Pith. sign in

REVIEW 4 major objections 7 minor 28 references

Box Allocation Optimization in Meal Kit Delivery

T0 review · 4 major / 7 minor · reviewed 2026-08-05 · deepseek-v4-flash

Pith's one-line read This paper introduces the Box Allocation Problem (BAP) for daily assignment of meal-kit orders to factories and claims that an open-source MILP solver certifies optimal allocations at up to 100,000 orders in under two minutes.

desk verdict Interesting industrial application, but the MILP is under-specified as written: a_{i,j,t} is never linked to x, so the objective and the optimality claims are undefined; still, the problem is real and the fix is straightforward, so it deserves a referee. read the letter →

arxiv 2509.06157 v1 pith:JU7KIMSR submitted 2025-09-07 math.OC

classification math.OC MSC 90C1190B0690C59
keywords BoxAllocationProblemmealkitdeliverymixed-integerlinearprogrammingCBCsolvertabusearchWMAPEproductionplanningsupplychainoptimization
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper proposes a new optimization problem, the Box Allocation Problem, for assigning recipe-box orders to production factories day by day. The goal is operational stability: each factory's mix of recipes should change as little as possible from one day to the next, so ingredient forecasts stay accurate and food is not wasted. The authors model BAP as a mixed-integer linear program with capacity, eligibility, and single-assignment constraints, and benchmark the open-source CBC solver against Tabu Search and an iterative pairwise-swap heuristic. On simulated instances with up to 100,000 orders, they report that CBC reaches proven optimal solutions in under two minutes, including under changing factory capacities and changing customer orders. If correct, the paper's practical claim is that a free solver can certify optimal daily allocations at commercial meal-kit scale, without custom heuristics.

What carries the argument

The load-bearing object is the MILP formulation in Eqs. (1)-(5): binary variables x_{o,j,t} decide which factory j receives order o on day t, with constraints (2) full capacity for all factories except the catch-all, (3) eligibility (an order goes only to a factory that can make every recipe in it), and (4) one factory per order. The objective is WMAPE site—the weighted mean absolute percentage error of recipe quantities per factory between consecutive days. The paper's computational engine is the CBC solver, whose primal heuristics (Feasibility Pump, Coefficient Diving) find the optimum at the root node and whose limited branch-and-bound certifies it. Because the objective's recipe quantiti

What would settle it

Run the released code on a two-factory, two-order instance where both orders contain recipe 1 and are eligible for both factories, and print the objective for both feasible assignments. If the objective value is identical for both assignments, the model has no decision-variable dependence, and the reported optimality is vacuous. The GitHub repository named in the data-availability statement is the place to look.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central claim is that BAP is a genuine new problem—distinct from bin packing and vehicle routing because its objective is temporal—and that it can be solved exactly and quickly. The objective minimizes the site-level Weighted Mean Absolute Percentage Error (WMAPE), i.e. the sum over factories and recipes of absolute day-to-day changes in allocated recipe quantities, divided by the day's total recipe quantity. The MILP (Eqs. 1-5) controls the a_{i,j,t} quantities indirectly through binary assignment variables x_{o,j,t}, with constraints forcing full factory capacity, recipe eligibility, and exactly one factory per order. The reported experiments claim that the CB

Load-bearing premise

The load-bearing premise is that the recipe quantities a_{i,j,t} appearing in the objective are determined by the assignment variables x_{o,j,t}, even though the paper writes no equation linking them, and that each order consumes exactly one unit of factory capacity and contributes one unit of each of its recipes.

