The evidence behind the candidates.

Candidate decision-method contribution; novelty unresolved. Proposed work; no experimental result is assumed.

The unresolved question

Can separate tests of process liquid and reused gypsum seeds choose corrective actions more economically than calibrated crystallization kinetics and a tuned operating rule?

What earlier work establishes

Antiscalant removal, seeded precipitation, seed reuse, kinetic diagnosis, dual-probe monitoring and seed attrition have precedents. The proposed contribution is the incremental value of a specific crossed assay in an operating decision. N03 N04 N05 N06 N07 N08 N09 N10 N11 N01 N02

The comparison

• C0: tuned host treatment and seed-management rule
• C1: calibrated kinetics and practical seed-state rule
• T: C1 with the crossed liquid/seed assay

A crossed liquid-by-seed response informs a frozen action table. Its own water, analysis, delay and handling costs are charged to the proposed policy.

What determines the decision

Operating burden per qualifying volume of treated concentrate at matched throughput, plus downstream RO compatibility.

72 development assay vessels; 36 confirmation campaigns of three decision cycles; 216 diagnostic vessels within the proposed-policy campaigns; 18 downstream RO compatibility runs. 1,368 principal panel equivalents.

Feed and reference assay

Use one gypsum-limited inland brackish RO concentrate family, rather than pooled seawater and inland feeds. Verify the limiting solids, major ions, alkalinity, additives and competing fouling. Cross matched reference/process liquid with fresh/reused seeds. Match calcium and sulfate activities, major-ion matrix, pH and temperature; reference preparation must not silently alter the causal factor. Measure seed mass, size and effective area. Include unseeded reference/process controls and transfer/carryover checks. A washed seed may lose the surface condition under study. A low growth response is an operational association until validated mechanistic controls distinguish inhibition, aggregation and active-area loss.

Development and decision logic

Use three independently collected development lots and four controlled response families: reference behavior, liquid-associated inhibition, seed-associated deterioration, and their combination. Each family receives four crossed cells and two unseeded controls, giving 72 assay vessels. The perturbations and concentrations are qualified by the host and are not claimed to represent field prevalence. Obtain three chemical panel equivalents per vessel to estimate growth response over a qualified interval. Fit a parsimonious action table using liquid effect, seed effect, interaction and uncertainty. If the assay is indeterminate, fall back to C1. If realistic states favor the same action, or assay uncertainty/delay prevents useful discrimination, stop diagnostic development.

Strongest feasible comparison

Give C0, C1 and T the same permitted actions: continued operation, qualified inhibitor-removal treatment, seed purge/makeup, and feasible conditioning or attrition. Include an action only when the host can implement it safely and compare it fairly. Calibrate C1 against the kinetics and practical seed-management precedents, rather than an arbitrary fixed dose. Make routine sensors and operating limits identical. Only T receives the extra crossed-assay information, and pays its full burden. Attrition is an established alternative to investigate, not an invention claimed by the program.

Independent confirmation and quantities

After freezing all policies, use six later feed collections, two independently established seed histories per collection and three policies: 36 campaigns. Each campaign performs three successive decision cycles, preserving its inventory. Lots are the highest-level blocks; the two seed histories are nested units, not twelve unrelated feeds. Randomize policies/order and reactor assignment within blocks. T has 12 campaigns, and each decision uses four crossed cells plus two unseeded controls: 216 additional diagnostic vessels. Allocate four process panels to each of 108 process cycles, and three to each diagnostic vessel. Finish with 18 separate downstream RO compatibility runs, one per feed-lot/policy combination, with four panels each. Principal capacity is 72×3 + 108×4 + 216×3 + 18×4 = 1,368. The downstream sample rule is fixed before selection; do not choose only the best campaign.

Endpoints and decision

The primary contrast is T versus C1 in variable operating burden per qualifying treated-concentrate volume at fixed throughput. Record physical amounts and delay alongside cost, including all assay, reagent, seed, energy, sludge, cleaning, reject and downtime burdens. Advance only when uncertainty supports the host-agreed useful margin and all service/residual limits pass. Report C0 secondarily. Report actual and membrane-specific normalized RO flow/salt passage at frozen reference conditions, stage by stage where applicable; do not normalize away a real osmotic-pressure benefit. The public ASTM scope and LG bulletin guide method selection, but do not establish full ASTM compliance. A three-cycle campaign is an initial decision test, not a 30-cycle reuse or lifetime claim.

Resources and work sequence

Use a 12-funded-month planning envelope: months 1–3 qualify feed, references and actions; 4–5 develop policies; 6–9 complete confirmation; 10–11 test downstream compatibility; 12 independent review. Use crossed-assay and conventional seed-management apparatus. Include nine collection events, seed/reference consumables, additional chemical panels and specific assay-qualification work. The scientific lead must revise capacity if reaction times, variance, matrix-specific analyses or apparatus scheduling exceed this ceiling.

