Draft — Correctness for Scientific Computing Systems (CS2)

MOTIVATION LETTER

The Conjunctive Consolidation Threshold model, a tripartite pharmacological framework for reward-memory encoding prevention in addiction, depends on numerical simulations that must be provably correct. My ODE/RK45 and Bayesian MCMC validation produced an encoding probability reduction from 0.855 to 0.122, an 85.8 percent decrease with super-additivity of 12.8 percentage points. If the numerical solver introduces even small errors in the stiff ODE system, the clinical trial architecture I designed on Zenodo (DOI 10.5281/zenodo.20492472) could produce false conclusions about dosing regimens. The Correctness for Scientific Computing Systems programme, jointly run by NSF and DOE, directly addresses this vulnerability.

My TOPOLOGIX platform applies persistent homology and bipartite simplicial complexes to drug-protein interaction analysis. The hERG cardiotoxicity MVP uses TDA pipelines where correctness of the filtration construction and barcode computation determines whether a compound is flagged as toxic or cleared for development. A single miscomputed persistence pair could lead to a false negative in cardiac safety screening. These are not abstract concerns. They are concrete failure modes in my daily research workflow.

I am an independent researcher based in Lagos, Nigeria, with a B.Pharm from the University of Ibadan and provisional patent protection on the CCT core architecture filed Q3 2026. My collaborators include Kent Berridge at Michigan, Samuel Gershman at Harvard who endorsed my arXiv submission, Nathaniel Daw at Princeton, and Marcelo Mattar at NYU. I seek to partner with a US-based formal methods researcher who holds an NSF grant to apply property-based testing, runtime verification, or probabilistic correctness proofs to the CCT simulation code and the TOPOLOGIX TDA pipeline.

The broader impact of this collaboration extends beyond my two platforms. Nigeria has fewer than fifty researchers working at the intersection of computational science and formal verification. Through workshops hosted at ZYCO and the University of Lagos, I will train early-career Nigerian scientists in correctness-aware scientific computing, using my open-source tools as teaching materials. The GATE platform for BCI neural-stimulation safety evaluation, released under Apache 2.0, will serve as a case study for runtime verification in neurotechnology.

This proposal matches the CS2 programme mandate to address correctness challenges that are core to DOE mission and essential to NSF mission of ensuring broad scientific reliability. My work sits at the intersection of computational pharmacology, topological data analysis, and numerical simulation verification. The programme provides the mechanism to formalize what I have already built empirically.

RESEARCH STATEMENT

The Conjunctive Consolidation Threshold model formalizes addiction as a failure of reward-memory encoding prevention. The model posits that three pharmacological mechanisms must act simultaneously at a critical consolidation window to prevent the encoding of drug-reward associations. My foundational paper on OSF (DOI 10.17605/OSF.IO/KG7B5) specifies the neurobiological basis. The formal mathematical specification (OSF 10.17605/OSF.IO/EMY4U) defines the ODE system governing dopamine, glutamate, and opioid receptor dynamics during the consolidation window. The Bayesian population dynamics paper (Zenodo 10.5281/zenodo.20492472) validates the model against clinical trial simulations.

The numerical core of the CCT model is a system of coupled nonlinear ODEs solved via Runge-Kutta 45 with adaptive step sizing. The parameter space spans 14 dimensions, and the Bayesian MCMC sampler runs 8 chains of 10,000 samples each on an HPC cluster managed through Nextflow and SLURM. All five pre-registered hypotheses H1 through H5 were confirmed, but the correctness of these results depends on the numerical stability of the solver across the full parameter range. A stiffness detection failure or an adaptive step error could shift the encoding probability threshold by enough to change clinical recommendations.

TOPOLOGIX addresses a different correctness challenge. The platform constructs bipartite simplicial complexes from drug-protein interaction data, then computes persistent homology to identify topological features that correlate with cardiotoxicity. The hERG cardiotoxicity MVP uses Ripser and Gudhi for barcode computation. The correctness challenge here is twofold: the filtration construction must preserve the underlying biological topology, and the barcode computation must be numerically stable under noise in the input distance matrix. A false persistence pair in dimension 1 could misclassify a safe compound as toxic.

