efcf041f6e___RS_HL 2188.JPG · SHA-256 1a40422c…69c4.
Healthy · leaf 75	Published RGB image · 256 × 256 px	78debbd4-43b4-437d-8fd8-86910b947d34___RS_HL 4459.JPG · SHA-256 3904614b…ac6f.
Healthy · leaf 61	Published RGB image · 256 × 256 px	411e3372-e40e-44ed-aa47-972afabd15f7___RS_HL 2225.JPG · SHA-256 07c17d2f…24f.
Leaf scorch · leaf 69	Published RGB image · 256 × 256 px	212433a4-4bda-450e-8026-02ffa42f9f32___RS_L.Scorch 1551.JPG · SHA-256 7c4e91ed…ce81.
Leaf scorch · leaf 16	Published RGB image · 256 × 256 px	16311953-0608-43c1-829d-d78b990a0fa4___RS_L.Scorch 0995.JPG · SHA-256 5d5f1729…7c55.
Leaf scorch · leaf 74	Published RGB image · 256 × 256 px	f8c43823-8efa-4f97-8e37-8ab7e0115fd0___RS_L.Scorch 1604.JPG · SHA-256 18ac7e23…8e6f.
Leaf scorch · leaf 60	Published RGB image · 256 × 256 px	d0d0377c-6c41-4bb4-abd8-f21553d37c09___RS_L.Scorch 1459.JPG · SHA-256 041c90ad…26a.

2.9.4Split by physical leaf

Photographs of the same physical leaf stay together. The deterministic seed 20260729 assigns leaf groups to 70% training, 15% validation and 15% test partitions. This prevents near-duplicate photographs of one leaf from appearing in both training and test data.

2.9.5Image augmentation

The Python preprocessing script creates the examples below from one published leaf-scorch image. Rotation, scale, brightness, contrast, saturation, blur and sensor noise change the complete frame by a recorded amount. They do not paint new spots or remove existing tissue. Only the training partition receives random augmentation; validation and test images keep their original pixels apart from the fixed resize and normalisation. [R54]

Figure 4

Deterministic augmentation examples

The Python script applies small, recorded changes to one published image. It does not draw, remove or recolour a symptom.

OriginalNo transform GeometryRotation +6° · scale 0.96 · translation 2.5% / 2% Brightness and colourBrightness 1.07 · contrast 1.10 · saturation 0.96 Blur and sensor noiseGaussian blur 0.7 px · noise σ = 2/255

Values

Series	Value	Note
Original	No transform	Published PlantVillage RGB source image. The repository JPEG remains unchanged.
Geometry	Rotation +6° · scale 0.96 · translation 2.5% / 2%	A deterministic affine transform reproduces a small indexing and framing difference.
Brightness and colour	Brightness 1.07 · contrast 1.10 · saturation 0.96	The transform changes the complete frame uniformly; it does not add a local lesion.
Blur and sensor noise	Gaussian blur 0.7 px · noise σ = 2/255	The deterministic example represents a small optical and sensor difference.

2.9.6MobileNetV2 and the MLP classifier

The reference model receives a 224 × 224 sRGB image. MobileNetV2, initialised with ImageNet-1K weights, converts the image into a 7 × 7 × 1,280 feature map. Global average pooling produces a 1,280-value vector. The MLP maps that vector to 256 ReLU6 units, applies dropout 0.25, and produces one logit for every declared plant-health class. [R31] [R53]

Why this model fits the camera task

MobileNetV2 uses depthwise convolutions and inverted residual blocks. With the 256-unit classifier, it contains 2,586,434 parameters. The same camera is evaluated at every indexed plant position, so model size and single-image latency matter alongside classification performance. [R31]

The classifier head

Global average pooling converts the 7 × 7 × 1,280 encoder output into one 1,280-value feature vector. The multilayer perceptron learns combinations of those features that separate the declared classes. ReLU6 bounds each hidden activation between 0 and 6; dropout removes 25% of hidden activations at random during fitting. The final linear layer returns two logits, which softmax converts into class probabilities.

The 1,280 × 256 hidden layer has 327,680 weights and 256 biases. The 256 × 2 output has 512 weights and 2 biases. The MLP therefore contributes 328,450 trainable parameters.

Encoder comparison

MobileNetV2, EfficientNetB0 and ResNet50 were tested with the same frozen ImageNet protocol, 256-unit ReLU6 MLP, physical-leaf split and three deterministic seeds. All three reached 100% mean accuracy on the 199-image controlled test. The natural-background PlantDoc check separated them: mean healthy-class recall was 84.38% for MobileNetV2, 74.65% for EfficientNetB0 and 55.90% for ResNet50.

MobileNetV2 also required the fewest parameters and had the lowest measured CPU latency: 2.59 million parameters and 60.59 ms median, compared with 4.38 million and 83.35 ms for EfficientNetB0, and 24.11 million and 171.27 ms for ResNet50. Latency was measured with TensorFlow 2.18 inside a three-core Docker limit on an Intel Xeon E5-2699 v3. Processor choice changes throughput, not the mathematical definition of accuracy.

Five-fold grouped cross-validation

A second test used all 1,232 images tied to 190 physical leaves. The folds were stratified by class; every leaf appeared in one test fold and never in that fold’s training or validation data. MobileNetV2 and the MLP were refitted in every fold. Aggregate accuracy was 99.84%, balanced accuracy 99.87% and macro F1 99.83%; two of 776 leaf-scorch images were assigned healthy and all 456 healthy images were assigned correctly. Mean PlantDoc healthy recall across the five fitted models was 80.00% with a standard deviation of 8.38 percentage points.