Itemized costs and common assumptions

Candidate control contribution; novelty unresolved. Proposed work; no experimental result is assumed.

The unresolved question

Does a consistency check across flow, pressure and salt measurements improve accepted-water operation under unseen faults beyond a tuned normalization-based controller?

What earlier work establishes

RO normalization, predictive maintenance and automatic quality diversion are established. A prediction score or the use of AI is not the proposed contribution. N01 N02 N09 N12 N32 N33 S06 S07

The comparison

• C0: tuned operating rules
• C1: mechanistic, normalization-based control with fault fallback
• T: C1 plus the prespecified joint consistency rule

A frozen joint consistency rule decides whether the controller can trust its state estimate or must use a conservative fallback.

What determines the decision

Qualifying delivered water per calendar operating time under the same quality and resource limits.

18 development runs and 54 held-out confirmation runs across three controllers, three challenge families and six feed/time blocks; 288 principal panel equivalents.

Specific intervention

Fit the conventional process model on development data. Compute prespecified residuals between measured and predicted salt/flow/pressure behavior and their uncertainty. T uses their joint consistency, rather than a single alarm, to accept the estimate, continue the qualified policy or abstain to the same safe fallback available to C1. Freeze residual definitions, thresholds, persistence, recovery rules and permitted setpoints. Both C1 and T have the same instruments, quality limits, pressure ceilings and opportunity to tune fault handling. Do not suppress a competent baseline feature simply to create a difference.

Development and confirmation

Use three complete development blocks with six runs each: 18 runs for model/threshold qualification, including matched reference and fault cases. On six later independent feed/time blocks, compare three controllers under three families: routine qualified variation, sensor drift/dropout, and a feed-composition shift that can fool a salt-only proxy. This gives 54 confirmation runs, nominally six hours each. Reset/verify carryover, balance apparatus assignment and randomize within blocks. A hazardous contaminant is not required; use the host-approved safe matrix and analytical challenge. Historical replay can reject a weak policy but is not causal proof of a changed physical plant. Physical confirmation supplies that comparison.

Measurements and analysis

Allocate four principal panels per development/confirmation run: 72×4 = 288, plus the proposed separate commissioning/blind/QC and nonsalt-method allowances. Choose feed, accepted product and rejected/residual timing during method qualification. Independently measure quality at the delivery boundary and count false acceptance, good-water loss, cleaning, interventions, electricity and calendar downtime. Primary T–C1 comparison is accepted service with no deterioration beyond prespecified absolute quality-risk and resource limits; a lower failure count is not a claim of zero risk. Analyze six blocked contrasts, not sensor samples. Freeze the smallest useful service gain and admissible quality margin with the host; power may require more blocks.

Stop and resource gate

Stop if C1 already provides equivalent qualified service, reference data cannot identify failures, or the joint residual method merely replicates existing practice without a useful unresolved comparison. Use a 12-month staffing and equipment envelope, nine feed/data collection allowances, independent controller audit and feed-specific analytical qualification. No autonomous control of a real production site or certification of reuse water is included.

Itemized costs and common assumptions

Comparative validation and operating-envelope research. Proposed work; no experimental result is assumed.

The unresolved question

Which qualified membrane and source/sink conditions can deliver the same useful water and cooling service at a lower whole-system burden?

What earlier work establishes

Waste-heat desalination and Janus membrane architectures are established. Combined oil/surfactant exposure also appears in later public research; it is not by itself an originality claim. N13 N14 N15 N24 S24 N34 S09 S22

The comparison

• Qualified conventional MD material
• Best feasible electric treatment plus equal cooling service

Measure the boundary at which a material’s stability, absolute throughput, cleaning burden and cooling demand change the preferred process.

What determines the decision

Accepted water and equal cooling service with absolute flux, maintenance, heat and electricity separately accounted.

48 material screens, 72 equal-service comparisons and six 30-cycle endurance campaigns; 972 principal panels.

Contribution and prior-art test

Treat the material screen as qualification/validation. The hypothesis is a condition-dependent process ranking, not a new coating, first waste-heat apparatus or first combined-contaminant challenge. Register a source/sink/service cell where uncertainty could plausibly reverse the choice. If no useful uncertainty remains after the desk model and literature comparison, do not buy a second demonstration rig. A hydrogel-modified PTFE preprint reports a combined synthetic challenge; its selected results and figure were checked, but that does not establish field lifetime, general organic purity or the suitability of its chemistry here.

Methods proposed and bounded

Use the detailed C methods in this protocol: four obtainable architectures, clean/surfactant/oil/combined challenges, three independent preparations, later equal-service feed lots and separate endurance collections. Compare absolute accepted output as well as normalized stability. Select one qualified MD material on development data and freeze it before equal-service confirmation. Six endurance campaigns are a ceiling; if the selected and conventional materials coincide, use the defined narrower branch and make no distinct-material claim. Apply release and cleaning gates before material selection. The proposed >5% modeled cost advantage remains a practical decision margin for host review, not proof of scientific novelty.

