Sampling the biological model
Three independent runs passed their fixed sampling checks. This supports computation under the specified model, not independent biological validation.
Loading study data…
Read the sampling report
The data snapshot couldn't load. You can still read the linked reports below.
OPEN SCIENCE. OPEN QUESTIONS.
Follow the evidence.
Question the result.
A shared map of cancer research, with models that put its ideas to the test. Explore the literature, inspect the results, and see where the next useful question begins.
New to the project? Start with the plain-language guide.
THE COLLECTION
A broad map.
Explicit boundaries.
—records in the collectionIndexed + text-recovered streams
—indexed records in these chartsMeSH-defined census and adjacent descriptors
—measurable mechanismsWithin the current descriptor mapping
01 / THE LITERATURE
Explore the evidence atlas.
Compare publication patterns across the indexed census. Clinical-trial shares describe how papers are labelled, not how effective a treatment is.
Trial share = trial-labelled records / all records tagged with that mechanism. Tags can overlap.
Loading mechanisms…TRIAL SHARE
FOLLOW A QUESTION
How does a mechanism appear in the literature?
Select a mechanism to inspect its trial share, publication trend, and associated cancer sites.
What this map can—and cannot—tell you
These charts use the indexed census records. The broader collection includes additional text-recovered records that are not part of these aggregates. Record-level browsing of the census is not offered.
Descriptor widths vary: a larger count can reflect a broader label. Mechanism and cancer-site tags overlap. Missing descriptors, including those for TTFields and bioelectric modulation, mean “not measurable here,” not zero research.
— records are labelled as clinical trials: — of the indexed census, or — of the — records with informative study-design labels. Both denominators matter.
Inspect the study-design classification ↗02 / THE MODELS
Show the checks.
Keep the failures.
A useful model has to survive more than a convincing plot. Here is where the current sampling work stands.
Challenging new geometries
Two rotated-box runs failed weight-concentration checks; one also failed regional-mass accuracy. Every declared region was observed.
Loading study data…
Read the failure analysis
Connecting to an actual assay
A candidate ML210 assay is identified. Raw replicate identities and a model-to-assay mapping are still needed before evaluation.
Still an open questionExperimental validity remains unresolved
See the missing inputs
INSPECT THE LATEST CHECK
One sampler.
Three harder shapes.
Nine independent learned runs. Fixed seeds and budgets. All results retained.
Protocol frozen before evaluationLoading the committed results…
These are synthetic sampling checks, not treatment comparisons. Passing finite moment and region checks cannot establish complete target coverage, precise tails, or clinical validity. No pooled posterior is published. View the calibration ledger ↗
03 / THE NEXT QUESTIONS
A roadmap you can help move.
Progress means resolving a specific uncertainty. These are the next useful contributions, with the evidence needed to call them complete.
01
Computational methods
02
Experimental validation
03
Model experiments
04
Evidence & discovery
Make sampling robust to correlation.
Develop a proposal revision, freeze a separate protocol, and compare it with the unchanged baseline.
Validate one independent observable.
Resolve raw replicate metadata, assay timing, and the mapping between measurements and model outputs.
Test what controls immune exposure.
All 20 new seed blocks retain the activation contrast without half-maximal saturation. A new controlled study separates source patterns, release, timing, and fixed recipient availability. Next: challenge layout and recipient-loss assumptions; independent validation remains pending.
Earn better discovery claims.
Measure evidence quality and test candidate ranking against popularity and random baselines.
GO ONE LEVEL DEEPER
Start where your curiosity is.
The manuscript ↗
The full argument, quantitative results, and limitations, with citations you can follow.
The Python dashboard ↗
Browse the historical 4,830-article archive and census aggregates. Starts only when you choose to launch it.
Data & methods ↗
Acquisition instructions, committed reports, and the boundary between public artifacts and bulk data.
Help test an idea ↗
Bring a missing paper, a counterexample, a better method, or an experiment that could change the conclusion.
THE WORK IS OPEN
Good questions belong here.
Built for people willing to follow the evidence—including where it says we don't know.