The proposed collaboration with a US formal methods researcher will address three specific correctness properties. First, property-based testing of the ODE solver against known analytical solutions for reduced versions of the CCT system. Second, runtime verification of the TOPOLOGIX filtration construction using invariants derived from simplicial complex theory. Third, probabilistic correctness bounds on the Bayesian MCMC sampler using concentration inequalities.

The technical approach follows the methodology of the formal verification community. For the ODE system, I will implement a property-based testing suite in Python using the Hypothesis library, generating random parameter vectors within biologically plausible ranges and checking that the solver output satisfies conservation laws and monotonicity constraints derived from the model equations. For the TDA pipeline, I will implement runtime monitors that check the simplicial complex construction against the Vietoris-Rips axioms at each filtration step. For the MCMC sampler, I will compute empirical convergence diagnostics and compare them to theoretical bounds from the Markov chain literature.

The broader impact plan has three components. First, a two-day workshop at ZYCO in Lagos on correctness-aware scientific computing, targeting 30 early-career Nigerian researchers with backgrounds in computational biology and data science. Second, release of the correctness verification toolkit as an open-source Python package under Apache 2.0, integrated with the existing GATE platform. Third, a case study publication in a computational science journal documenting the correctness verification process for the CCT model, co-authored with the US formal methods partner.

The timeline spans 24 months. Months 1-3: establish collaboration agreement and define correctness properties. Months 4-9: implement property-based testing for CCT ODE solver and runtime verification for TOPOLOGIX TDA pipeline. Months 10-15: implement probabilistic correctness bounds for MCMC sampler and run verification suite on full parameter space. Months 16-20: conduct Lagos workshop and release open-source toolkit. Months 21-24: write case study publication and final report.

BUDGET NARRATIVE

The requested funds support a 24-month collaboration between myself as independent researcher in Lagos, Nigeria, and a US-based formal methods research group. Personnel costs cover my stipend at 3,000 USD per month for 24 months, totaling 72,000 USD. This replaces my current income as National Product Manager at Synthcare and allows full-time dedication to the correctness verification work.

Travel costs of 8,000 USD cover one round-trip flight from Lagos to the US partner institution for a two-week working visit, plus domestic travel within Nigeria for the Lagos workshop. Equipment costs of 5,000 USD cover a dedicated workstation with GPU for running the verification suite on the full CCT parameter space. Workshop costs of 10,000 USD cover venue rental, catering, and materials for the two-day Lagos workshop for 30 participants. Publication costs of 3,000 USD cover open-access fees for the case study publication. Indirect costs at 10 percent of direct costs total 9,800 USD.

Total requested amount: 107,800 USD.

BIOGRAPHICAL SKETCH

Eniola Ayodele Olutogun. B.Pharm, University of Ibadan, 2021. CGPA 5.1 out of 7.0, German equivalent 1.9. Licensed pharmacist, Pharmacists Council of Nigeria.

Independent researcher, Lagos and ZYCO, 2025 to present. Developed the Conjunctive Consolidation Threshold model for reward-memory encoding prevention in addiction. Three sole-authored preprints on OSF and Zenodo. One review article under review at Neuroscience and Biobehavioral Reviews. One co-authored paper in Alcohol, Elsevier, under review. Provisional patent on CCT core architecture, Q3 2026.

Platforms built: IMPRINT for addiction-liability screening. TOPOLOGIX for topological data analysis of drug-protein interactions, with hERG cardiotoxicity MVP. GATE for BCI neural-stimulation safety evaluation, released under Apache 2.0.

Employment: National Product Manager, Synthcare, March 2026 to present. Clinical Pharmacist, Ramset Pharmacy, January to March 2026. Research Assistant, CDDDP, NMDA and insulin docking studies. Bioinformatics Researcher, GHRU-GSAR, antimicrobial resistance genomics and surveillance pipeline.

Collaborators and endorsements: Kent Berridge, University of Michigan. Samuel Gershman, Harvard University, arXiv endorsement. Nathaniel Daw, Princeton University. Marcelo Mattar, New York University.

