independent · donated compute · universal curative focus

Sensei Ewok Research Lab

Small tools. Careful questions. Research in the open.

What this isA small, independent software lab. We write research tools, test them in public, and say plainly what each one does and does not do.
What it is notNot a medical organization, not a clinic, and not affiliated with the Cystic Fibrosis Foundation. Nothing here is medical advice.

EXPLORE THE LAB Open the 3D view of the CFTR protein

A letter to people living with CF, and to those who care alongside them.

A still picture. The interactive 3D view needs JavaScript and WebGL2; the facts it shows are in the caption below and in the table on the model page.

Membrane-spanning part 1, residues 81 to 365
Nucleotide-binding part 1, residues 423 to 646
Membrane-spanning part 2, residues 859 to 1155
Nucleotide-binding part 2, residues 1210 to 1443
Gray: placed, outside the four marked parts
Dotted gray: not placed in this structure
Flat band: helix or strand, as UniProt annotates

CFTR is a human protein, and variants of it are linked to cystic fibrosis. Structure 6MSM places 1,181 of its 1,480 amino acids. Residue 508 is present in this structure. A model fitted to electron-microscope data, not a photograph.

Four-Pillar Lab Compass · Persona Gateways

Four pillars. One curative mission.

Navigate by your role: community orientation and global support, care-team clinic and trial pathways, discipline-indexed biophysics workbenches for CF researchers, and open-source verification for builders.

Adaptive Command Deck · All 4 Reader Pathways

Select your role above to adapt the page, filter the pathway cards, and surface direct 1-click launchers for your primary questions.

Genotype Pathway Finder (Classes I–VII) → 79 Recruiting Trials & Visit-Prep → 52-Model Curative Bottleneck Radar → 113 Sources & 138 Caught Mistakes →

01 · Patients & Families 65 Roses · 22 Countries

Sourced History, Curative Map & Community Support

From Dorothy Andersen's 1938 description to triple-combination therapy and the 6–12% still waiting. Plain-language variant classes, our Tree of Life horizon map, 22-country support directory, and AI safety boundaries.

The CF Story → Tree of Life Map → Global Resources (45) → AI in Care Safety →

02 · Care Teams & Clinicians 79 Recruiting · 57,288 Patients

Recruiting Trials, Visit-Prep & Multi-Registry Trajectories

Bilingual clinic visit-prep across 79 recruiting studies and 940 screened trials, plain-language endpoint decoders, 10-year ECFSPR and national registry modulator eligibility denominators, and clinical AI audits.

940-Trial Explorer → 10-Yr Registries → Clinic Discussion Guides → Clinical AI Audit →

03 · CF Researchers 6 Disciplines · 33 Modules

Discipline-Indexed Biophysics & Curative Workbenches

Dedicated workbenches organized across 6 scientific disciplines: Structural Biology (4 Cryo-EM PDBs & 209 variants), RNA & Gene Editing (Class I PTC & ASO splicing), Mucus Rheology, LNP/mRNA, AAV/Stem Cells, and Phage/Host Defense.

3D CFTR Channel → Nonsense PTC Sim → Airway LNP Delivery → 940-Trial Explorer →

04 · Builders & About Lab 113 Sources · 138 Caught · MIT / CC0 / CC BY 4.0

Zero-Hallucination Verification, Council Loop & Boundaries

Every public number rests on a primary source in our 113-entry catalog and a SHA-256 lockfile. Inspect our 138-row mistake ledger, multi-agent council benchmarks, T-0105 review record, and what the site collects (nothing).

Evidence & 138 Ledger → Open Tools & CLI → AI Loop & Council → Mission & Review Record → Contact Boundary →

FLAGSHIP DISCOVERY · ALPHAEVOLVE-CF MULTI-OBJECTIVE PARETO & FORMAL INVARIANT ENGINE (T-0426) Tier T1 Exploratory Mechanistic Simulation · Deb NSGA-II + Saltelli-Jansen QMC Sobol + WebGPU WGSL Compute + SMT Prover

Why Single-Barrier Therapies Stall at 0.65% WT — and How Evolutionary Phase-Sequencing Unlocks 6.86x Epistatic Rescue

Quasi-Monte Carlo Sobol analysis (Saltelli-Jansen) demonstrates that 88.2% of Class I CFTR rescue variance depends on temporal coupling across 6 serial barriers. Our client-side NSGA-II solver and formal SMT verifier identifies a conflict-free 24-hour sequence predicting 17.84% WT chloride flux in silico with 0 modeled safety violations.