Shared work and stop

Share the source/sink feasibility model and, where practical, apparatus qualification with G. Charge shared equipment, founder time and analytical development once in a combined award. Keep C’s material/treatment comparison distinct from G’s cooling-service study and disclose shared observations. If a conventional treatment delivers equal service as well or better, select it and publish the operating boundary. No laboratory result is converted into a data-center partnership or river-restoration claim.

Implementation: Material qualification

Use four architectures where obtainable: conventional hydrophobic, omniphobic, a hydrophilic surface on a hydrophobic substrate, and a hydrophilic surface on an omniphobic substrate. Obtain commercial or partner-supplied specimens with preparation and composition records; custom membrane synthesis is outside this estimate. Use three independently prepared specimens per architecture/challenge. Four architectures × four challenges × three preparations gives 48 separate screening runs. Randomize run order within preparation and feed blocks. Keep membrane area, hydraulics, feed composition and thermal conditions matched; report thickness and pressure drop. Fix feasible screening temperatures after source/sink commissioning; do not assume the paper’s operating point represents a low-grade heat source.

Implementation: Challenges and material gate

Use a clean matrix, surfactant alone, oil alone, and combined oil/surfactant. A synthetic matrix is suitable for mechanism screening; three real development feed lots establish relevance and feasible spike levels. Set concentration ramps and durations during commissioning and lock them before screening comparisons; allow up to ten operating hours per screening run. The paper’s challenge concentrations are literature anchors, not environmental exposure claims. Before confirmation, qualify cleaning compatibility and composition-specific leaching, particle release and analytical detection limits. Fluorinated chemistries require an appropriate analyte list, handling and waste route; this is not approval to reproduce the paper’s coating recipe. A material that fails the agreed water-contact criteria does not advance.

Implementation: Equal-service confirmation

Select one qualified MD material and the strongest feasible electricity-powered water-treatment baseline using development data. Freeze recovery, accepted-water requirements, source heat availability, cooling duty, return-temperature limits and accounting boundaries. Use source settings of 40, 50 and 65 °C and two sink conditions chosen within the verified apparatus envelope. Three later independent feed lots × two independent apparatus/coupon preparations per lot × three source settings × two sinks × two processes gives 72 six-hour comparison runs, paired within each condition. Count the electric baseline’s independent cooling service too. Do not extrapolate an infeasible operating cell: report it as unavailable. Fresh batches and the two processes must supply the same defined service, not merely run for the same duration.

Implementation: Endurance, measurements and inference

Use three additional feed collections in three matched pairs: the selected MD material versus the strongest qualified conventional MD material. Six independently prepared campaigns each receive 30 six-hour operating cycles with a prespecified cleaning, rinse and restart procedure. Record recovery after each cleaning and cumulative accepted water per area and campaign. If the selected material is itself the conventional comparator, run three endurance campaigns and report no distinct-material comparison; the six-campaign budget is a ceiling. This 30-cycle screen does not establish field lifetime. In all stages record absolute flux, normalized flux, independent product conductivity, salt rejection, pressure, temperature, heat, auxiliary electricity, all external water and inventory changes. Report thermal kWh and electric kWh separately. Quality failures do not count as accepted production. Budget three chemistry panels per screening run, four per comparison run and three per endurance cycle: 144 + 288 + 540 = 972 principal panels. These cover feed/product/residual checks; 48 separate composition-specific release packages are an additional service allowance. Freeze the exact analytes and sampling times with the laboratory.

Implementation: Decision and resources

Advance only if the matched service and quality requirements pass and the modeled full water cost is at least 5% lower than the strongest feasible baseline under the declared assumptions, with uncertainty reported. Assess paired contrasts by independent preparation/feed block; repeated cycles are not independent replicates. Qualify this practical margin and the initial sample sizes with the scientific lead before confirmation. Otherwise report a negative or inconclusive result, or select the established process. Plan 18 funded months: months 1–4 source/sink and methods, 5–8 qualification, 9–13 comparisons, 14–16 endurance, 17–18 independent review. Use an isolated heat emulator with bypass and safe shutdown. No operating data center or reactor connection is included.

Itemized costs and common assumptions

Conditional target-specific validation; material novelty not defined. Proposed work; no experimental result is assumed.

The unresolved question

For one native feed and one product or removal service, does the proposed recovery route outperform the strongest conventional route after regeneration, material loss and finishing?

What earlier work establishes

Magnetic sorbents, selective resins, boron pretreatment and magnesium precipitation have precedents. No exact new material composition is supplied or presumed. N16 N17 N29 N30 S23 S13 S14

The comparison

• Best feasible selective resin or precipitation route
• Proposed route at the same native feed and acceptance specification

A target-specific comparison of cumulative qualifying output and total capture/regeneration/finishing burden, including lost support material.

What determines the decision

Cumulative qualifying product or treated volume, with whole-composite and target inventories and complete service cost.