Skills: Python with scipy, numpy, ODE and RK45, PyMC and MCMC, pandas. R. Topological data analysis with Ripser and Gudhi. Computational neuroscience with NEURON and Brian2. Structural biology with AlphaFold, RDKit, ADMET and QSAR, GROMACS, AutoDock. High-performance computing with Nextflow, SLURM, HPC. Database and web development with Supabase, Postgres, JavaScript, Node.js.

Selected publications and preprints: Olutogun, E.A. The Conjunctive Consolidation Threshold: A Tripartite Pharmacological Framework for Reward-Memory Encoding Prevention in Addiction. OSF, 2025. DOI 10.17605/OSF.IO/KG7B5. Olutogun, E.A. Formal Mathematical Specification of the CCT Model. OSF, 2025. DOI 10.17605/OSF.IO/EMY4U. Olutogun, E.A. Bayesian Population Dynamics and Clinical Trial Architecture for the CCT Model. Zenodo, 2026. DOI 10.5281/zenodo.20492472.

DATA MANAGEMENT PLAN

Three data types will be generated during this project. First, the CCT model simulation outputs including ODE solver trajectories, MCMC posterior samples, and encoding probability estimates. Second, the TOPOLOGIX TDA pipeline outputs including persistence barcodes, simplicial complex filtrations, and classification results for the hERG cardiotoxicity dataset. Third, the correctness verification suite outputs including property-based test results, runtime monitor logs, and probabilistic correctness bounds.

All data will be stored on Zenodo with DOIs, following the existing practice established for the CCT preprints. Simulation code and verification tools will be stored on GitHub under the AmunRaPtah account, with versioned releases archived on Zenodo. The TOPOLOGIX and GATE platforms are already open-source under Apache 2.0 on GitHub.

Data will be shared at the time of publication of the case study paper. Intermediate data will be shared with the US partner via a private Supabase instance with access controls. No personally identifiable information or protected health data will be collected. All simulation data uses synthetic patient populations generated from prior distributions.

The data management plan follows the NSF guidelines for scientific computing data. Raw simulation outputs will be stored in HDF5 format. Processed data will be stored in CSV and JSON formats. Code will be stored in Python and R scripts with dependency specifications in conda environment files.

CHECKLIST

- [ ] Confirm eligibility as independent researcher not affiliated with US institution
- [ ] Identify and contact US-based formal methods researcher willing to serve as partner
- [ ] Obtain letter of collaboration from US partner institution
- [ ] Verify that NSF 24-571 allows international researchers as PI or requires US-based PI
- [ ] Prepare full proposal document with all sections as above
- [ ] Prepare budget justification with detailed cost breakdown
- [ ] Prepare biographical sketch in NSF format
- [ ] Prepare data management plan in NSF format
- [ ] Obtain current CV with all publications and preprints
- [ ] Obtain letters of support from Kent Berridge, Samuel Gershman, Nathaniel Daw, or Marcelo Mattar
- [ ] Confirm deadline date on NSF website: August 11, 2026
- [ ] Register for NSF FastLane or Research.gov account
- [ ] Verify ORCID iD 0009-0001-9272-6735 is current
- [ ] Prepare statement of current and pending support
- [ ] Confirm that provisional patent filing does not create conflict with open-source requirements

EDITOR NOTES

- Eligibility risk: NSF 24-571 typically requires US-based PI or co-PI. Eniola is an independent researcher in Nigeria. The proposal must be structured as a collaboration where the US partner is the PI and Eniola is a senior personnel or subawardee. Verify this before submission.

- The budget total of 107,800 USD is an estimate. The programme does not specify a maximum amount, but NSF grants in this area typically range from 100,000 to 500,000 USD. Confirm that the amount is appropriate for the scope.

- The provisional patent on CCT core architecture filed Q3 2026 may create complications with the open-source requirement of the NSF. Verify that the patent covers a specific implementation while the verification work covers the mathematical framework, which is not patentable.

- Eniola is not yet enrolled in an MSc programme. The profile states application for October 2026 start at MUG or Graz, Austria. If the grant starts before enrollment, the US partner must be the lead institution. If enrollment starts during the grant period, the Austrian institution may need to be involved.

- The Lagos workshop budget of 10,000 USD for 30 participants may be low. Verify local costs for venue, catering, and materials in Lagos. Consider adding a line for participant travel support if attendees come from outside Lagos.