The image classifier assigns one of its declared visual labels to the current frame. Biomass forecasting, nutrient-state estimation and stress attribution use separate models with repeated images and measured cultivation variables.

2.9.7Model fitting

1. Fit the MLP.The ImageNet encoder remains frozen while the classifier learns from the grouped training images.
2. Fine-tune the final encoder blocks.A lower learning rate adjusts the highest-level visual features; validation macro F1 controls early stopping.
3. Calibrate probability.One temperature value is fitted to validation logits after the model weights stop changing. [R55]
4. Lock the test.Architecture, preprocessing, class thresholds and the low-confidence rule are fixed before test images are opened.
5. Report each image domain separately.PlantVillage, PlantDoc and FlavoRotor results use separate tables, because their camera conditions differ.

2.9.8Controlled-image benchmark

The reproducible TensorFlow 2.18.0 run completed on 29 July 2026. After five epochs, the locked model classified all 199 images in the grouped PlantVillage test partition correctly: 72 healthy and 127 leaf scorch. The test partition contains 30 physical leaf groups that were absent from training and validation. Accuracy, macro F1 and balanced accuracy are each 1.0000 for this controlled two-class benchmark. The fitted temperature is 0.500584 and the ten-bin expected calibration error is 0.0000361.

PlantVillage photographs isolated leaves against a controlled background. The result above therefore measures discrimination between those two published Strawberry classes under the same acquisition style; it is not substituted for a measurement from the FlavoRotor camera. [R30] [R56]

Uncertainty around the measured scores

Every classification score is estimated from a finite test set. The controlled test observed 199 correct assignments from 199 images, but its exact two-sided 95% interval is 98.16–100%. Class recall has wider intervals because each class contains fewer observations: 95.01–100% for 72 healthy images and 97.14–100% for 127 leaf-scorch images. The interval states how much precision the test count provides; it does not change the observed confusion matrix.

2.9.9Natural-background check

The frozen model was then applied without retraining to all 96 images in PlantDoc's Strawberry leaf class. These photographs contain natural backgrounds, changing scale, partial leaves and varied lighting. The model assigned 79 images to healthy and 17 to leaf scorch, which gives healthy-class recall of 79 / 96 = 0.8229. PlantDoc does not publish a matching Strawberry leaf-scorch class, so this check reports recall for its healthy class rather than two-class accuracy. [R51] [R56]

Figure 8

Natural-background Strawberry images from PlantDoc

The same frozen model was applied to PlantDoc images labelled Strawberry leaf. Three low-confidence-order and three high-confidence-order examples are shown.

PlantDoc image 1Upstream label: healthy · predicted leaf scorch PlantDoc image 2Upstream label: healthy · predicted leaf scorch PlantDoc image 3Upstream label: healthy · predicted leaf scorch PlantDoc image 4Upstream label: healthy · predicted healthy PlantDoc image 5Upstream label: healthy · predicted healthy PlantDoc image 6Upstream label: healthy · predicted healthy

Values

Series	Value	Note
PlantDoc image 1	Upstream label: healthy · predicted leaf scorch	Fragaria-virginiana-6.jpg · P(healthy) 0.000000025 · SHA-256 49738f34…93b1.
PlantDoc image 2	Upstream label: healthy · predicted leaf scorch	102_0829.JPG.jpg · P(healthy) 0.000008215 · SHA-256 1e5a93bc…6097.
PlantDoc image 3	Upstream label: healthy · predicted leaf scorch	img_0164.jpg · P(healthy) 0.000009111 · SHA-256 47344a43…69a3.
PlantDoc image 4	Upstream label: healthy · predicted healthy	indian-strawberry-leaf.jpg · P(healthy) > 0.999999999999 · SHA-256 0d2b7902…d94e.
PlantDoc image 5	Upstream label: healthy · predicted healthy	Strawberry+leaves.jpg · P(healthy) > 0.999999999999 · SHA-256 368f4a9c…af54.
PlantDoc image 6	Upstream label: healthy · predicted healthy	strawberry-plant-leaves-strawberry-red-spots-on-strawberry-plant-leaves.jpg · P(healthy) 1.000000 · SHA-256 5cb64d2e…728c.

2.9.10Metric definitions

Precision answers: of the images assigned to one class, how many are correct? Recall answers: of the images that truly belong to that class, how many were found? F1 combines both values. Macro F1 gives every class the same weight, so the larger leaf-scorch class cannot hide weak performance on healthy leaves. The report also contains the complete confusion matrix, balanced accuracy, image count and physical-leaf count for every class. [R55] [R56]

2.9.11Following one plant through time

Repeated images of the same plant form a time series. Projected canopy area, calibrated colour, developmental stage and image quality are stored beside temperature, light, pH, EC and recipe version at the same timestamp. Growth rate compares the same plant at two recorded times. It does not compare unrelated plants photographed on different days. [R52]

Datasets used to test longitudinal methods

A separate model for each measured endpoint

2.9.12Stored model output

Every model output stores the image hash, plant ID, model version, preprocessing version, probability for every class, calibrated confidence and image-quality score. A low-quality or low-confidence image is marked for review together with the reason. The original image and complete class-probability vector remain available beside the final label.

2.10.1Purpose

The perimeter lighting communicates operating state, warnings, service conditions and user interactions without changing the central cultivation-light recipe. Colour assignments are interface states and are versioned in the software configuration.