7% Hypertonic Saline → rhDNase I Lag (h) 0.85 h

Post-15 Hz HFCWO Vest → LNP/AAV Inhalation Window (min) 12 min

SMG1i NMD Escape + sup-tRNA Readthrough Dose 45%

Apical ENaC Clamp + ETI Fold-Potentiation 60%

Restored Chloride Flux (ΔIsc) 17.84% WT

Epistatic Synergy Ratio (vs 2.60% Additive) 6.86x (88.2% Epistatic)

Thixotropic Mucin Mesh Pore ξ(t) 105.4 nm (67% perm)

Periciliary ASL Depth & ΔLCI 2.5 9.05 µm (ΔLCI -2.74)

Ribosomal TRSS Safety & CYP3A4 Load TRSS 87.7% · CYP 0.57x

Formal SMT Proof Status (4 Invariants) PROVEN SAT (0/4 Violations)

INV-01 · rhDNase I Ionic Protection (Δt ≥ 0.75 h) PASS · rhDNase 95% active

INV-02 · HFCWO Thixotropic Pore Window ([5, 25] min) PASS · 105.4 nm pore window

INV-03 · Housekeeping Stop-Codon Safety (TRSS ≥ 80%) PASS · TRSS ≥ 80% safe

INV-04 · CYP3A4 Ceiling (≤ 1.0x) & ASL Floor (≥ 7.0 µm) PASS · ASL ≥ 7.0 µm & CYP ≤ 1x

WebGPU WGSL @compute (workgroup_size=64) + Float64Array · N=256 QMC CI [15.3%–20.5% WT]

NSGA-II 2D Pareto Frontier & Sobol Variance Partition (88.2% Epistatic)

Schematic, not a structure · Click any green Pareto node to load regimen · Ribbon: N=256 QMC P10–P90 CI · Tier T1 Exploratory Simulation (Not medical advice)

24-Hour Conflict-Free Circadian Regimen & LNP Permeation Window

Schematic, not a structure · Grounded in NCT05668741, NCT05248230, DOI: 10.1126/science.1223012 & DOI: 10.1038/s41598-025-16178-z

Interactive WebGL2 Biophysical Mucin-Mesh Permeation & 3D Pareto Epistatic Surface

Switch between real-time #version 300 es shader simulation of 72 nm LNP/AAV nanoparticles diffusing across thixotropically widened mucin pores into the ciliated periciliary brush, or the rotatable 3D 3-objective Pareto surface (ΔIsc vs TRSS vs CYP3A4).

WebGL2 #version 300 es · Pore ξ=105.4 nm (67% LNP Permeation) · Periciliary h_ASL=9.05 µm · ΔIsc=17.84% WT

3D Turn (Yaw) 3D Tilt (Pitch)

6-Barrier Serial Epistatic Survival Waterfall Prover (Why Monotherapy Stalls at 0.65% WT vs 17.84% WT Sequenced)

All 6 Serial Gates Aligned · Cumulative Yield 2.40% → 17.84% WT (6.86x Synergy)

Because airway and intracellular delivery barriers act in series (∏ η_k), leaving any single gate unconditioned collapses overall CFTR rescue to the 0.65% WT baseline. Move the sliders above to see which serial barrier becomes the active bottleneck.