MOTIVATION LETTER

The Conjunctive Consolidation Threshold model, a tripartite pharmacological framework for reward-memory encoding prevention in addiction, depends on numerical simulations that must be provably correct. My ODE/RK45 and Bayesian MCMC validation produced an encoding probability reduction from 0.855 to 0.122, an 85.8 percent decrease with super-additivity of 12.8 percentage points. If the numerical solver introduces even small errors in the stiff ODE system, the clinical trial architecture I designed on Zenodo (DOI 10.5281/zenodo.20492472) could produce false conclusions about dosing regimens. The Correctness for Scientific Computing Systems programme, jointly run by NSF and DOE, directly addresses this vulnerability.

My TOPOLOGIX platform applies persistent homology and bipartite simplicial complexes to drug-protein interaction analysis. The hERG cardiotoxicity MVP uses TDA pipelines where correctness of the filtration construction and barcode computation determines whether a compound is flagged as toxic or cleared for development. A single miscomputed persistence pair could lead to a false negative in cardiac safety screening. These are not abstract concerns. They are concrete failure modes in my daily research workflow.

I am an independent researcher based in Lagos, Nigeria, with a B.Pharm from the University of Ibadan and provisional patent protection on the CCT core architecture filed Q3 2026. My collaborators include Kent Berridge at Michigan, Samuel Gershman at Harvard who endorsed my arXiv submission, Nathaniel Daw at Princeton, and Marcelo Mattar at NYU. I seek to partner with a US-based formal methods researcher who holds an NSF grant to apply property-based testing, runtime verification, or probabilistic correctness proofs to the CCT simulation code and the TOPOLOGIX TDA pipeline.

The broader impact of this collaboration extends beyond my two platforms. Nigeria has fewer than fifty researchers working at the intersection of computational science and formal verification. Through workshops hosted at ZYCO and the University of Lagos, I will train early-career Nigerian scientists in correctness-aware scientific computing, using my open-source tools as teaching materials. The GATE platform for BCI neural-stimulation safety evaluation, released under Apache 2.0, will serve as a case study for runtime verification in neurotechnology.

This proposal matches the CS2 programme mandate to address correctness challenges that are core to DOE mission and essential to NSF mission of ensuring broad scientific reliability. My work sits at the intersection of computational pharmacology, topological data analysis, and numerical simulation verification. The programme provides the mechanism to formalize what I have already built empirically.

RESEARCH STATEMENT

The Conjunctive Consolidation Threshold model formalizes addiction as a failure of reward-memory encoding prevention. The model posits that three pharmacological mechanisms must act simultaneously at a critical consolidation window to prevent the encoding of drug-reward associations. My foundational paper on OSF (DOI 10.17605/OSF.IO/KG7B5) specifies the neurobiological basis. The formal mathematical specification (OSF 10.17605/OSF.IO/EMY4U) defines the ODE system governing dopamine, glutamate, and opioid receptor dynamics during the consolidation window. The Bayesian population dynamics paper (Zenodo 10.5281/zenodo.20492472) validates the model against clinical trial simulations.

The numerical core of the CCT model is a system of coupled nonlinear ODEs solved via Runge-Kutta 45 with adaptive step sizing. The parameter space spans 14 dimensions, and the Bayesian MCMC sampler runs 8 chains of 10,000 samples each on an HPC cluster managed through Nextflow and SLURM. All five pre-registered hypotheses H1 through H5 were confirmed, but the correctness of these results depends on the numerical stability of the solver across the full parameter range. A stiffness detection failure or an adaptive step error could shift the encoding probability threshold by enough to change clinical recommendations.

TOPOLOGIX addresses a different correctness challenge. The platform constructs bipartite simplicial complexes from drug-protein interaction data, then computes persistent homology to identify topological features that correlate with cardiotoxicity. The hERG cardiotoxicity MVP uses Ripser and Gudhi for barcode computation. The correctness challenge here is twofold: the filtration construction must preserve the underlying biological topology, and the barcode computation must be numerically stable under noise in the input distance matrix. A false persistence pair in dimension 1 could misclassify a safe compound as toxic.