2.10.2Engineering requirements

• exact LED family and revision recorded in the bill of materials;
• maximum and typical channel current measured on the installed assembly;
• voltage drop measured at the first, middle and final segment;
• surface and enclosure temperature recorded at worst-case command;
• power-supply headroom, connector rating and conductor cross-section documented;
• brightness limited for night operation and camera acquisition;
• status meanings remain accessible through text or the dashboard and do not rely on colour alone.

2.10.3Validation plan

Bench validation records current, voltage, temperature and command latency for representative patterns. Optical validation records spectrum and PPFD at plant positions with the central lamp off and on. Trial protocols state whether perimeter lighting was disabled, constant or included in the measured recipe.

2.10.4Permitted claims

The website may state that the v2.0 architecture includes addressable perimeter status lighting. It may not state that the subsystem supplements photosynthesis, improves flowering or delivers a defined wavelength until a component-specific spectrum and plant-position photon map are published.

2.11.1Scope

Industrial design is included in the research documentation because enclosure decisions can alter access, airflow, light distribution, camera visibility, contamination risk and mechanical balance. Aesthetic intent is recorded separately from measured cultivation performance.

2.11.2Layered architecture

The v2.0 concept separates a stationary external frame from the rotating cultivation layer. This allows plant movement to remain visible while fixed guards, service panels, lighting and cable routes remain referenced to the structure. The original prototype and v2.0 reports provide the primary engineering record. [I01] [I02]

2.11.3Serviceability requirements

2.11.4Verification plan

Verification includes assembly and disassembly trials, loaded rotation, splash observation, cable-clearance inspection, access-time measurement, cleaning inspection and confirmation that service work does not invalidate sensor, pump, camera or light calibration. Deviations are recorded as maintenance events in the experiment log.

3.1.1Explanation

A three-roller rotor compresses a flexible tube. Advancing the compression moves liquid while the liquid remains inside the replaceable tube. The motor provides a precise command, but only calibration determines the volume that actually exits the tube.

3.1.2Documented design

3.1.3First-order model

With d_i = 3.2 mm, L_eff = 25 mm and N_e = 3, the report model gives A_t ≈ 8.04 mm² and V_rev,ideal ≈ 0.603 mL/rev.

3.1.4Motor-command increment

3.1.5Why calibration is mandatory

Tube recovery, occlusion, viscosity, suction head, outlet pressure, roller geometry, motor torque, microstep non-linearity and tube wear all alter delivered volume. Peristaltic-pump modelling and published multi-channel systems therefore use physical calibration rather than geometry alone. [R18] [R19]

3.1.6Progress classification

The pump geometry and four-channel module are substantial v2 engineering progress. The supplied report documents CAD, component selection and the analytical model. No traceable FlavoRotor gravimetric dataset accompanies the report, so delivered-volume performance remains unclaimed until CR-PMP-001 is published.

3.1.7FlavoRotor design provenance

The system-specific configuration on this page is traced to the supplied FlavoRotor engineering records. [I02] [I03]

3.1.8CAD documentation

3.2.1Principle

Each pump dispenses into a vessel on a calibrated balance. Mass gain is converted to volume using fluid density at the measured temperature. The test is repeated across channel, dose, speed, tube condition and hydraulic head.

The gravimetric chain follows traceable liquid-volume and uncertainty principles: balance performance, test-liquid density, evaporation, timing, repeatability and calibration state are recorded. Large published peristaltic-pump datasets show why repeated measurements and drift analysis are necessary, but their performance values are not transferred to FlavoRotor. [R37] [R38] [R47]

3.2.2Test matrix

3.2.3Calibration statistics

Repeatability, bias, residual analysis and method precision are reported using declared statistical procedures rather than a single R² value. [R46] [R50]

3.2.4Channel model

3.2.5Predefined engineering acceptance gates

These are FlavoRotor acceptance criteria, not claimed achieved performance. Published multi-channel pump data guide the method but are not copied as FlavoRotor results. [R19]

3.2.6FlavoRotor design provenance

The system-specific configuration on this page is traced to the supplied FlavoRotor engineering records. [I03]

3.3.1Explanation

Four pumps allow four liquids to be added independently. The channels control liquid volumes. They do not directly control sweetness, acidity or aroma.

3.3.2Architecture

The documented motor-control architecture uses STEP/DIR microstepping drivers of the A4988 class. Driver selection defines command generation and protection requirements; it does not determine pump volumetric accuracy. [R44]

3.3.3Safe channel definition

3.3.4Dose traceability

event_id, timestamp, system_id, recipe_id, recipe_version,
channel_id, fluid_id, fluid_batch, calibration_id,
requested_volume_mL, commanded_steps, speed_rpm, direction,
reservoir_volume_before_L, pH_before, EC_before, temperature_before,
mixing_delay_s, pH_after, EC_after, fault_state

3.3.5FlavoRotor design provenance

The system-specific configuration on this page is traced to the supplied FlavoRotor engineering records. [I02] [I03]

3.3.6Module interior

3.4.1Principle

Each stock solution is defined by the concentration of every relevant chemical species, not by a marketing label. The controller calculates the amount added to the reservoir from calibrated dose volume and stock composition.

3.4.2Stock compatibility

Concentrated calcium stocks are separated from concentrated phosphate or sulphate stocks unless compatibility has been demonstrated, because precipitation can remove nutrients and obstruct tubing. Stock identity, concentration, solvent, preparation date, lot and storage conditions are recorded.

3.4.3Recipe solving

When several stocks contribute to several ions, the system solves a constrained non-negative dosing problem rather than assigning one pump to one sensory attribute.