B1 · rhDNase I Ionic Cleavage (η_DNase)95.3% (vs 8.3% simultaneous)

Δt = 0.85 h lag avoids 7% NaCl divalent Mg2+/Ca2+ enzyme inhibition

B2 · Thixotropic Mucin Mesh Sieving (η_mucus)67.2% (vs 1.5% unconditioned)

Pore ξ = 105.4 nm vs 72 nm vector at 12 min post-15 Hz HFCWO shear

B3 · SMG1i NMD mRNA Escape (β_NMD)51.7% (vs 18.0% unshielded)

Sub-toxic SMG1i (45%) preserves PTC mRNA pool (TRSS 87.7%)

B4 · Sup-tRNA PTC Readthrough (ε_RT)23.6% (vs 2.0% basal)

Engineered sup-tRNA (45% cargo dose) outcompetes eRF1/eRF3 stop termination

B5 · ETI Co-Translational Folding (η_fold)68.3% (vs 19.8% unassisted)

VX-445/VX-661 (3.45x fold) stabilizes near-cognate Trp/Cys readthrough channels

B6 · Apical Gating & ENaC Hydration (P_open)45.0% · ASL 9.05 µm

VX-770 gating + 60% ENaC clamp restores h_ASL (ΔLCI -2.74)

Counterfactual Clinical Trial Failure Forensics (T-0428) & Single-AAV IS110 Bridge-RNA Co-Evolution (T-0427)

Inspect the exact bottleneck that stalled monotherapy in historical trials or dual-AAV vectors. Load the AlphaEvolve counterfactual rescued regimen directly into the live engine.

HISTORICAL UNSAT (0.65% WT · Pore 58.6 nm < 72 nm vector). Nebulized without coordinated ≥45 min saline-rhDNase lag or 15 Hz shear window. Steric mucin mesh trapped LNP cargo.

ALPHAEVOLVE COUNTERFACTUAL SAT (17.84% WT · Pore 105.4 nm · +17.19% WT gain). Enforce 0.85 h saline-rhDNase lag (INV-01). Nebulize 12 min post-15 Hz HFCWO (INV-02) with SMG1i/ENaC support.

Explore NCT03375047 in 940-Trial Registry →

Airway Clearance & Nebulizer Order Inspector (T-0435) · SIMPLIFY Post-ETI Context (NCT06350474)

Inspect why physical and enzymatic order matters across daily respiratory therapies, and review questions for your CF care team or respiratory therapist (educational reference, not a medical device or prescription).

BIOPHYSICAL BOTTLENECK PREVENTED: Nebulizing 7% hypertonic saline or tobramycin immediately after dornase alfa (without a ≥45 min enzymatic cleavage window and airway clearance) elevates luminal ionic strength and washes out rhDNase before extracellular neutrophil DNA polymers are cleaved (DOI: 10.1073/pnas.87.23.9188).

SEQUENCED MECHANISM (INV-01 & INV-02 PASS · 95% rhDNase Activity): 01 Bronchodilator opens conducting airways → 02 7% Hypertonic Saline draws osmotic water into the periciliary brush (DOI: 10.1126/science.1223012) → 03 Wait ≥45 min, then nebulize Dornase alfa → 04 15 Hz Vest (HFCWO, high-frequency chest wall oscillation) / PEP (positive expiratory pressure) clearance → 05 Inhaled antibiotic or vector into cleared airways.

Questions for Your CF Care Team (Clinic Visit Prep)

1. Hypertonic Saline & Dornase Alfa Sequencing: How long should we wait between nebulizing 7% hypertonic saline and dornase alfa so high ionic strength does not reduce enzyme activity?
2. Post-Vest (HFCWO, High-Frequency Chest Wall Oscillation) Inhalation Timing: In inhaled genetic therapy trials (such as NCT05668741 or NCT05248230), how is airway clearance coordinated prior to vector nebulization?
3. Multi-Modal Class I Nonsense Trials: Are there recruiting studies evaluating translational readthrough or mRNA replacement alongside apical hydration or modulator therapy?

Independent research, not medical advice. The lab does not assess efficacy or recommend treatments; ClinicalTrials.gov does not evaluate safety or efficacy; listing on the registry does not denote government endorsement. Tier T1 Exploratory Mechanistic Simulation: never change the timing of prescribed therapies without consulting your accredited CF care center.