Up to 24 development runs and six campaigns of up to 100 cycles; 2,202 principal panels at the full ceiling.

Target and contribution gate

Choose magnesium recovery or boron removal from assays and a receiver’s requirement; do not fund both as one study. If the candidate is itself conventional and differs only by feed application, classify the result as transfer/validation. For a material claim, disclose the actual composition, support fraction, accessible preparation and regeneration mechanism, then compare each element against the magnetic-particle and resin precedents. No generic magnetic-bead purchase can substitute for that definition. A treatment service may have value without a saleable mineral.

Comparison and proposed methods

Use the detailed D development, paired later-feed campaigns, direct liquid/solid lithium balances and seven analytical checkpoints. Compare recoverable output per initial whole-composite mass as well as surviving material, recording attrition, carryover and replacement. Match acceptable treated volume and product requirements, not merely brine charged or percent removal. Boron pretreatment remains conditional on native chemistry; there is no automatic transfer of salt-lake results to seawater. The eight-panel checkpoint map must include actual incompatible fractions without unvalidated compositing.

Decision and overlap

Test the difference from the conventional route. A positive modeled net value under the proposed lower-value/higher-cost stress case is a commercial screening criterion, not an originality test. Stop a candidate with no attributable difference or no acceptable product/residual path. If E uses the same magnesium/boron development, make D that shared qualification package rather than repeat it as an independent discovery. Preserve six-campaign/100-cycle ceilings only for a justified distinct D question; narrow and recost shared execution before funding.

Implementation: Choose one service and develop the comparator

Collect three independent development feed lots. Measure target and competing species, salinity, suspended material and existing additives. Obtain a proposed customer’s dry-product specification or boron-removal requirement, then choose magnesium recovery or boron removal. No customer, feed supplier or sale is secured. For magnesium, screen native feed versus boron-selective-resin pretreatment and NaOH versus lime precipitation: two pretreatment states × two reagent routes × three lots × two independent preparations = 24 development runs. Tune dose, pH, residence time, settling, wash and resin conditioning within the assay-derived chemical envelope. Use appropriate speciation and blanks. If native boron does not justify treatment, omit the resin half and document the 12-run branch. For a boron-removal service, use up to 24 runs to tune the selective-resin benchmark and candidate at matched treated-water specification; do not also conduct the magnesium program.

Implementation: Confirmation and reuse

Freeze one candidate process and the strongest feasible conventional route from development. Use three later independent feed lots, with one independently prepared candidate train and one conventional train for each lot: six campaigns, three per arm. Each campaign follows up to 100 feed/regeneration cycles; preserve and record inventories between cycles. Randomize paired execution order and block by feed lot. Track breakthrough by acceptable treated volume per resin/material volume, rather than counting all charged brine as successfully treated. Match product/service specifications and report residence time, bed volumes and pressure losses. Material replacement or a failed quality gate ends the campaign; do not replace failed runs silently.

Implementation: Direct balances and sampling

Weigh every feed, reagent, recovered liquid, wet cake, wash, regeneration fraction, purge and retained inventory. Measure lithium in liquids and digested solids; wet-minus-dry cake mass is a secondary entrainment estimate, not the lithium balance. Track B, Mg, Ca, Li and introduced material with blanks, validated digestion, matrix spikes and detection limits. Require a commissioning-derived uncertainty budget; unresolved closure outside that interval prevents a recovery claim. Fully assay eight allocated stream/panel equivalents at each of 24 screening runs and at cycles 1, 5, 10, 25, 50, 75 and 100 in each confirmation campaign. Those seven checkpoints replace the routine three-panel set for that cycle. Other cycles receive three panels for feed, treated liquor and regeneration output. At the full ceiling: 24 × 8 + 6 × 7 × 8 + 6 × 93 × 3 = 2,202 principal panels. Full analytical closure is assessed at checkpoints; mass and volume inventories are recorded every cycle. The eight-panel allocation must be mapped to actual fractions and solids before execution; obtain added capacity if separate incompatible fractions require it.

Implementation: Product, water and value

Qualify dry composition and relevant impurities with solid characterization and independent checks. Count all external water: reagent preparation, resin conditioning, acid/alkali regeneration, rinsing, cleaning and retained moisture. Internal recycling is an inventory transfer; usable effluent volume is a separate denominator from feed charged. Evaluate net process value after capture, regeneration, separation, drying, finishing, residual treatment, transport, replacements and water/energy costs. Only constituents originally present in the feed can count as recovered value. Mineral sales remain zero in the research funding model; scenario value is not booked income.

Implementation: Decision and resources

Proceed only when the customer/service requirement is met and modeled net value remains positive with product/service value 30% lower and operating costs 20% higher. Report the candidate’s difference from the conventional route and uncertainty; a profitable conventional route can be the preferred outcome. Three independent campaigns per arm are an initial precision allowance, not proof of statistical power. The scientific lead must confirm the decision margin and power before confirmation. Plan 18 funded months, including development and qualification, staged cycle campaigns and independent product/value review. Six collection events, regeneration consumables, 72 solid-characterization samples and 84 special analytical packages are planning allowances. No additional target, resin lifetime beyond 100 cycles, pilot plant or market commitment is included.