3.4.4Ion coupling

Stock design accounts for the fact that fertilizer salts introduce coupled ions and that precise individual-ion control requires more information than bulk EC. [R34] [R35] [R36]

3.5.1Control hierarchy

3.5.2Variables suitable for direct feedback

• reservoir level or working volume;
• bulk EC, with temperature compensation;
• pH, with calibrated electrode and mixing delay;
• solution temperature;
• dissolved oxygen if a validated sensor is installed.

Sweetness, acidity of harvested tissue and aroma are not direct loop variables because they are not measured continuously in the reservoir.

3.5.3Discrete PI form

3.5.4Dose sequence

1. Validate sensor state and reservoir volume.
2. Calculate a bounded stock-volume request.
3. Verify channel calibration and stock identity.
4. Deliver dose and log actuator command.
5. Wait the measured mixing time.
6. Acquire stable pH/EC readings.
7. Apply another correction only if all limits remain valid.

3.5.5pH control note

Because pH is logarithmic and buffering varies with solution composition, the controller uses small empirical dose increments and measured response rather than converting pH error directly into a fixed acid/base volume. Open-source pH-stat work supports this calibration and logging approach. [R20]

3.6.1Mandatory interlocks

3.6.2Tube lifecycle

Tube service life is not published as a fixed 1,000-hour value unless validated for the exact tube, occlusion, fluid, speed and duty cycle. Replacement is triggered by calibrated-flow drift, visible damage, loss of occlusion, contamination risk or the validated service threshold.

3.6.3Cleaning record

cleaning_id, system_id, channel_id, cleaning_agent,
concentration, contact_time, rinse_volume, verification_method,
operator, timestamp, next_allowed_fluid_class

3.6.4FlavoRotor design provenance

The system-specific configuration on this page is traced to the supplied FlavoRotor engineering records. [I03]

4.1.1Explanation

Flavour is not one sensor value. It is the combined experience produced by taste, retronasal aroma, texture, temperature, trigeminal sensations and context.

4.1.2FlavoRotor outcome model

Discrimination, descriptive profiling and consumer liking answer different questions and are documented separately. [R25] [R29] [R40] [R41] [R42] [R48] [R49]

4.1.3Core rule

4.1.4Measurement map

Terminology and method selection follow sensory-analysis standards. A chemical difference can help explain perception, but it does not replace a sensory test. [R29] [R41] [R42] [R48]

4.1.5A measurable sensory fingerprint

A target is stored as a versioned set of measurements for one crop and cultivar at a defined harvest stage. It combines a small number of primary chemical and sensory variables with physical plant state and the complete cultivation history.

4.2.1Definition of control

FlavoRotor defines a reproducible flavour result when a requested target can be translated into a versioned cultivation recipe that produces a statistically and sensorially bounded outcome across independent cycles.

4.2.2Controllable inputs

The platform can programme light, nutrient-stock additions, pH management, bulk EC limits, solution temperature and rotation schedule; the supplied engineering reports also document the rotating drum, sensing, monitoring and proposed four-channel dosing architecture. [I01] [I02] [I03]

4.2.3Measured outcomes

Outcomes are crop-specific: basil may be evaluated through selected aroma volatiles and descriptive aroma; arugula through glucosinolate-related phytochemicals, pungency and bitterness; lettuce through bitterness, texture and quality; mint through essential-oil composition and menthol-related descriptors; strawberry through soluble solids, titratable acidity, volatile profile, firmness and sensory response. [R01] [R02] [R05] [R10] [R12] [R15] [R28]

4.2.4Recipe model

The model is trained only after single-factor and interaction experiments. It is never seeded with invented nutrient-to-flavour coefficients.

4.2.5Recipe release gate

A recipe is released only when the machine input was calibrated, the protocol was frozen before analysis, the result includes uncertainty and effect size, sensory evidence is appropriate to the claim, and at least one independent replication succeeds.

4.2.6From influence to repeatable targeting

1. Measure influence.Change one calibrated input and measure the chemical and sensory response against a matched control.
2. Map the response.Repeat across treatment levels and independent cycles to estimate direction, magnitude and interaction with cultivar and growth stage.
3. Define a target.Freeze the chemical, sensory and physical acceptance ranges before a new cultivation run begins.
4. Test prospectively.Run the frozen recipe on new biological material and compare the harvest with the predefined target.
5. Replicate.Repeat on another cycle, unit and operator with the same physical targets and calibrated local commands.

4.3.1Explanation

pH changes the chemical environment around the roots. It affects nutrient speciation, solubility, microbial conditions and uptake. It is not a direct “sweetness dial”.

4.3.2Published crop studies

Lettuce studies show that relatively small pH changes can alter physiological performance and tissue composition. [R06] [R07] These studies justify a pH trial but do not demonstrate a universal taste setting.

4.3.3Operational approach

4.3.4How to test pH and flavour correctly

Use at least three pH treatments while keeping elemental formulation, EC, DLI, temperature, harvest age and post-harvest handling matched. Measure tissue composition, biomass, chemistry and blinded sensory response. A difference in growth alone is not a flavour result.

4.4.1Explanation

EC measures how well the solution conducts electricity. It is useful for detecting overall dilution or concentration, but it cannot state how much potassium, nitrate, calcium or magnesium is present individually.

4.4.2Ion drift

Plant uptake is selective. A controller can hold bulk EC near target while individual ions diverge. This is directly documented in hydroponic nutrient-dynamics research. [R17]

4.4.3Required controls

• known initial elemental composition;
• logged stock additions and water additions;
• reservoir-volume accounting;
• periodic solution replacement or laboratory verification;
• crop-specific tissue or solution analysis for research claims;
• EC used as a bulk constraint, not an ion sensor.