Editorial extensions

If this is right

  • If CBC solves BAP optimally at 100,000 orders in under two minutes, meal-kit operators can re-optimize allocations daily without buying commercial solvers.
  • Stable day-to-day recipe mixes make ingredient purchasing match actual production, lowering over-procurement and spoilage.
  • The proxy objective—minimizing consecutive-day changes—gives a computable route to the company's real goal of minimizing cumulative deviation from the final hard allocation at LD3.
  • The observed decrease of WMAPE with order volume means consolidation into larger allocation cycles should improve operational stability, not hurt it.
  • Under capacity shocks and order edits, an exact re-solve recovers optimality within days, whereas a greedy or ID-based carry-over rule accumulates error.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: if a_{i,j,t} were made explicit as the sum of x over orders containing recipe i, the same MILP would become a textbook assignment formulation; the practical claim then partly rests on CBC's ability to exploit that structure—an extension the paper does not state.
  • Editorial inference: BAP's consecutive-day proxy is a temporal smoothing objective; the same formulation could be applied to other perishable, multi-site production settings (bakery, pharma, fresh produce) where day-to-day recipe or SKU mix stability drives waste.
  • Editorial inference: the reported inverse scaling effect suggests testing larger horizons or longer lead times may yield further error reduction, going beyond the 15-day frame the paper considers.
  • Editorial inference: using WMAPE global as the optimality benchmark is only valid when the factory-level decomposition can be permuted freely; as stated, Eq. (8)'s triangle inequality lower bound does not certify optimality unless site-level and global objectives coincide—an assumption the experiments appear to verify but the formulation does not prove.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 7 minor

Summary. The paper introduces the Box Allocation Problem (BAP) for meal kit delivery: assign orders to factories over a 15-day horizon while minimizing day-to-day recipe allocation variation, subject to capacity and recipe-eligibility constraints. It claims to formulate BAP as a MILP in Eqs. (1)-(5), compares the CBC solver against Tabu Search and Iterative Targeted Pairwise Swap on simulated instances up to 100,000 orders, and reports that CBC consistently reaches the WMAPE global lower bound in under two minutes, including under dynamic capacity and demand changes. The central computational claim is that CBC produces provably optimal allocations at commercial scale.

Significance. If the formulation and results were correct, the paper would be a useful applied contribution: it identifies a temporal allocation problem with a clear sustainability motivation, benchmarks an exact solver against heuristics, and proposes a lower-bound optimality check. The paper also makes its data and code available on GitHub, which is a strength. However, the mathematical formulation as written is not well posed: the objective in Eq. (1) is not a well-defined function of the decision variables, is not a MILP as claimed, and the stated lower-bound argument in Eq. (8) is invalid as written. These issues are load-bearing because the optimality claims, tables, and figures all depend on them. The significance of the paper therefore cannot be assessed until the model is correctly specified and the experiments are repeated or verified under that specification.

major comments (4)
  1. [Section 2.2, Eq. (1)] The objective is not a well-defined function of x. The notation lists a_{i,j,t} and a_{i,j,t-1}, but no constraint or definition links them to x_{o,j,t}. The sentence after Eq. (5) that these quantities are 'indirectly controlled' is not a mathematical relation. With only constraints (2)-(5), f(x) is constant with respect to the decision variables, so the reported optima in Table 6 and Figures 8-19 are undefined. The authors must add linking constraints, e.g., a_{i,j,t} = sum over orders o containing recipe i of x_{o,j,t}, and state how multi-unit recipe quantities within an order are counted.
  2. [Section 2.2, Eq. (1)] Even after adding a linking definition, Eq. (1) is not a MILP. It contains absolute values and a ratio, and the denominator retains the free index j while the numerator sums over j; as written the quotient is a vector indexed by j, not a scalar objective. The paper must specify a scalar objective (e.g., sum over factories of per-factory ratios, or a single ratio with a global denominator) and describe the standard linearization using auxiliary variables. This is necessary to substantiate the claim that CBC solves a MILP.
  3. [Section 2.2, Eqs. (6)-(8) and Table 3] The claimed lower-bound relation is not established. WMAPE site as defined in Eq. (1) has denominator sum_i a_{i,j,t} for a factory j, while Eq. (8) uses the global denominator sum_i a_{i,t} and labels the ratio WMAPE site. The triangle inequality only gives |sum_j u_j| <= sum_j |u_j|; it does not yield Eq. (8) when denominators differ. Consequently, the numerical check 'WMAPE site = WMAPE global' in Table 6 is not a valid optimality certificate. If the intended site metric is actually total absolute deviation divided by global volume, the terminology and equations must be corrected consistently.
  4. [Section 4.2.1-4.2.2] The experimental setup is incompletely specified. The model defines T = {-18,...,-3}, but the benchmark and scalability tests use 'two consecutive days LD12 and LD11'; it is unclear whether T is restricted to two days and what t and t-1 denote. The reported model size of 300,600 columns for 100,000 orders and 3 factories is also inconsistent with 100,000 x 3 x 2 = 600,000 (or 4.5 million over 15 days). The authors must specify the instance generator, factory capacities, initialization of previous-day allocations, and the exact model passed to CBC for each reported run.
minor comments (7)
  1. [Section 2.1 / Notation] The set T = {-18,-17,...,-3} contains 16 days, not 15 as stated throughout. Either exclude one endpoint or correct the count.
  2. [Section 2.2] The quantities a_{i,t} and a_{i,t-1} are used in Eqs. (6)-(8) but are not defined relative to a_{i,j,t}. Add a_{i,t} = sum_j a_{i,j,t}.
  3. [Table 6] The reported improvement percentages are inconsistent with the rounded values: (0.074-0.054)/0.074 is approximately 27.0%, not 26.21% for ITPS. Recompute or state the exact unrounded values used.
  4. [References / Figures] Reference [16] is cited as an 'Internal company document' and is used for the problem definition and Figures 2-4. This is not verifiable by readers. Please replace it with public sources or provide a detailed description in the paper.
  5. [Throughout] The terms 'B&B' and 'CBC' are used interchangeably. Clarify that CBC is the solver implementing branch-and-bound, and use one consistent name for the exact method.
  6. [Section 4.2.2] No information is given about the number of heuristic runs or seed variability. Report means, standard deviations, or at least the number of repetitions for the heuristic results.
  7. [Data Availability] The paper says data are on GitHub, but the reproducibility of the experiments would be greatly improved by including the exact CBC model files or a script that generates instances and solves them.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the optimality claims rely on an external lower bound, and the cited internal source is motivational, not load-bearing.