Itemized costs and common assumptions

Candidate operating-rule contribution; novelty unresolved. Proposed work; no experimental result is assumed.

The unresolved question

Does an impurity- and inventory-aware wash/purge rule reduce total external freshwater use on real concentrate beyond optimized conventional recycled or countercurrent washing?

What earlier work establishes

Displacement, recycled-brine and countercurrent washing already exist, including lithium-rich brine patents and industrial practice. Two wash stages alone are not a contribution. N18 N19 N20 N29 N30 S23 S10 S11 S14

The comparison

• B0: strongest optimized conventional wash and purge policy
• T: same selected topology with the proposed inventory/impurity decision rule

On a common wash topology, a frozen inventory/impurity rule selects reuse, fresh makeup or purge while meeting the same product and residual limits.

What determines the decision

Total external freshwater makeup, product qualification, native lithium retention and downstream scale burden.

Up to 42 development runs; 24 confirmation campaigns of ten cycles; 2,820 principal panels.

Separate topology from the candidate rule

Keep the 24-run ceiling for treatment order and justified boron pretreatment. Use the 18 wash-development runs to compare optimized fresh washing, conventional displacement/recycle and established countercurrent washing. All are conventional comparators; choose the strongest feasible topology, then use that same topology for B0 and T. Fit and freeze T’s inventory balance, impurity estimate, decision interval, conservative uncertainty bounds and reuse/fresh/purge table using development data. B0 receives competent conventional impurity limits, purge practice and equal sensor access. T must add a useful decision beyond the disclosed conventional washing and recycle rules.

Confirmation and sampling

Use six later lots, two independently prepared trains per arm/lot, 24 campaigns and ten cycles per campaign. Compare the locked policies on fresh aliquots with preserved wash inventory. Hold chemistry, product specification and permitted topology equal. Allocate the same up-to-ten principal panel equivalents per development run/confirmation cycle: (42+240)×10 = 2,820. Trace the proposed direct lithium, boron, magnesium, calcium, solid and regeneration-fraction map; count reagent dilution, resin conditioning/regeneration, wash, cleaning, purge and retained liquid. An absent stream is unused capacity, not a fabricated sample.

Decision and meaningful failure

The existing >10% lower-confidence-bound freshwater reduction, >=95% native lithium retained in the designated recoverable liquid, matched product/brine specifications and downstream scale guardrail are proposed decision margins. Validate their measurability before confirmation. Infer paired effects at lot/campaign level. Ten cycles establish only the observed transient horizon; they do not prove steady-state accumulation or stable indefinite recycle. If impurity or water inventory still trends materially at the endpoint, report the transient result and recost a justified extension before a steady-state claim. If B0 is equivalent, keep B0. Keeping lithium is not producing a saleable lithium product.

Implementation: Feed, chemistry and development

Select one real SWRO concentrate source, with three independent development lots and six later confirmation lots. First compare calcium-first versus magnesium-first separation and native feed versus boron-selective-resin pretreatment, all with a common optimized fresh wash. Two orders × two boron states × three lots × two independently prepared trains gives up to 24 chemistry runs. Enable the resin branch only if native boron, speciation and the intended product requirement justify it; otherwise use the 12 native-feed runs. Do not spike confirmation feed to manufacture a boron problem. Select chemistry on qualifying products, lithium retention and total water/reagent/energy burden. Then tune three wash options at that chemistry: optimized fresh wash, a competent conventional displacement/recycled wash, and the established countercurrent wash. Three wash arms × three lots × two independent trains gives 18 additional development runs, for a 42-run ceiling.

Implementation: Apparatus and confirmation

Use two controlled stirred precipitation stages, solids separation, weighed rinse reservoirs and a matched downstream scale-challenge loop. Start with a nominal one-litre working batch, qualifying mixing and residence time. Include a small selective-resin column and conditioning/dosing connections only for the justified branch. Freeze chemistry, reagent dosing, resin operating limits, permitted wash-volume ranges, recycle boundaries, policy coefficients and cleaning rules after development. Select the strongest feasible conventional topology from fresh, displacement/recycle and countercurrent washing. Give B0 and T that same topology, chemistry, sensor access and permitted actions. B0 uses a competent conventional wash/purge policy; T adds the frozen inventory/impurity decision rule. Confirm B0 versus the inventory-aware policy on the selected conventional topology on six later lots with two independent trains per arm and lot: 24 campaigns. Each campaign processes ten fresh aliquots while maintaining its wash inventory: 240 repeated cycles. Both arms use the same selected chemistry and boron-treatment state. Cycles and aliquots within a campaign are not independent replicates.