4.4.4EC experiment rule

An EC treatment is reproducible only when the recipe used to reach that EC is also specified. Raising EC with a balanced nutrient concentrate is biologically different from raising EC with NaCl or a single salt, even when the meter reads the same value.

4.5.1Explanation

Adding one fertilizer changes every ion carried by that salt. A KNO₃ dose changes both potassium and nitrate; it is not a pure “sweetness” command.

4.5.2Mass balance

4.5.3Separate N, P and K limitation

The USDA Bibb lettuce dataset changes one target nutrient concentration at a time and reports fresh mass through day 32 after transplant. The day-32 response is not monotonic for every nutrient. Nitrogen rises from 1.16 g at 5 mg·L⁻¹ to 250.73 g at 132 mg·L⁻¹, then falls to 74.11 g at 264 mg·L⁻¹. Phosphorus rises from 9.14 g at 1 mg·L⁻¹ to 250.73 g at 31 mg·L⁻¹. The potassium series contains a wide interval at 42 mg·L⁻¹, so that treatment mean should not be read without its uncertainty. [R58]

The graph preserves each nutrient's actual concentration scale and the authors' 95% confidence intervals. It demonstrates two practical points: nutrient response can be curved rather than linear, and equal EC values do not imply equal elemental availability.

4.5.4Nutrient solution and tissue chemistry

El-Nakhel and colleagues tested green and red lettuce with calcium-, magnesium- or potassium-dominant macrocation ratios. The complete design contains three biological replicates in each of six cultivar-by-solution cells. The published measurements include total nitrogen, sulphate, six organic acids and total chlorophyll. [R59]

A balanced two-way fixed-effects ANOVA was recomputed from all 18 published observations. For malate, the nutrient-solution effect was F(2, 12) = 106.39 with Holm-adjusted p = 3.94 × 10⁻⁷, and the cultivar-by-solution interaction was F(2, 12) = 21.40 with adjusted p = 9.93 × 10⁻⁴. For total chlorophyll, the interaction was F(2, 12) = 25.38 with adjusted p = 4.89 × 10⁻⁴. The interaction means the solution effect changes with cultivar; one universal nutrient-to-chemistry coefficient would discard that structure.

Estimating plant nutrient state

Visible colour is useful but not chemically specific. Nitrogen limitation, water stress, senescence, exposure error and disease can all alter RGB appearance. FlavoRotor therefore joins four records at the same plant and time: the delivered elemental formulation, pH and EC history, the repeated image, and a reference measurement such as tissue mineral composition or chlorophyll. A supervised model predicts a declared laboratory endpoint, not an undefined label such as “nutrient health”.

Evaluation keeps all observations from one plant or cultivation cycle in the same fold. The report includes MAE for continuous chemistry, balanced accuracy for declared deficiency classes, calibration of uncertainty and performance for each cultivar and growth stage.

4.5.5Why EC is insufficient

EC is an indirect bulk response to all dissolved ions. Closed systems can maintain a target EC while individual nutrients become deficient or excessive. [R17] [R34] [R35] [R36]

4.5.6Four-channel implication

The four FlavoRotor channels must be assigned to chemically defined and compatible fluids. Channel labels describe the liquid, not an expected flavour. The formulation must account for coupled ions, precipitation risk, source-water composition and the limited degrees of freedom available with four reservoirs. [I03] [R34] [R35]

4.5.7Validation

Recipe trials report the full elemental formulation, source water, pH, EC, solution replacement, delivered stock volumes and tissue composition. Sensory conclusions are made only after chemical and blinded sensory measurements.

4.6.1Light variables

Hydroponic Italian Large Leaf basil provides direct evidence that lighting quality can alter key aroma volatiles. [R01]

Far-red photons can contribute to canopy photosynthesis when combined with shorter wavelengths, so FlavoRotor records them rather than reducing the light description to “red/blue percentages”. [R33]

4.6.2Experimental design

To test a spectral effect, DLI, temperature, cultivar, nutrient formulation, plant age and harvest handling remain matched. A treatment is reported as measured photon distributions, not LED control percentages.

4.6.3Required treatment definition

A light treatment is fully specified only when spectrum, PPFD, DLI, photoperiod, fixture geometry, plant position, leaf temperature and treatment timing are recorded. Percent dimmer settings are device commands, not transferable biological units.

4.7.1Scope

Root-zone behaviour is defined by more than pH and EC. FlavoRotor records solution temperature, dissolved oxygen when available, immersion timing, drainage, mixing time, reservoir volume, solution age and root architecture.

4.7.2Temperature

Root-zone temperature can alter growth and soluble-solids measurements, with cultivar-dependent responses reported in lettuce. [R11]

4.7.3Reservoir age and composition

Recycled solution can accumulate unwanted or slowly consumed ions while bulk EC remains near target. [R36]

4.7.4Sequential immersion

The rotating geometry introduces periodic root wetting and drainage documented in the FlavoRotor system records. This is treated as an experimental factor and potential confounder rather than assumed to improve oxygenation or flavour. [I01] [I02]

4.7.5Required measurements

• solution temperature at defined locations and intervals;
• reservoir volume and replacement events;
• mixing-time validation after each dose;
• root-zone exposure duration per revolution;
• root images and root dry mass;
• water and nutrient balance.

4.7.6Coupled variables

Root-zone temperature changes oxygen solubility and root metabolism; pH changes nutrient speciation and availability; EC describes bulk conductivity but not individual ions; flow and immersion determine renewal around the root surface. These variables are therefore logged together and are not interpreted independently when they covary.

4.8.1Explanation

A treatment can increase the concentration of a compound per gram while reducing total plant growth. Both outcomes must be reported.