full rationale

The paper's central claims are that BAP is a new MILP problem and that the CBC solver achieves optimal solutions, verified by equality of WMAPE site with WMAPE global. The lower-bound inequality (8) is derived from the triangle inequality, an external mathematical fact, so achieving that bound is a legitimate optimality certificate rather than a circular reduction. The objective (1) and constraints (2)-(5) do not explicitly link a_{i,j,t} to x_{o,j,t}; the text asserts they are 'indirectly controlled' after Eq. (5). This is a serious underspecification and a reproducibility flaw, but it is not a circular step: no equation is defined in terms of the result being derived. Reference [16] is an internal company document authored by co-author Loic Genest; it is used for problem motivation and figures, but the MILP formulation and solver comparison are independently presented in this paper, so the self-citation is not load-bearing for the mathematical results. The heuristic iteration counts (ITPS=1500, TS=500) were tuned on 30 generated instances of the same distribution used in evaluation; this is standard hyperparameter tuning and could inflate heuristic performance, but it does not convert a fitted input into a prediction by construction. The WMAPE global lower bound is not an input; it is computed from the same allocation variables and provides an external check. Consequently, no step in the derivation chain reduces to its own inputs, and the claimed optimality is supported by a genuine mathematical inequality.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The mathematical core has no fitted constants. The burden is in the unstated link between x and a, in the synthetic instance generator parameters, and in the tooling assumption that CBC proves optimality as implemented. The lower-bound optimality certificate is a genuine theorem and is not circular.