Implementation: Measurements and water accounting

Measure external fresh-water input per litre of feed and per litre of qualifying remaining brine, with inventories corrected at the start and end. Include reagent dilution, resin preconditioning and regeneration, all washes, cleaning, purge and wet-cake water. Meter internal transfers separately and assign no water-saving credit simply for filling a reservoir. Weigh and assay lithium in every relevant liquid and digested solid; report both lithium retained in the designated recoverable liquor and whole-system closure. Measure B, Mg, Ca, introduced materials and product impurities. Report settling/filtration time, bed volumes, breakthrough, electricity and any drying burden. Do not assume a salt-lake resin flow rate works for SWRO concentrate.

Implementation: Analytical allocation

Allow up to ten chemistry panel equivalents for each of 42 development runs and 240 confirmation cycles: 2,820 principal panels. The provisional stream map is feed, remaining brine, two solids, two wash streams, rinse purge, acid regeneration eluate, alkaline conditioning discharge and final neutral rinse. Some fractions occur only at regeneration events; capacity for absent fractions is not a fabricated sample or result. Reagent-input blanks and qualification/QC are additional. Freeze event sampling and compatibility before execution; do not composite incompatible fractions to fit a budget. Obtain a validated solids digestion and an uncertainty-based closure criterion. If native lithium and the validated assay uncertainty cannot resolve the retention criterion, do not claim that it passed. A failure to close the lithium or water balance must be resolved or reported as inconclusive.

Implementation: Decision and scope

Advance only if the lower 95% confidence bound on reduction in total external fresh-water makeup exceeds 10% versus B0, at least 95% of feed lithium remains in the designated recoverable liquid with uncertainty accounted for, and both arms meet the same mineral and treated-brine specifications. The downstream scale endpoint must be no more than 10% worse than B0, or within a prespecified absolute tolerance where B0 is near zero. Use paired feed-lot contrasts and campaign-level inference; test whether the pilot variance supports these margins before confirmation. Report the conventional process as preferred if it performs as well or better. Plan 12 funded months: 1–2 feed/specifications/methods, 3–5 chemistry and wash development, 6–10 confirmation, 11–12 independent review. No mineral income or saleable lithium product is assumed. Capacity, material-release criteria and laboratory quotes must be resolved before testing.

Itemized costs and common assumptions

Candidate system-control contribution; novelty unresolved. Proposed work; no experimental result is assumed.

The unresolved question

Does a transport-aware routing rule reduce actual off-spec water entering the accepted tank beyond competent early-warning and diversion controls?

What earlier work establishes

Conductive-spacer wetting sensing, threshold alarms, salinometers, dump valves and persistence delays have precedents. Neither early detection nor automatic diversion is new as a broad concept. N21 N22 N23 N24 N32 N33 S24 N34 S06

The comparison

• C0: optimized conductivity routing
• C1: early sensing plus tuned threshold/persistence and fault fallback
• T: C1 plus a transport/inventory model

A frozen residence-time and contamination-inventory estimate controls routing and safe recovery on matched hardware.

What determines the decision

Absolute off-spec volume accepted, good water diverted and recovery time at the delivered-water boundary.

12 commissioning runs and 108 confirmation runs; 864 principal plus 96 commissioning panels.

Define what T adds

Measure the full residence-time distribution, reference sensor response, valve delay and tank carryover. T propagates contamination-risk inventory through that calibrated flow path, routes conservatively when uncertainty overlaps the acceptance limit, and resumes collection only after its prespecified flushing/recovery condition. A fixed persistence timer or a larger holding tank is not the claimed advance. Give all arms identical sensing, tank volume, valve geometry, membrane and permitted actions; C1 includes competent persistence, hysteresis and fault diversion. Lock the model, thresholds and reset criteria before confirmation.

Adversarial test

Use six challenge families: gradual wetting, abrupt integrity failure, benign feed/temperature variation, sensor drift/dropout, oil-only fouling and combined oil/surfactant exposure. Twelve commissioning runs qualify ranges; six fresh coupons per controller/family give 108 confirmation runs, blocked across later feed and membrane lots. Oil-only is a negative wetting control only if independent truth confirms it. Hold out trajectories. The separate fast reference channel, synchronized logging and chemical sampling in the detailed methods are mandatory. Eight time-point panels per confirmation run are chemistry checks; continuous reference measurements establish the transient endpoint.

Primary contrast and stop

Use T versus C1 as the primary contrast. Predeclare an absolute exposure margin, quality gate and good-water-loss allowance for the selected service. Evaluate both harmful water accepted and unnecessary diversion; a controller that dumps everything cannot win. If baseline bad-water acceptance is effectively zero, report the floor and test a prespecified noninferiority/retention question only if it was chosen before confirmation; otherwise stop or redesign. Twelve commissioning runs are an allowance, not a guarantee that the latency model can be identified. The 2017 material paper’s slow sampling cannot validate this controller, and later combined-challenge materials do not prove delivered-water protection.