4.8.2Response model

Mint species show species-dependent essential-oil and antioxidant responses under salinity, accompanied by growth effects. [R15]

Arugula EC trials report simultaneous changes in growth, nutritional quality and flavour-related phytochemicals, demonstrating why yield and chemistry must be analysed together. [R05]

4.8.3Measured water-stress sequence

Fevgas and colleagues published 145 hourly soil-moisture records with RGB, thermal and pseudo-colour lettuce images. The normally irrigated sequence starts at 75% and ends at 73%. The non-irrigated sequence starts at 75% and ends at 8%. A difference of at least 10 percentage points persists from 14 January 2024 at 09:24:45 (UTC+2). [R60]

The lower panel applies one fixed pixel rule to the authors' pseudo-colour outputs: R > 180, G > 180, B < 130 and R + G > 420. Yellow overlay occupies 35.13–52.28% of the detected canopy in the 13 non-irrigated outputs and 0.02–4.11% in the irrigated outputs. This is a measurement of the published visualisation, not a newly trained disease or stress classifier.

Figure 19

Published pseudo-colour outputs for irrigated and non-irrigated lettuce

The authors’ images are shown at their original aspect ratio. No generated plant image or synthetic stress region is added.

Irrigated · capture 1Published pseudo-colour output Irrigated · capture 7Published pseudo-colour output Irrigated · capture 13Published pseudo-colour output Non-irrigated · capture 1Published pseudo-colour output Non-irrigated · capture 7Published pseudo-colour output Non-irrigated · capture 13Published pseudo-colour output

Values

Series	Value	Note
Irrigated · capture 1	Published pseudo-colour output	Source file Normal_1.jpg. Full frame retained; no generated or edited plant content.
Irrigated · capture 7	Published pseudo-colour output	Source file Normal_7.jpg. Full frame retained; no generated or edited plant content.
Irrigated · capture 13	Published pseudo-colour output	Source file Normal_13.jpg. Full frame retained; no generated or edited plant content.
Non-irrigated · capture 1	Published pseudo-colour output	Source file Stressed_1.jpg. Full frame retained; no generated or edited plant content.
Non-irrigated · capture 7	Published pseudo-colour output	Source file Stressed_7.jpg. Full frame retained; no generated or edited plant content.
Non-irrigated · capture 13	Published pseudo-colour output	Source file Stressed_13.jpg. Full frame retained; no generated or edited plant content.

Separating water, nutrient and disease signals

A colour change is not assigned a cause from RGB alone. Water stress is checked against reservoir level, root-zone contact, temperature and moisture or water-potential measurements. Nutrient state is checked against the delivered formulation, pH, EC and tissue analysis. Disease labels require symptom-specific expert or laboratory confirmation. The same image can contribute features to each analysis, but each endpoint has its own reference measurement.

Time to detection

For a new cultivation run, detection time is measured from the recorded treatment change to the first alert that remains above threshold for a declared number of consecutive captures. The report includes false-alert rate in control plants, sensitivity, median detection delay and an interval across biological replicates. This distinguishes early detection from a visually strong endpoint image.

4.8.4FlavoRotor rule

No salinity or water-stress recipe is released without biomass, tissue water, visual quality, chemical endpoints and sensory confirmation. Severe stress is not used merely to create a larger analytical signal.

4.9.1Why this matters

A cultivation treatment can be overwhelmed by differences introduced during harvest, storage or sample preparation. Post-harvest conditions are therefore controlled as part of the experiment rather than treated as logistics.

Lettuce quality and shelf-life responses depend on nutrient treatment, season and post-harvest handling. [R26]

4.9.2Required harvest record

4.9.3Sensory samples

Samples receive blind random codes, balanced serving order and identical preparation. Test-room conditions and the selected discrimination, descriptive or hedonic method are documented. [R40] [R41] [R42] [R48] [R49]

4.10.1Control hierarchy

4.10.2Direct feedback loops

Closed-loop control is appropriate for directly measured variables such as pH, reservoir level, solution temperature, rotation speed and bulk EC. Individual-ion control requires ion-specific measurement or a constrained mass-balance model validated by chemical analysis. [R17] [R34] [R35] [R36]

4.10.3Learned sensory mapping

The sensory predictor is trained from completed FlavoRotor experiments. Its input data include genotype, developmental stage, measured environmental history, solution composition and harvest handling. Cross-validation is separated by cultivation cycle to prevent samples from the same run appearing in training and test sets.

4.10.4Safety constraints

• no automatic dose with an expired channel calibration;
• no correction while the mixing delay is active;
• bounded dose and runtime per event;
• sensor plausibility and redundancy checks;
• fault-safe state after communication loss;
• full event logging with recipe and firmware versions.

4.10.5State-conditioned treatment

A recipe contains time limits and plant-state conditions. For example, a light phase can begin when a validated imaging model detects the required developmental stage, provided plant-health checks pass and the minimum and maximum calendar limits are respected.

5.1.1Rotation in the plant frame

The FlavoRotor drum turns around a horizontal axis. Gravity remains vertical and close to 9.81 m·s⁻², while each plant module changes orientation relative to that vector. Rotation therefore produces a periodic directional stimulus, not reduced gravity. The encoder record links angle, angular speed and direction to every image and sensor sample. [I01] [I02] [R22] [R32]

5.1.2How a plant detects reorientation

Gravity-sensing cells contain dense, starch-rich amyloplasts. After reorientation, the amyloplasts move towards the new lower side of the cell. That physical change alters gravity signalling and directional auxin transport. Unequal growth on opposite sides of the organ produces curvature: primary roots usually bend with gravity, while shoots usually bend against it. [R63]

The biological input depends on both orientation and exposure time. A slow cycle permits a longer dwell at each angle; a faster cycle changes direction more often. Drum speed is therefore reported together with acceleration ramps, stop duration and the complete angle-time series.