Biological Scale Research Pipeline · 6 Interactive Stops

Follow the lab's curative biophysics from DNA nonsense codons to continental patient registries

Interactive Research Constellation · 6 Biological Threads · 19 Evidence-Locked Nodes

The Tree of Life: Curative Research Horizon

An interactive map of how six independent lines of cystic fibrosis research grow from foundational biology toward durable restoration for every CFTR genotype. Select any node to inspect its evidence tier, SHA-256 lockfile, and open workbench.

Six research threads converging on 100% genotype restoration: DNA & Gene Editing, mRNA & Nonsense Readthrough, Protein Folding & Channels, Airway Mucus Delivery, Clinical Trial Endpoints, and Global Registry Equity.

Open Full Tree of Life with Guided Quests →

Multi-Registry Denominator

Who is still waiting: the persistent 6–12%

Triple-combination modulators revolutionized care for common mutations like F508del. But modulators require a full-length protein substrate. For Class I nonsense alleles, canonical splice defects, and large deletions, modulators alone cannot restore channel function.

United States · CFFPR (2024)

8.0% ineligible across all approved modulators in 2024. 89.6% prescribed of eligible (26,277 individuals). Among adolescents/adults ineligible by genotype, 14.9% are Black and 23.7% are Hispanic (vs 3.4% and 10.3% of registry).

CFF Patient Registry 2024 Annual Data Report, pp. 8, 58.

Canada · CCFR (2024)

7.6% of individuals not eligible by age and genotype for modulators. 3,087 individuals prescribed ETI.

Canadian CF Registry 2024 Annual Data Report, p. 38.

United Kingdom · UKCFR (2024)

5.8% of adults and 11.5% of children ineligible for approved modulators by genotype.

UK CF Registry 2024 Annual Data Report, pp. 24–26.

Australia · ACFDR (2025)

5.8% of adults and 11.5% of children ineligible for modulators by genotype.

Australian CF Data Registry Annual Report 2025, pp. 5, 39, 68.

Open 10-Year ECFSPR & Multi-Registry Explorer →

T2 Receipt · Multi-Registry 6–12% Unserved Denominator & GitHub Proof

Primary Sources: CFFPR 2024 Annual Data Report (pp. 8, 58: 8.0% ineligible; 26,277 of 29,314 prescribed), CCFR 2024 (p. 38: 7.6% ineligible), UKCFR 2024 (pp. 24–26: 12.4% never modulated), ACFDR 2025 (pp. 5, 39, 68: 5.8% adult / 11.5% pediatric ineligible), ECFSPR 2024 (Ch. 9, pp. 81–88).

Scope & Non-Pooling Rule: Each national registry defines its denominator independently; figures are never summed or pooled across borders. Catalog SHA-256: 63eab6bb789bfb8096c6bc456e8809e6c79c7010ab83439cc933c77505a02cde.

GitHub Proof: cf-research/landscape/multi-registry-eligibility-2026.md ↗ · multi-registry-eligibility-claims.json ↗

Pillar 02 · Dedicated Biophysics Workbench (/cf/ptc/)

Class I Nonsense Mutation Readthrough Simulator (G542X & W1282X)

Modeling the biophysical trifecta for Class I premature stop codons: SMG1-mediated NMD transcript salvage, engineered suppressor-tRNA opal readthrough, and allosteric ETI corrector folding rescue.

Tri-Modal Conductance (ΔI_sc) 10.4%

Salvaged mRNA Pool 62.4%

Proteome Safety (TRSS) 96.2 / 100

ERAD Escape Rate 72.0%

Open Full Interactive PTC Readthrough Workbench → Inspect 3D Cryo-EM Coupling on /cf/cftr/ →

Investigational biophysical model based on PDB 6MSM/8EIQ Cryo-EM coordinates, RefSeq NM_000492.4, and Manfredi et al. (Biochemistry 2025). Research simulation only; not clinical guidance.

Where would you like to start?

A lab is a set of questions in motion.

Six interactive modules and research doors. Each page says what it covers and what it leaves out.

Curative research & biophysics

Class I nonsense mutation modeling, aerosol delivery formulation, Cryo-EM perturbation maps, and multi-registry access surveillance.

G542X / W1282X ECFSPR 100/100 Audit 5 Registries

Autonomous multi-agent architecture

Local worker loops on Ollama, dual-provider frontier review councils (Opus + Gemini), deterministic test gates, and zero token waste on idle.