Implementation: Apparatus and comparators

Use one characterized SWRO concentrate family in a contained direct-contact MD rig, nominal membrane area 0.02 m². Match flow path, active sensor availability, holding volume, valve geometry and membrane/spacer construction across controllers. C0 uses conductivity-triggered routing; C1 uses a tuned early-warning threshold plus conductivity; T adds the proposed transport-delay model and conservative fault diversion. Tune both baselines on the same development access as T. Use 12 commissioning runs, two allocated per challenge family, to qualify ranges, freeze threshold/timing rules and identify feasible exposure trajectories. Qualification may require revising this planning allocation before confirmation; never train on confirmation runs.

Implementation: Six challenge families

Confirmation covers (1) gradual surfactant-driven wetting, (2) abrupt membrane-integrity failure, (3) harmless feed or temperature variation within the qualified envelope, (4) sensor dropout or drift, (5) oil-only fouling, and (6) combined oil/surfactant exposure. Use six independently assigned fresh coupons per controller and family: six families × three controllers × six coupons = 108 runs. Block the six coupons across three later feed lots and at least three membrane preparations/lots where available; balance controllers within those blocks. Hold out complete concentration/time trajectories and feed lots from tuning. Set spike levels from the actual feed, apparatus limits and pilot response. If an intended oil-only negative control actually wets, classify it by the independent reference rather than relabeling the result to fit the hypothesis. A combined challenge is a test here, not a demonstrated result from the Janus paper.

Implementation: Independent truth and timing

Use a separate calibrated reference conductivity/quality channel and independently measured flows/tanks, synchronized with early sensors, valves and alarms. The reference must not share the proposed trigger as its truth signal. Before confirmation, measure sensor response time, sample transport delay, mixing, valve actuation and tank carryover using controlled tracer/fault checks. Choose a reference logging interval no greater than the smaller of one second and one tenth of the shortest relevant breakthrough interval; verify actual sensor bandwidth and uncertainty. If response is too slow, enlarge the protected hold interval or redesign before testing. Continuous samples are correlated observations, not replicates. Eight laboratory chemistry time points per confirmation run (864 panels) and per commissioning run (96 panels) verify reference calibration and salt/ion behavior; they do not independently establish subsecond timing.

Implementation: Water quality and endpoints

Freeze accepted-water criteria for the intended use before confirmation; a commissioning alarm such as 100 µS/cm at 25 °C is not a potable standard. Add oil/organic carryover checks appropriate to the challenge chemistry: 72 targeted packages allow two samples per oil-only or combined confirmation run. Specify analytes and sensitivity during method qualification; conductivity cannot exclude nonsalt contaminants. Compute off-specification water accepted as the time integral of accepted flow times independently determined failure status, with uncertainty and tank assays. Report contaminant mass, absolute bad-water volume, good water diverted, cumulative accepted production and absolute flux. For organic analytes without a validated continuous surrogate, conservatively bound the affected interval between chemical samples; do not claim fine-time resolution for those endpoints.

Implementation: Decision and resources

For prespecified genuine salt-breakthrough families, the lower 95% confidence bound on reduction in bad water accepted must exceed 50% versus C1. Use run-level contrasts with family and feed/material blocks; report every family separately as well as any prespecified pooled estimate. Qualifying water diverted must be no more than 5% during independently confirmed nonfailure periods. The 95th-percentile end-to-end routing latency plus measurement margin must remain below the earliest qualified breakthrough time. If C1 already prevents measurable release, use an absolute detection-limit bound and report no demonstrated percentage advantage. Do not induce a harsher fault merely to rescue that claim. Combined-exposure organic outcomes are separately bounded and cannot establish drinking-water certification. Plan 12 funded months: 1–3 qualification, 4–5 commissioning/freeze, 6–10 confirmation, 11–12 independent audit. These counts require pilot-based precision review and laboratory/equipment quotes; no live water supply is used.

Itemized costs and common assumptions

Equal-service integration validation with a conditional control question. Proposed work; no experimental result is assumed.

The unresolved question

When do source interruption, hot sinks and real demand make heat-assisted recovery better or worse than conventional cooling plus water treatment?

What earlier work establishes

Data-center heat-driven desalination, telemetry scheduling, RO preheating and seawater cooling have precedents. Building an isolated coupling is a validation task, not a first-invention claim. N13 N14 N15 N24 S09 S20 S22

The comparison

• B0: best feasible conventional cooling plus treatment
• B1: simpler recovered-heat use, including qualified RO preheating
• T: isolated MD with a prespecified bypass/dispatch rule

Map failure and benefit boundaries at equal cooling reliability and accepted-water service; advance a new controller only if an explicit difference survives existing scheduling prior art.

What determines the decision

Freshwater consumption and auxiliary electricity at equal cooling and delivered-water service, including downtime.

54 six-hour trajectory runs and three paired seven-day campaigns; 288 principal panels across 324 platform hours and 42 configuration-days.