5.1.3Mechanical acceleration

At these example settings, centripetal acceleration is less than 0.07% of g. The dominant physical input is the changing direction of the gravity vector in plant coordinates. Vibration, airflow, liquid movement and start-stop transients are measured separately because they can also change plant growth. [R21]

5.1.4Rotation and root-zone exposure

Drum angle also determines when each root module enters and leaves the nutrient solution. For every position, the run record stores immersion depth, immersed duration, drainage duration and retained liquid mass. A biological comparison must match average light and root-zone exposure between rotating and control plants; otherwise orientation, illumination and hydroponic contact change together.

5.1.5Controlled rotation experiment

Primary endpoints are chosen before cultivation: root-tip angle or shoot curvature for orientation response, plus one growth or chemistry endpoint. Encoder angle, three-axis acceleration, plant images, light exposure, immersion and air velocity are synchronised by timestamp. The analysis uses the plant or independent cultivation cycle as the experimental unit. [R02] [R21] [R63]

5.2.1Mechanical record

The rotation record contains commanded and measured speed, angular position, direction, acceleration and stop duration. A three-axis accelerometer fixed at the plant module records vibration and transient acceleration. Each test reports mean speed, speed ripple, peak acceleration, RMS vibration and the difference between commanded and measured angle.

5.2.2Plant exposure record

Mechanical measurements are synchronised with PPFD, air velocity and root-zone contact. One plant-position record therefore identifies the gravity direction in plant coordinates, incident light, immersion state and local air movement at the same time. This prevents a response caused by light or root-zone exposure from being assigned to rotation alone.

5.2.3Plant response

Root-tip angle, shoot curvature and elongation are measured from indexed image sequences. Growth, root architecture and plant chemistry are analysed as separate endpoints. The experimental unit is one plant or one independent cultivation cycle; repeated frames from the same plant are not counted as independent biological replicates. Gravitropic interpretation follows the measured angle-time history and the known statolith–auxin response of roots and shoots. [R21] [R63]

5.2.4Reporting

The report publishes the complete speed profile, acceleration trace, light and immersion records, sample count, biological replicate count and analysis code. Effect estimates are reported with confidence intervals. A rotation setting is identified by drum speed, direction, radius, ramp profile, operating duration and stop schedule rather than by a device preset name.

5.3.1Biological basis

Plants can change morphology, growth and metabolism in response to repeated mechanical stimulation, a field commonly described through thigmomorphogenesis. [R21]

Controlled mechanical stimulation in basil has been associated with metabolic and sensory changes, which provides a direct rationale for a FlavoRotor trial. [R02]

5.3.2What rotation may introduce

• periodic reorientation of stems and leaves;
• airflow relative to the canopy;
• vibration from drive components;
• acceleration and deceleration events;
• leaf-to-leaf or leaf-to-structure contact;
• periodic root immersion and drainage.

5.3.3Required controls

A valid experiment uses a static light-matched control, a vibration-matched control where feasible, measured airflow, identical root-zone exposure and recorded acceleration profiles. Rotation cannot be isolated by comparing two systems that also differ in light and watering.

5.3.4Mechanisms to separate

Mechanical stimulation is biologically plausible, but the published basil stimulus is not equivalent to FlavoRotor motion. The experiment must measure and match the physical stimulus before attributing a plant response to rotation. [R02] [R21]

6.1.1Selection criteria

A reference crop is selected by device fit, cycle length, commercial relevance, measurable sensory chemistry, published evidence and experimental tractability.

Basil has the strongest initial combination of short cycle, hospitality relevance and direct light/mechanical sensory literature. [R01] [R02] [R03]

Strawberry is retained as a high-value application but follows root-zone and flowering validation because cultivation-system and cultivar effects are substantial. [R12] [R13] [R14] [R28]

6.2.1Why basil first

Basil fits the device geometry, has a relatively short cycle, is used directly in premium hospitality and has published hydroponic evidence linking controlled light to aroma volatiles and mechanical treatment to sensory/metabolic response. [R01] [R02]

6.2.2FlavoRotor starting condition

6.2.3BAS-LGT-001 — spectral treatment

Use matched DLI and environmental conditions while changing a predefined spectral component. Primary outcome: selected volatile compounds by GC–MS. Secondary outcomes: fresh/dry mass, colour and blinded aroma discrimination. [R01] provides the literature anchor, not the expected FlavoRotor result. [R01]

6.2.4BAS-MEC-001 — rotation/mechanical treatment

Compare rotating and matched static controls after engineering light and root-zone equivalence has been demonstrated. Primary outcome: a predefined volatile or sensory descriptor; secondary outcomes: morphology and biomass. [R02] [R21]

6.2.5BAS-NUT-001 — nutrient formulation

Use complete elemental formulations and tissue analysis. Do not alter one stock bottle and label the response “potassium sweetness”. Cultivar, total ionic strength and nitrogen form remain explicit. [R03] [R04] [R08]

6.3.1Literature-matched starting condition

[R05] found the balance of growth and quality in the named cultivar was strongest around 1.5–1.8 mS/cm, while 2.1 mS/cm is better treated as a higher-strength experimental condition rather than a default. [R05]

6.3.2Outcomes

6.3.3Claim rule

A higher glucosinolate concentration can support a mechanism for changed pungency, but sensory testing is still required. “More phytochemical” is not automatically “better tasting”.

