Biomedical claims, audited before they move.

Company identity Named accountability, controlled routing. Registered-agent or business mailing channels only; private residence is not published. Source Source Ledger Map every public statement to a traceable source, status label, and missing-artifact note. Stress Break Tests Challenge QC filters, normalization defaults, and donor-level support before a claim moves. Verdict Verdict Line Survives, caveat, and held labels come only after the audit trail is written. 02 Source mapped Track datasets, tissue, platform, access, metadata, and missing artifacts. 01 Claim received Identify the claim, decision context, and what would make it useful or unsafe. Provenance Source map Every public statement needs a source state: shown, request-only, missing, or held. Audit Audit Us The public website and its source are built to be checked together. Preprocessing Did the anchor move with disease? Housekeeping genes and scaling assumptions can erase the biology they were meant to stabilize. Filtering QC sensitivity Filters are checked for removing stressed or relevant cells before analysis. Replication Cohort tiering Findings are bucketed by cross-cohort support before claims are written. Reviewer path Need the full review path? Open the deeper page for source maps, method rows, figures, and status labels. Contact Start with one source. Send role, claim, source link, and decision context. No patient files or treatment requests. Official channels Official Channels Public company correspondence should be verified before sensitive material moves. Security Security Contact The public site is static and intentionally small. Security disclosures currently use leng.jixiang@ninenames.com. Claim Claim triage Sort the claim, decision context, and unsafe promises before method review starts. Context Decision context Ask what decision depends on the claim: review, funding, strategy, publication, or care boundary. Boundary Scope boundary Name what we can review, what stays out, and what needs a controlled handoff. Index Source index A compact list of papers, accessions, repositories, figures, and missing artifacts. Dataset Accession check Verify whether a named dataset or repository is reachable, current, and actually relevant. Literature Paper trail Separate what a paper shows from what later summaries, slides, or citations imply. Access Data availability Mark whether source material is shown, request-only, missing, blocked, or held. Metadata Metadata caveat Record donor, tissue, stage, platform, access, and caveats before trusting a matrix. Raw files Raw file gate Flag when only processed matrices exist and raw reads are needed to test a failure mode. QC QC caps Audit whether fixed filtering caps removed stressed or relevant biology before analysis. Stress Mitochondrial stress Treat high mitochondrial signal as a question, not automatic trash. Background Ambient RNA Check whether background RNA or sample handling explains a tempting signal. Cells Doublet check Ask whether mixed cells or multiplets created a fake hybrid state. Pipeline Preprocessing path Trace what happened before the public matrix existed. Normalize Normalization anchor Test whether the reference used to stabilize the data moved with disease or stress. Anchor Housekeeper drift Check whether classic housekeeping genes are stable enough for the claim. Baseline Baseline shift Ask whether the comparison baseline moved under the claim. Scaling Scaling choice Compare whether a signal survives reasonable normalization and scaling choices. Donors Donor statistics Require cell-level signals to survive donor-level evidence. N Donor counts Show how many independent donors carry the signal, not just how many cells. Stats Pseudoreplication Block claims that confuse many cells with many independent samples. Magnitude Effect size Prefer magnitude, uncertainty, and replication tier over dramatic p-values. Assay Batch / platform Ask whether instrument, chemistry, site, or batch explains the signal. Chemistry Assay version Track chemistry and pipeline version before comparing studies. Stage Disease stage Separate mechanism from severity, stage, sampling window, and survivor effects. Age Aging filter Test whether the signal is aging, frailty, or sampling rather than disease mechanism. Comparator Cross-disease filter Ask whether the same signal appears across diseases where the story should differ. Annotation Cell-type boundary Check whether annotation choices created or erased the cell state. Repeats Repeat-read loss Check whether repeats or retroelement signal vanished before counting. Annotation Annotation loss Ask whether the feature existed but the annotation layer could not count it. Matrix Matrix blind spot Mark signals that cannot be judged from the processed matrix alone. Structure Protein-fold claim Treat predicted structure as evidence with confidence limits, not a verdict. Confidence Structure confidence Check model confidence, domains, disorder, and whether the claim exceeds the tool. Variant Domain / variant Ask whether the variant touches the claimed domain or mechanism. Support Wet-lab support Separate computation, association, perturbation, and direct experimental support. Citation Citation chain Find where careful language became stronger after repeated citation. Figure Figure provenance Tie each public figure to source state, caption limits, and missing material. Visuals Public figure pack Use simple public-safe figures as method orientation, not final proof. Mechanism Mechanism map Map rival explanations before turning a signal into a biological story. Downgrade Status downgrade Lower claim strength when stronger checks fail or support narrows. Correction Correction loop Make visible what changed, why it changed, and what would change it again. Caveat Caveat label Attach the uncertainty where the reader sees the claim. Gate Replication gate Hold claims that need another cohort, assay, or donor-level check. Held Held material Name that material exists without publishing controlled or unsafe details. Ledger Public ledger Future public findings need source, status, safety, and translation review first. Packet Reviewer packet A controlled handoff can include source map, checks, caveats, and request-only material. Diligence Trust / diligence Route partner or funder questions through scope, confidentiality, and decision context. Email First email frame Start with role, claim, source