free parameters (5)
  • Eligibility proportions (30% F1, 60% F2, 100% F3) = 30/60/100 percent
    Synthetic order generator uses these shares; the scalability results depend on the resulting easy problem structure.
  • Factory capacity fractions (F1, F2, F3) = 25%, 50%, 25%
    Capacities are chosen to match eligibility proportions; exact solver performance depends on these instance characteristics.
  • Heuristic iteration counts = ITPS 1500, TS 500
    Tuned on 30 sets of 10,000 orders before the benchmark; heuristic quality comparisons are conditional on this tuning.
  • Recipe group boundaries and order size = Groups 1-29, 30-49, 50-89, 90-100; 1 to 4 recipes per order
    Define the BAP instances and make F3 a catch-all factory, simplifying the assignment structure.
  • Order change rates in temporal variation test = 5% deletions, 30% recipe modifications per day
    Dynamic results are specific to these assumed rates; no sensitivity analysis is reported.
assumptions (5)
  • standard math Triangle inequality: WMAPE_global <= WMAPE_site (Eq. (8))
    Used as a universal lower bound; equality of a feasible solution with this bound is a valid optimality certificate for the pooled objective.
  • domain assumption Recipe quantities are additive over orders: a_{i,j,t} = sum of x over orders containing recipe i
    This link is never written in the paper, yet the objective and all reported results depend on it.
  • domain assumption Each order consumes exactly one unit of factory capacity
    Constraint (2) equates a sum of binary order variables to an integer capacity; unequal order sizes would break the model.
  • domain assumption Eligibility is all-or-nothing: an order is allowed at a factory only if every recipe in the order is allowed there
    Constraint (3) enforces this; it is a business rule from the internal document, not a generic property.
  • domain assumption The CBC solver returns proven optima for the implemented model
    No formal certificate or machine-checked proof is provided; the lower-bound check is indirect and code is not pinned to a commit.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Box Allocation Optimization in Meal Kit Delivery." pith.science (2026). https://pith.science/paper/JU7KIMSR

@misc{pith2026250906157,
  author       = {Pith},
  title        = {Pith review of: Box Allocation Optimization in Meal Kit Delivery},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JU7KIMSR}},
  note         = {Machine review of arXiv:2509.06157}
}
abstract

This study introduces the Box Allocation Problem (BAP), a novel optimization challenge in the $1.4 billion UK meal kit delivery market. BAP involves assigning orders across multiple production facilities to minimize daily recipe variations while adhering to capacity and eligibility constraints over a 15-day planning horizon. We formulate BAP as a mixed-integer linear programming (MILP) problem and systematically compare the performance of the COIN-OR Branch and Cut (CBC) solver with heuristic methods, including Tabu Search and Iterative Targeted Pairwise Swap. Scalability experiment on instances with up to 100,000 orders show that CBC consistently achieves optimal solutions in under two minutes, maintaining optimality even under dynamic conditions with fluctuating factory capacities and changing customer orders. By reducing day-to-day recipe discrepancies, this approach supports more accurate ingredient forecasting, decreases food waste, and improves operational efficiency across multi-factory network. These results provide the first comprehensive solution framework for temporal allocation problems in meal kit delivery operations.

Figures

Figures reproduced from arXiv: 2509.06157 by the authors.

Figure 1
Figure 1. Recipe box [27] Within this operational landscape emerges the Box Allocation Problem (BAP), a critical optimization challenge that involves strategically distributing recipe boxes ( [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Recipe box allocation (a) Capacity constraint [16] (b) Eligibility constraint [16] [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Main constraints [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: Temporal aspect of BAP [16] 3 [PITH_FULL_IMAGE:figures/full_fig_p003_4.png]
Figure 5
Figure 5. Figure 5: Objective of proxy optimization BAP shares similarities with two well-known combinatorial optimization problems: BPP and CVRP. These problems rely on a network structure, where nodes represent key entities (factories, storage bins, or vehicles) and edges define decisio…
Figure 6
Figure 6. Figure 6: Similarities between BAP, BPP, and CVRP [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Types of box swaps This article first examines the application of the exact method to address BAP. The CBC solver, developed under the Computational Infrastructure for Operations Research (COIN-OR) project, offers a robust open￾source solution for MILP problems [6]. It…
Figure 8
Figure 8. Figure 8: Comparison of optimization time [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 10
Figure 10. Figure 10: WMAPEs of ITPS [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 12
Figure 12. Figure 12: Composition of total orders through days [PITH_FULL_IMAGE:figures/full_fig_p011_12.png]
Figure 13
Figure 13. Figure 13: shows WMAPE site when B&B continuously optimizes allocation between two consecutive days. For instance, LD16 is optimized based on the B&B allocation of LD17, LD15 is optimized based on LD16’s allocation, and so on. It is evident that, except for some initial days whe…
Figure 15
Figure 15. Figure 15: WMAPE site and WMAPE global through days [PITH_FULL_IMAGE:figures/full_fig_p012_15.png]
Figure 17
Figure 17. Figure 17: WMAPE site and WMAPE global with and without order changes Also in this scenario, a simple allocation strategy can be applied since the IDs of real orders remain unchanged across days. For example, if order O1 is assigned to F1 and O2 to F2 on LD14, the same assignmen…
Figure 18
Figure 18. Figure 18: B&B and ID-based allocation without or￾der changes [PITH_FULL_IMAGE:figures/full_fig_p013_18.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 22 canonical work pages