6.4.1Longitudinal lettuce growth

A growth model needs repeated measurements from the same plant. Karimzadeh and Ahamed's dataset links 18 identified lettuce heads to 30 daily biomass measurements, RGB canopy images and environmental readings. The sequence contains 540 biomass observations and 1,443 environmental records. [R57]

Figure 13

The same controlled cultivation study observed through time

Full-frame canopy images from four dates show why repeated, indexed imaging is more informative than unrelated photographs.

Day 1 after transplantPublished full-frame RGB canopy image Day 15 after transplantPublished full-frame RGB canopy image Day 27 after transplantPublished full-frame RGB canopy image Day 31 after transplantPublished full-frame RGB canopy image

Values

Series	Value	Note
Day 1 after transplant	Published full-frame RGB canopy image	Source file aalto-canopy-01.png. The complete frame is shown without synthetic content or symptom editing.
Day 15 after transplant	Published full-frame RGB canopy image	Source file aalto-canopy-02.png. The complete frame is shown without synthetic content or symptom editing.
Day 27 after transplant	Published full-frame RGB canopy image	Source file aalto-canopy-03.png. The complete frame is shown without synthetic content or symptom editing.
Day 31 after transplant	Published full-frame RGB canopy image	Source file aalto-canopy-04.png. The complete frame is shown without synthetic content or symptom editing.

The mean fresh biomass rises from 3.37 g on day 1 after transplant to 234.56 g on day 30. The figure also retains the daily minimum and maximum, because a mean alone hides variation between plants.

Three-day biomass forecast

The forecast uses the most recent five daily masses to estimate biomass three days later. Three models are evaluated on the same 414 cases: persistence carries the latest mass forward; a five-day line extrapolates the local trend; ridge autoregression uses five masses, four daily mass increments and day after transplant.

Each outer fold withholds one complete plant. The ridge penalty is selected inside the training fold by withholding each of the remaining plants in turn. Measurements from the evaluated plant therefore cannot choose its coefficients or regularisation strength.

The ridge model reduces mean absolute error by 83.2% relative to persistence and by 48.0% relative to the five-day linear trend. Its plant-cluster bootstrap 95% interval is 2.67–3.10 g. This benchmark measures interpolation within one published cultivation study. A FlavoRotor growth model is re-evaluated by crop, cultivar, camera geometry and cultivation cycle.

How the forecast is used

The forecast creates an expected mass and an uncertainty range for the next observation. A measured plant that repeatedly falls outside that range is inspected together with its image sequence, pH, EC, light, temperature, dose history and root-zone record. The residual identifies an unusual trajectory; it does not name the cause by itself.

Independent longitudinal datasets

HydroGrowNet follows three 30-day Batavia lettuce cycles with daily images, pH, EC and water temperature. A second multi-sensor dataset follows 45 plants over 42 days under three nitrogen concentrations and two irrigation rates, with RGB, 3D, multispectral, SPAD and fluorescence records. These datasets add camera, cultivar and treatment variation that is absent from the 18-plant forecast benchmark. [R61] [R62]

6.4.2Cultivar rule

Closed-soilless lettuce research also shows that genotype and macrocation supply interact in shaping the bioactive profile. This supports factorial crop-by-nutrient experiments rather than a universal nutrient rule. [R27]

Lettuce responses are strongly cultivar-dependent. Every trial names the cultivar and does not combine cultivars as interchangeable replicates.

6.4.3LET-PH-001

6.4.4LET-EC-001

Use a named cultivar and treatment strengths derived from [R08] or [R09] rather than a universal 1.2–1.6 mS/cm statement. In [R08], growth response differed between the tested lettuce and basil cultivars; [R09] showed functional-metabolite responses were genotype-dependent. [R08] [R09]

6.4.5LET-FLV-001

A confirmatory experiment can test the combined pre-harvest nitrogen limitation and controlled-light treatment reported by [R10]. Primary outcomes should include sensory sweetness/bitterness and the chemical variables used in the source study. [R10]

6.4.6LET-RTZ-001

Root-zone temperature is explicitly controlled because [R11] found cultivar-dependent effects on growth and °Brix. °Brix is reported as an instrumental endpoint, not automatically as perceived sweetness. [R11]

6.5.1Initial FlavoRotor screening

6.5.2Required trade-off analysis

Report essential-oil concentration, total essential-oil amount per plant and biomass. Stress can increase concentration while reducing total usable yield.

6.6.1Why strawberry is phase two

Strawberry requires a longer cycle, flowering and fruit set, stronger root-zone oxygen control, pollination management and post-harvest standardisation. A 2025 system comparison found the tested substrate system outperformed the tested water-culture systems, so the FlavoRotor root zone must be validated before flavour treatment claims. [R14]

6.6.2Two different literature anchors

[R13] reported the two-thirds treatment as the best compromise for the tested Kuemsil crop, but that result is not universal. [R13]

6.6.3Programme sequence

1. Validate survival, flowering, fruit set and root-zone oxygen under one conservative recipe.
2. Compare root support/medium configurations before nutrient-strength optimisation.
3. Test nutrient strength in one named cultivar.
4. Run an N×K factorial only after stable baseline production.
5. Measure yield, °Brix, titratable acidity, firmness, volatiles and blinded sensory profile together.

[R12] supports the N×K interaction design; [R28] supports combining volatile, quality and sensory analysis across cultivars. [R12] [R28]

7.1.1Objective

Demonstrate that one crop and cultivar can be grown repeatedly under a fixed recipe with acceptable environmental and biological variability.

7.1.2Minimum design