Dual-Provider Council 138 Ledger Rows 60/60 Claim Gate

Open source tools & repositories

The lab's three public repositories (cf-lab, cf-skills, cf-research), the evidence tool, and catalog-enforced safety policies.

cf-evidence-loop EMA & UKRI Unpaywall & S2

Community resources & support

Help pages from cystic fibrosis organizations across countries, on a searchable map or list, with no tracking or advertising.

15 Countries 45 Verified Links EN / ES-MX

3D CFTR channel & gating workbench

Inspect four published Cryo-EM structures in the browser, measure residue distances, and toggle normal ATP gating versus F508del ER retention.

1,480 Residues F508del Ringed Zero CDN

Trial endpoint atlas & delivery explorer

Filter 940 interventional CF trials by phase, 8 canonical endpoint classes, PRISMA screening flow, and verbatim ClinicalTrials.gov quotes.

PRISMA 2020 8 Endpoint Classes SHA-256 Locked

If you would rather just read something: A letter to people living with CF, and those who care alongside them

If you want the counts: the trial endpoint atlas

What we have built

Three public repositories, in the open.

Each with its license, what it does, and what it does not.

Control repo · MIT
cf-lab

Shared instructions for AI agents, their skills (short guides an agent can load), the security rules, and an open task board.
Skills repo · CC0
cf-skills

An evidence tool that answers a research question as a record, with the source, the date and the limits attached.
Research repo · CC BY 4.0 notes, MIT code
cf-research

Research notes, guides to reading patient-registry reports, a catalog of what each source permits, and small checking tools.

More about the tools

Evidence, not claims

Counting what we check

The catalog is our list of sources and what each one permits. Our tools may only use a source the catalog allows. Here it is as it stands; the numbers are counted by a script, not typed.

113sources in the catalog

21a tool may download, within each site's rules

24APIs with a recorded usage policy

64a person must download

3refused by the publisher

1need a written request

These counts were taken from the lab's research repository as it stood on October 24, 2026 (version 5e0e412).

We also keep a ledger of the mistakes we have caught, with how each was found. Read the ledger or see the whole evidence page.

Letter

A letter to people living with CF, and those who care alongside them

To people living with CF, and to those who care alongside them,

This letter is written slowly, for you to read in whatever moment you have.

We think of the mornings that start with treatments before anything else can, and the evenings that end the same way. The kit packed for every trip. The waiting rooms, the forms filled in again, the phone calls chasing a prescription. The nights someone sits up, listening. And the jokes that only make sense to the people in the room. No two lives with CF are alike, and this is only some of what we think of.

This community is made of all of you: people living with CF; parents, partners, siblings and friends; care teams who learn your names; researchers; people who gave their time to studies; people who raised money; people who shared what they learned so someone else did not have to start from nothing; people who simply turned up. Thank you, for all of it, as things are. What you have built together is real, and it is yours. If you are new to it, or reading this alone, it is yours too.

Some people are not here to read this. We hold them close, and the people who love them.

You do not owe anyone an inspiring story. Ordinary days count. Hard days count. Rest counts. So does a good laugh at the wrong moment.

We are a small, independent lab, not a clinic, and not part of the Cystic Fibrosis Foundation or any other organization. We work in the open, on donated time, so that evidence about CF is easier to read and check, with plain words about what we do not know. Our work is research, not medical advice. You may take what is useful here and leave the rest.

With love and respect,
the lab

How we work

Working with AI,
checked by tests.

One agent proposes a change. Another looks for mistakes. We run checks before accepting it. Agreement alone is not proof. Prototypes, notes and code stay visible, so anyone can check the work.

The AI-loop field guide · The evidence loop · Local agent benchmarks

Local agent benchmarks

Ollama · GPU memory

Measured on the lab's machine across deterministic code checks, prompt injections, and fact recall depth probes. See all 4 benchmark panels

Independent research. No clinical advice.

Follow the work

Follow the signal.

Code & projectsGitHub Research notes & demosYouTube Optional: give to CF research through the Foundation, Washington chapterDonate