  1. [1]

    and Tecchiolli, G

    Battiti, R. and Tecchiolli, G. (1994) ’The reactive tabu search’,ORSA Journal on Computing, 6(2), pp. 126-140. Available at:https://doi.org/10.1287/ijoc.6.2.126

  2. [2]

    (2006)Primal heuristics for mixed integer programs

    Berthold, T. (2006)Primal heuristics for mixed integer programs. Diploma thesis. Technische Universität Berlin. Available at:https://www.researchgate.net/publication/258846101_Primal_Heuristics_ for_Mixed_Integer_Programs

  3. [3]

    (1999) ’Branch and bound algorithms-principles and examples’,Department of Computer Sci- ence, University of Copenhagen, pp

    Clausen, J. (1999) ’Branch and bound algorithms-principles and examples’,Department of Computer Sci- ence, University of Copenhagen, pp. 1-30. Available at:https://imada.sdu.dk/~jbj/DM85/TSPtext.pdf

  4. [4]

    Available at:https: //clearspider.net/blog/meal-kit-delivery-supply-chains/

    Clear Spider (2024)How Meal Kit Delivery Services Manage Their Supply Chains. Available at:https: //clearspider.net/blog/meal-kit-delivery-supply-chains/

  5. [5]

    and Lodi, A

    Fischetti, M., Glover, F. and Lodi, A. (2005) ’The feasibility pump’,Mathematical Programming, 104(1), pp. 91-104. Available at:https://doi.org/10.1007/s10107-004-0570-3

  6. [6]

    and Lougee-Heimer, R

    Forrest, J. and Lougee-Heimer, R. (2005) ’CBC user guide’, inINFORMS Tutorials in Operations Research, pp. 257-277. Available at:https://doi.org/10.1287/educ.1053.0020

  7. [7]

    Available at:https://www

    GeeksforGeeks (2021)How to Calculate Weighted MAPE in Excel?. Available at:https://www. geeksforgeeks.org/how-to-calculate-weighted-mape-in-excel/

  8. [8]

    1276-1290

    Gendreau, M., Hertz, A.andLaporte, G.(1994)’ATabuSearchHeuristicfortheVehicleRoutingProblem’, Management Science, 40(10), pp. 1276-1290. Available at:https://doi.org/10.1287/mnsc.40.10.1276

Show all 28 references
  1. [9]

    (2003) ’An introduction to tabu search’, in Glover, F

    Gendreau, M. (2003) ’An introduction to tabu search’, in Glover, F. and Kochenberger, G.A. (eds.) Handbook of metaheuristics. Boston: Springer, pp. 37-54. Available at:https://doi.org/10.1007/ 0-306-48056-5_2

  2. [10]

    (1986) ’Future paths for integer programming and links to artificial intelligence’,Comput- ers & Operations Research, 13(5), pp

    Glover, F. (1986) ’Future paths for integer programming and links to artificial intelligence’,Comput- ers & Operations Research, 13(5), pp. 533-549. Available at:https://doi.org/10.1016/0305-0548(86) 90048-1

  3. [11]

    (1989) ’Tabu search—part I’,ORSA Journal on Computing, 1(3), pp

    Glover, F. (1989) ’Tabu search—part I’,ORSA Journal on Computing, 1(3), pp. 190-206. Available at: https://courses.cs.umass.edu/cics521-cg/docs/tabu-glover-1.pdf

  4. [12]

    and Laguna, M

    Glover, F. and Laguna, M. (1997)Tabu search. Boston: Springer Science & Business Media. Available at: http://dx.doi.org/10.1007/978-1-4615-6089-0

  5. [13]