The proposed collaboration with a US formal methods researcher will address three specific correctness properties. First, property-based testing of the ODE solver against known analytical solutions for reduced versions of the CCT system. Second, runtime verification of the TOPOLOGIX filtration construction using invariants derived from simplicial complex theory. Third, probabilistic correctness bounds on the Bayesian MCMC sampler using concentration inequalities.

The technical approach follows the methodology of the formal verification community. For the ODE system, I will implement a property-based testing suite in Python using the Hypothesis library, generating random parameter vectors within biologically plausible ranges and checking that the solver output satisfies conservation laws and monotonicity constraints derived from the model equations. For the TDA pipeline, I will implement runtime monitors that check the simplicial complex construction against the Vietoris-Rips axioms at each filtration step. For the MCMC sampler, I will compute empirical convergence diagnostics and compare them to theoretical bounds from the Markov chain literature.

The broader impact plan has three components. First, a two-day workshop at ZYCO in Lagos on correctness-aware scientific computing, targeting 30 early-career Nigerian researchers with backgrounds in computational biology and data science. Second, release of the correctness verification toolkit as an open-source Python package under Apache 2.0, integrated with the existing GATE platform. Third, a case study publication in a computational science journal documenting the correctness verification process for the CCT model, co-authored with the US formal methods partner.

The timeline spans 24 months. Months 1-3: establish collaboration agreement and define correctness properties. Months 4-9: implement property-based testing for CCT ODE solver and runtime verification for TOPOLOGIX TDA pipeline. Months 10-15: implement probabilistic correctness bounds for MCMC sampler and run verification suite on full parameter space. Months 16-20: conduct Lagos workshop and release open-source toolkit. Months 21-24: write case study publication and final report.

BUDGET NARRATIVE

The requested funds support a 24-month collaboration between myself as independent researcher in Lagos, Nigeria, and a US-based formal methods research group. Personnel costs cover my stipend at 3,000 USD per month for 24 months, totaling 72,000 USD. This replaces my current income as National Product Manager at Synthcare and allows full-time dedication to the correctness verification work.

Travel costs of 8,000 USD cover one round-trip flight from Lagos to the US partner institution for a two-week working visit, plus domestic travel within Nigeria for the Lagos workshop. Equipment costs of 5,000 USD cover a dedicated workstation with GPU for running the verification suite on the full CCT parameter space. Workshop costs of 10,000 USD cover venue rental, catering, and materials for the two-day Lagos workshop for 30 participants. Publication costs of 3,000 USD cover open-access fees for the case study publication. Indirect costs at 10 percent of direct costs total 9,800 USD.

Total requested amount: 107,800 USD.

BIOGRAPHICAL SKETCH

Eniola Ayodele Olutogun. B.Pharm, University of Ibadan, 2021. CGPA 5.1 out of 7.0, German equivalent 1.9. Licensed pharmacist, Pharmacists Council of Nigeria.

Independent researcher, Lagos and ZYCO, 2025 to present. Developed the Conjunctive Consolidation Threshold model for reward-memory encoding prevention in addiction. Three sole-authored preprints on OSF and Zenodo. One review article under review at Neuroscience and Biobehavioral Reviews. One co-authored paper in Alcohol, Elsevier, under review. Provisional patent on CCT core architecture, Q3 2026.

Platforms built: IMPRINT for addiction-liability screening. TOPOLOGIX for topological data analysis of drug-protein interactions, with hERG cardiotoxicity MVP. GATE for BCI neural-stimulation safety evaluation, released under Apache 2.0.

Employment: National Product Manager, Synthcare, March 2026 to present. Clinical Pharmacist, Ramset Pharmacy, January to March 2026. Research Assistant, CDDDP, NMDA and insulin docking studies. Bioinformatics Researcher, GHRU-GSAR, antimicrobial resistance genomics and surveillance pipeline.

Collaborators and endorsements: Kent Berridge, University of Michigan. Samuel Gershman, Harvard University, arXiv endorsement. Nathaniel Daw, Princeton University. Marcelo Mattar, New York University.

Skills: Python with scipy, numpy, ODE and RK45, PyMC and MCMC, pandas. R. Topological data analysis with Ripser and Gudhi. Computational neuroscience with NEURON and Brian2. Structural biology with AlphaFold, RDKit, ADMET and QSAR, GROMACS, AutoDock. High-performance computing with Nextflow, SLURM, HPC. Database and web development with Supabase, Postgres, JavaScript, Node.js.