Comparator and scope

Use dry, evaporative/hybrid and isolated seawater-cooling qualification where feasible. Freeze the strongest host-applicable B0 and the simpler B1; a wasteful evaporative baseline is not permitted when a better feasible alternative exists. Specify IT-equivalent load, allowed coolant return, reliability and water demand. Source/sink telemetry-based scheduling alone is already disclosed. T’s policy and any claimed incremental feature must be written against that precedent before confirmation. If no distinct rule remains, complete G as comparative validation only.

Methods and decision

Use three later real-feed lots, six trajectories and three configurations, then three paired seven-day B0/T campaigns. Include interruption, hot sink, restart, demand shift and zero-useful-temperature-difference conditions without perfect future knowledge. Keep separate freshwater withdrawal/consumption, saline inputs, exports, returns and source-attributed inventory. Count the emulator’s electricity separately from a modeled recovered-heat supply. The proposed >10% freshwater saving and <=5% auxiliary-electricity increase require uncertainty bounds and matched service. When B0 freshwater use is near zero, declare the savings case absent; do not subtract exported water twice.

Share or stop

Perform the C/G feasibility model once. Share compatible metering and thermal apparatus only after documenting capacity and schedule, with no duplicated costs or independent-evidence claims. If the desk calculation shows no plausible service advantage, stop the MD branch before construction and retain the result. A nominal 2-kW emulator and 42 configuration-days do not establish data-center uptime, intake/discharge acceptability, river benefit or a safe nuclear interface.

Itemized costs and common assumptions

Candidate constrained-control contribution; novelty unresolved. Proposed work; no experimental result is assumed.

The unresolved question

Can a controller that accounts for uncertain forecasts and stored moisture avoid wasted regeneration while delivering more qualifying water than a tuned adaptive controller under the same resource limits?

What earlier work establishes

Weather adaptation, condenser control, forecast-assisted energy management and battery-constrained water generation already appear in research and patents. Adding a forecast or heat budget alone is not a defensible novelty claim. N25 N26 N27 N35 N36

The comparison

• C0: tuned fixed timing
• C1: competent weather/condenser-aware, forecast-assisted feasible schedule
• T: C1 plus the registered inventory/uncertainty rule

A prespecified moisture-inventory and uncertainty-aware run/wait/stop rule is compared on equal hardware and forecast access.

What determines the decision

Qualifying collected liquid over complete 24-hour periods at identical heat, electricity and cooling limits.

Nine independent beds and 54 chamber bed-days; six paired field units over two 14-day windows, totaling 168 unit-days; 222 principal water panels.

Exact policy comparison

C1 receives the same forecasts, state sensors and resource ceilings as T and may use competent forecast scheduling and fault fallback. T’s candidate addition is a conservative bound on releasable bed moisture and expected condenser capture, with a terminal inventory condition that prevents apparent gains from merely emptying the bed or tank. Freeze the calculation, uncertainty treatment, action thresholds and fallback using development profiles. Both policies pay for sensing/computation and obey the same constraints. The precise added rule must survive full-method comparison with adaptive and forecast-control precedents; otherwise classify the work as validation.

Conditions and independent units

Use the qualified commercial-bed approach and the 54 chamber bed-days across nine independently prepared beds. Allocate the six held-out trajectories to qualified reference weather, heat interruption, hot-sink limitation, forecast timing error, gradual condenser impairment and sensor-invalid/restart conditions. Do not combine several faults and call their effects isolated. Each bed receives repeated profiles in randomized order with validated inventory reset. Empty-device/conditioning runs belong to commissioning. Use three independently assembled C1/T pairs in two contrasting 14-day field windows after the chamber gate. Beds/devices and weather blocks determine inference; 168 unit-days are not 168 independent units.

Energy, water and decision

Measure actual collected liquid, bed/condenser inventory, rejects, external cleaning water, heat and all electricity/cooling. A chamber humidifier is an external water input. Equal starting/final stored moisture and time boundaries prevent borrowing output from another period. Compare the same end-use quality and the proposed >10% lower-confidence-bound daily-output improvement target against C1, subject to host-selected useful volume and resource constraints. Zero-output days remain in the calendar denominator. The proposed 100-cycle retention screen requires its own cycle log and material measurements; it is included in conditioning/bed-life allowances, not extra independent water-production replicates or annual-life evidence.

Prior-art and stop gate

A 2026 primary publisher abstract describes weather forecasting and hierarchical AWH setpoint retrieval for a modeled refrigeration system; a 2022 thesis discusses limited battery power and proposed predictive management. These adjacent systems defeat a broad forecast-control claim but do not validate this sorbent policy. Their headline energy values are not imported. If the selected bed cannot regenerate at the available heat grade, if the conventional policy is equivalent, or if a simpler supply better serves the user, stop this branch. Two field windows do not establish Phoenix annual supply or household self-sufficiency.

Itemized costs and common assumptions