link, and decision context only. Clinical Clinical boundary Evidence support does not become diagnosis, treatment, or emergency guidance. Safety No medical advice Keep public language out of prescriptions, emergencies, and individual-care instructions. Privacy No patient files Do not send patient files until role, purpose, recipient, and route are clear. Privacy Privacy route Route privacy, takedown, and data-rights questions away from public issues. Legal Formal notice Keep legal and formal correspondence on official channels, not map interactions. Security Security disclosure Send vulnerabilities through the responsible disclosure route, not public comments. Public GitHub issues Use issues for public corrections, translation bugs, build issues, and audit questions only. Language Translation review Mark language as draft until a human checks tone, jargon, and safety. Language Language risk A good claim can look unserious if the translation sounds inflated or strange. Controlled Controlled source boundary Keep controlled source tables and sensitive review material off the public site. Company Company record Keep the company name, director accountability, and official channel story aligned. Director Director accountability Use named accountability without publishing private residence or private research material. Repository Public repo The repository proves source, buildability, issue routing, and correction history. Handoff Request-only source Some source material can be named publicly but shared only after role and purpose are clear. Fix path Correction request Public mistakes should have a visible route for correction without exposing sensitive material. Ledger Public method ledger A plain record of what was checked, what was not checked, and what changed after review. Samples Sample manifest Make the countable unit visible: donor, sample, tissue, platform, timepoint, and missing fields. Boundary Consent boundary Do not move sensitive material until role, purpose, permission, and recipient are explicit. Breaks Negative evidence Keep the results that weakened the claim visible instead of burying them in prose. Conflicts Contradiction index List papers, datasets, or checks that pull against the attractive story. Origin Provenance score Show how close the public claim stays to source files, source methods, and source limits. Gaps Missing data map Name the absent files, absent metadata, or absent comparators before trust rises. Figure Chart rebuild Rebuild the picture from traceable inputs before treating the picture as evidence. Selection Sample selection Ask who entered the dataset, who was excluded, and what that does to the claim. Survival Survivor bias Test whether the remaining samples are the biology, or just the samples that survived selection. Strata Sex stratification Check whether a pooled result hides a sex-linked pattern or a sex-linked artifact. Tissue Tissue mismatch Separate what a blood, muscle, brain, cell, or model system can actually support. Timing Timepoint window Ask whether the signal depends on onset, stage, sampling interval, or treatment timing. Model Species transfer Mark where animal or model evidence stops and human evidence must begin. Model Cell-line boundary Name what an engineered or immortalized cell can prove, and what it cannot. Bridge Bulk / single-cell bridge Check whether bulk tissue and single-cell claims are answering the same question. Spatial Spatial context loss Ask what location, neighborhood, and tissue architecture disappeared before analysis. Isoform Isoform ambiguity Check whether gene-level summaries hide transcript-level differences. Pathway Pathway overfit Challenge pathway stories that can explain almost anything after the fact. Enrichment Gene-set inflation Ask whether a large gene set made weak evidence look broader than it is. Labels Ontology drift Check whether labels, disease names, or pathway terms changed meaning across sources. AI Model leakage Check whether training, tuning, or benchmark overlap made performance look cleaner. Benchmark Benchmark mismatch Ask whether the validation task matches the real decision that will use the claim. Contam Training contamination Check whether examples, labels, or answers leaked into model development. Structure Protein fold boundary Separate a proposed structure from a proven mechanism, target, or intervention. Docking Docking overclaim Keep docking as a hypothesis unless orthogonal evidence survives. Reagent Antibody specificity Ask whether the reagent marks the intended target under the actual assay conditions. Confidence Confidence ladder Show why a claim is low, cautious, useful, strong, or ready for handoff. Would break Falsifier list Name what result would weaken or break the claim before the next check starts. Next test Next replication Convert uncertainty into the smallest useful next check. Plan Replication plan Define the cohort, source, method, and stop rule needed for a stronger claim. Downgrade Downgrade trigger State the exact check that moves a claim from strong to cautious. Supports Support trigger State the exact check that lets a claim move from interesting to usable. Hold Hold trigger State what makes a claim stay private, unfinished, or not ready for public use. Question Safe question Start with the claim, source link, decision context, and boundary; leave sensitive files out. Rules Reviewer rules Agree scope, audience, confidentiality, and allowed outputs before deeper review. Hard read Red-team read A deliberate attempt to break the claim before money, trust, or action moves. Diligence Diligence brief Compress the review into what survived, what failed, what is missing, and what to ask next. Corrections Correction window Make it easy to challenge public wording without using unsafe channels. Domain Neurodegeneration lane Audit stage, tissue, donor count, aging, and cross-cohort claims before mechanism talk. Domain Muscle disease lane Separate injury, regeneration, inflammation, stage, and sampling before the story hardens. Domain Immune signaling lane Challenge whether immune signal is cause, state, handling, batch, or tissue mixture. Domain Rare disease lane Use smaller claims, clearer boundaries, and visible missingness when N is small. Mechanism Mechanism review lane Move from association to mechanism only when the bridge is named and tested. Language Translation quality lane Treat awkward or unsafe wording in any language as a public correction issue.