    Available at:https://www

    Langham Logistics (2020)The Complex Logistics Behind Meal-Kit Services. Available at:https://www. elangham.com/whats-for-dinner-the-complex-logistics-behind-meal-kit-services/

  6. [14]

    and Nobert, Y

    Laporte, G. and Nobert, Y. (1987) ’Exact algorithms for the vehicle routing problem’,Annals of Discrete Mathematics, 132, pp. 147-184. Available at:https://doi.org/10.1016/S0304-0208(08)73235-3

  7. [15]

    (1965) ’Computer solutions of the traveling salesman problem’,Bell System Technical Journal, 44(10), pp

    Lin, S. (1965) ’Computer solutions of the traveling salesman problem’,Bell System Technical Journal, 44(10), pp. 2245-2269. Available at:https://doi.org/10.1002/j.1538-7305.1965.tb04146.x

  8. [16]

    (2024)Internal material

    Loic, G. (2024)Internal material. Internal company document

  9. [17]

    and Vigo, D

    Lodi, A., Martello, S. and Vigo, D. (1999) ’Heuristic and metaheuristic approaches for a class of two- dimensional bin packing problems’,INFORMS Journal on Computing, 11(4), pp. 345-357. Available at: https://doi.org/10.1287/ijoc.11.4.345

  10. [18]

    and Toth, P

    Martello, S. and Toth, P. (1990)Knapsack Problems: Algorithms and Computer Implementations. Wiley. Available at:http://old.math.nsc.ru/LBRT/k5/knapsack_problems.pdf. 14

  11. [19]

    (1993) ’Metastrategy simulated annealing and tabu search algorithms for the vehicle routing problem’,Annals of Operations Research, 41(4), pp

    Osman, I.H. (1993) ’Metastrategy simulated annealing and tabu search algorithms for the vehicle routing problem’,Annals of Operations Research, 41(4), pp. 421-451. Available at:https://doi.org/10.1007/ BF02023004

  12. [20]

    and Krause, A

    Paulus, M.B. and Krause, A. (2023) ’Learning To Dive In Branch And Bound’,arXiv preprint arXiv:2301.09943. Available at:https://doi.org/10.48550/arXiv.2301.09943

  13. [21]

    (1939)Taboo

    Radcliffe-Brown, A.R. (1939)Taboo. Cambridge: Cambridge University Press

  14. [22]

    and Sanders, N.R

    Reid, R.D. and Sanders, N.R. (2012)Operations Management: An Integrated Approach. 5th edn. Wiley

  15. [23]

    Available at:https://rmsomega

    RMS Omega (2024)4 Challenges For Meal Kit Delivery Operations. Available at:https://rmsomega. com/4-challenges-for-meal-kit-delivery-operations/

  16. [24]

    Available at:https://www.statista.com/topics/8895/ meal-kits-in-the-uk/

    Statista (2024)Meal Kits in the UK - Overview. Available at:https://www.statista.com/topics/8895/ meal-kits-in-the-uk/

  17. [25]

    and Vigo, D

    Toth, P. and Vigo, D. (eds.) (2002)The Vehicle Routing Problem. SIAM Monographs on Discrete Mathe- matics and Applications. Available at:https://api.semanticscholar.org/CorpusID:209742615

  18. [26]

    Available at:https://unfccc.int/news/ food-loss-and-waste-account-for-8-10-of-annual-global-greenhouse-gas-emissions-cost-usd-1-trillion

    UN Climate Change News (2024)Food loss and waste account for 8-10% of annual global green- house gas emissions; cost USD 1 trillion annually. Available at:https://unfccc.int/news/ food-loss-and-waste-account-for-8-10-of-annual-global-greenhouse-gas-emissions-cost-usd-1-trillion

  19. [27]

    Available at:https://woop.co.nz/

    Woop (2024)Home. Available at:https://woop.co.nz/

  20. [28]

    and Al-Kinani, A

    Zangl, G., Graf, T. and Al-Kinani, A. (2006) ’Proxy Modeling in Production optimization’, Paper presented at theSPE Europec/EAGE Annual Conference and Exhibition, Vienna, Austria, 12-15 June. Available at: https://doi.org/10.2118/100131-MS. 15

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.