Selected publications and preprints: Olutogun, E.A. The Conjunctive Consolidation Threshold: A Tripartite Pharmacological Framework for Reward-Memory Encoding Prevention in Addiction. OSF, 2025. DOI 10.17605/OSF.IO/KG7B5. Olutogun, E.A. Formal Mathematical Specification of the CCT Model. OSF, 2025. DOI 10.17605/OSF.IO/EMY4U. Olutogun, E.A. Bayesian Population Dynamics and Clinical Trial Architecture for the CCT Model. Zenodo, 2026. DOI 10.5281/zenodo.20492472.

DATA MANAGEMENT PLAN

Three data types will be generated during this project. First, the CCT model simulation outputs including ODE solver trajectories, MCMC posterior samples, and encoding probability estimates. Second, the TOPOLOGIX TDA pipeline outputs including persistence barcodes, simplicial complex filtrations, and classification results for the hERG cardiotoxicity dataset. Third, the correctness verification suite outputs including property-based test results, runtime monitor logs, and probabilistic correctness bounds.

All data will be stored on Zenodo with DOIs, following the existing practice established for the CCT preprints. Simulation code and verification tools will be stored on GitHub under the AmunRaPtah account, with versioned releases archived on Zenodo. The TOPOLOGIX and GATE platforms are already open-source under Apache 2.0 on GitHub.

Data will be shared at the time of publication of the case study paper. Intermediate data will be shared with the US partner via a private Supabase instance with access controls. No personally identifiable information or protected health data will be collected. All simulation data uses synthetic patient populations generated from prior distributions.

The data management plan follows the NSF guidelines for scientific computing data. Raw simulation outputs will be stored in HDF5 format. Processed data will be stored in CSV and JSON formats. Code will be stored in Python and R scripts with dependency specifications in conda environment files.

CHECKLIST

- [ ] Confirm eligibility as independent researcher not affiliated with US institution
- [ ] Identify and contact US-based formal methods researcher willing to serve as partner
- [ ] Obtain letter of collaboration from US partner institution
- [ ] Verify that NSF 24-571 allows international researchers as PI or requires US-based PI
- [ ] Prepare full proposal document with all sections as above
- [ ] Prepare budget justification with detailed cost breakdown
- [ ] Prepare biographical sketch in NSF format
- [ ] Prepare data management plan in NSF format
- [ ] Obtain current CV with all publications and preprints
- [ ] Obtain letters of support from Kent Berridge, Samuel Gershman, Nathaniel Daw, or Marcelo Mattar
- [ ] Confirm deadline date on NSF website: August 11, 2026
- [ ] Register for NSF FastLane or Research.gov account
- [ ] Verify ORCID iD 0009-0001-9272-6735 is current
- [ ] Prepare statement of current and pending support
- [ ] Confirm that provisional patent filing does not create conflict with open-source requirements

EDITOR NOTES

- Eligibility risk: NSF 24-571 typically requires US-based PI or co-PI. Eniola is an independent researcher in Nigeria. The proposal must be structured as a collaboration where the US partner is the PI and Eniola is a senior personnel or subawardee. Verify this before submission.

- The budget total of 107,800 USD is an estimate. The programme does not specify a maximum amount, but NSF grants in this area typically range from 100,000 to 500,000 USD. Confirm that the amount is appropriate for the scope.

- The provisional patent on CCT core architecture filed Q3 2026 may create complications with the open-source requirement of the NSF. Verify that the patent covers a specific implementation while the verification work covers the mathematical framework, which is not patentable.

- Eniola is not yet enrolled in an MSc programme. The profile states application for October 2026 start at MUG or Graz, Austria. If the grant starts before enrollment, the US partner must be the lead institution. If enrollment starts during the grant period, the Austrian institution may need to be involved.

- The Lagos workshop budget of 10,000 USD for 30 participants may be low. Verify local costs for venue, catering, and materials in Lagos. Consider adding a line for participant travel support if attendees come from outside Lagos.

AI Draft — Correctness for Scientific Computing Systems (CS2)