Nine Names

Forensic biomedical review

Before a claim becomes expensive, we try to break it.

Nine Names audits the source map, pipeline defaults, donor and cohort support, and failure modes before a biomedical claim earns a safe handoff.

  1. 01 Claim received

    Identify the claim, decision context, and what would make it useful or unsafe.

  2. 02 Source mapped

    Track datasets, tissue, platform, access, metadata, and missing artifacts.

  3. 03 Defaults audited

    Treat QC filters, normalization anchors, and matrix construction as testable choices.

  4. 04 Donor/cohort checks

    Separate cell-level signals from donor-level and cross-cohort support.

  5. 05 Failure modes challenged

    Ask whether the signal is disease, age, assay, stage, selection, or artifact.

  6. 06 Safe handoff or hold

    Publish only public-safe summaries; controlled material waits for qualified review.

Recent audit spotlight: ALS

We audited whether ALS signals survived donors, cohorts, and pipeline checks.

ALS is the current reviewed example, not the company category. The spotlight shows how Nine Names turns a biological claim into checks before language.

Donor-level audit 80 dataset-by-axis tests; 23 cell-level flags; 3 donor-level p < 0.05.

Cell count is not donor count.

Cross-cohort tiering Five ALS-relevant cohorts across cortex, spinal cord, PBMC, and fibroblast data.

Signals are separated by source, assay, and replication posture.

QC / normalization stress Pipeline defaults were treated as audit targets, not background settings.

Filters, anchors, and pooled-cell certainty can all change the story.

Reviewer boundary Public summary first; source tables by qualified review.

No treatment advice or prescriptions; physicians and PIs decide.

Recent examples

More claims we have checked.

Each example shows a public-safe verdict and inspection rows; dense source tables wait for qualified review.

Corrected Public safety

Public boundary gate

Overstrong claims get held, narrowed, or routed to qualified review before publication.

Source map Public language points to sources and status labels.

Claims without source-clean backing stay request-only or held.

Clinical boundary No treatment advice, prescribing language, or public pressure.

Medical decisions stay with physicians, PIs, and qualified reviewers.

Evidence snapshot

Evidence snapshot from current audit

Axes tested
80
Pre-gate robust flags
23
Donor-level p < 0.05
3
Screened findings
2,255
Tiered findings
596
Replicated-status findings
287

What we actually did

What reviewers can inspect.

A claim moves only after sources, defaults, donor/cohort support, and failure modes are checked. The figures show the public-safe shape of that inspection.

Scatter plot showing cell-level signal against donor-level signal, with only a small subset surviving donor-level significance.
Donor gateThe public figure shows the donor gate, not a treatment claim.
Horizontal bar chart showing screened findings divided into not promoted, Tier R, Tier S, and Tier H evidence-status buckets.
Evidence statusScreened findings are sorted before public language is written.

Audit method

Checks before claims.

Each row names the check, the failure mode, and the output we expect before public language gets sharper.

Donor gate

Failure mode
Cell-level p-values can outrun donor evidence.
Output
donor p / donor count

Cohort tiering

Failure mode
Same-donor or same-assay support can look like replication.
Output
replication tier

Normalization audit

Failure mode
Housekeeping anchors and baselines can move with disease.
Output
stress flag

QC sensitivity

Failure mode
Filters can remove the stressed cells that matter.
Output
sensitivity note

Source map

Failure mode
Local-only or missing artifacts raise evidence risk.
Output
shown / request / held

Clinical boundary

Failure mode
Therapeutic language can turn into advice.
Output
rewrite / hold

Full source tables are request-only; the public page shows the gate shape.

Audit outcomes

Labels come after the work.

Survives

The narrow claim survived source, method, and donor/cohort checks.

Caveat

The signal is useful, but limited by cohort, assay, metadata, or review access.

Hold

The material stays off the public page until it is source-clean and safe.

Rewrite

The claim was too broad, too clinical, or too hot for the evidence.

Boundaries

What we do not do.

Public

Allowed
summary figures
Blocked
source tables, patient files

Qualified review

Allowed
source notes, reviewer packet
Blocked
anonymous download

Offline

Allowed
drafts, sensitive records
Blocked
web publication

Medical decisions

Allowed
physician/PI review
Blocked
Nine Names advice

Clinical decisions stay with physicians and PIs. No prescriptions, recruitment, or treatment instructions.

Contact

Review inbox

Send role, claim, and source question. Do not send patient files or treatment requests.

Review inbox review@ninenames.com Claim review only. No patient files.