The verification layer for science.

We check research claim by claim, and show you what holds up.

§ 1·How a paper moves

From paper to claim graph to verdict

Each claim is checked by the method that suits it best.

Cell migration

Target-gene knockdown slows tumor-cell migration in a wound-healing assay

Abstract

Cell migration drives invasion in many solid tumors. We show that knocking down the target gene slows tumor-cell migration in the wound-healing assay, and that the effect tracks the loss of expression across three comparisons.

2 Methods

Cells were seeded at 2 × 10⁵ per well and transfected after 24 h. The three biological replicates come from independent cultures. Wound closure was imaged every 6 h following the protocol of [9].

3 Results

Treatment reduced target expression by 40% at 48 h (Figure 3b). Treated wounds closed more slowly at every time point. The reduction remains significant after Bonferroni correction across Table 2’s three comparisons.

  1. Checked by Scientific reasoning models
    Knockdown slows tumor-cell migration. Contested, P(false) 0.62.
  2. Checked by Computer vision
    Knockdown cuts expression by 40%. Contested, P(false) 0.71. Necessary, confirmed. Supports Knockdown slows tumor-cell migration.
  3. Checked by Language models
    The assay follows the cited protocol. Upheld, P(false) 0.06. Supporting, confirmed. Supports Knockdown slows tumor-cell migration.
  4. Checked by Statistics
    The effect survives correction for three comparisons. Refuted, P(false) 0.93. Supporting, confirmed. Supports Knockdown slows tumor-cell migration.
  5. Checked by Heuristics
    The three replicates are independent. Refuted, P(false) 0.90. Necessary, confirmed. Supports The effect survives correction for three comparisons.
  • 1 upheld
  • 2 contested
  • 2 refuted
§ 2·What we build

Software that reads, hardware that records

A printed research paper on a desk, three of its sentences bracketed and numbered in red, with lines running from them into a small hand-drawn claim tree in the margin

Software

Reads research claim by claim.

Symby finds the claims in a paper, spots the shaky ones, and checks each one, starting with the cheapest test.

  • A check to run
  • A signed run
A self-recording barograph on a workbench, its pen tracing a line across the chart on a turning drum, a signed tag tied beside it

Hardware

Records what actually happened.

Symby sits beside your instruments and signs every run, so results are evidence, not someone’s word.

§ 3·Cheapest check first

The verification ladder

Each check climbs only as high as it needs to. Only the hardest reach the bench.

Spent1,244 units

Without the ladder4,000 units

Saved 2,756 units

  1. F0 ·1 unit a check

    Sanity check

    Does it hold up in a quick model?

  2. F1 ·10 units a check

    Existing data

    Does it match what’s already been measured?

    • The three replicates are independentrefuted
    • The assay follows the cited protocolupheld
  3. F2 ·100 units a check

    Re-run

    Do the authors’ code and data give their numbers?

    • The effect survives correction for three comparisonsrefuted
  4. F3 ·1,000 units a check

    Bench

    Does it happen again in a real lab, captured and signed?

    • Knockdown cuts expression by 40%contested
§ 4·Outputs

What comes back

  • Issues and weak points

    Statistical anomalies, image problems, unsupported citations, and irreproducible steps, each with quoted evidence.

    • 1 upheld
    • 2 contested
    • 2 refuted
  • Replication package

    Everything needed to re-run the work. Running needs no account.

    • Table 2’s analysis
    • Code and data
  • Lab protocol

    An experiment plan a lab can execute, compiled from the paper’s methods.

    • From the methods
    • The wound-healing assay
  • A living research record

    Claim graph, execution graph, protocol, data, and write-up in one versioned record. Fork it like code.

    • 5 claims in one graph
Fig. 1 · Everything to grow it again
Seed packets of plain brown paper on a worn wooden table: one lies open with its seeds spilled across the wood, two more are tied together with string under a paperclip, and beside them a seedling grows in a small clay pot
§ 5·Deep Scan

SymbyAI runs eight checks on every paper

20 credits per page

  1. Editorial & writing quality

    Checks grammar, clarity, and structure
  2. Image forensics

    Finds duplicated, spliced, or manipulated figures
  3. Code & data validation

    Catches errors in code, data, and statistics
  4. Claim verification

    Checks claims against their sources and the literature
  5. Plagiarism & sourcing

    Finds copied text and citations that don't support their claims
  6. Safety & compliance

    Flags personal data, ethics, conflicts of interest, and regulatory issues
  7. Literature synthesis

    Surveys related work and the gaps in it
  8. Research recommendations

    Suggests datasets, collaborators, and directions
Cell migration

3 Results

Expression fell in every knockdown lane. Treated wounds closed more slowly than controls at every time point (Table 2).

Figure 3b. Target expression in control and knockdown lanes, 48 h after transfection.
Table 2. Wound closure at 24 h, three comparisons.
Comparisontp
Knockdown vs. control2.310.012
Knockdown vs. scrambled2.080.026
Scrambled vs. control0.410.69

4 Discussion

Loss of the target gene slows migration without changing proliferation, which points to a role in adhesion turnover. Earlier work found the same decrease after this treatment [14].

A rescue experiment and a second cell line would show whether the effect is specific to the gene.

Declarations

The cell lines are commercial, and no human or animal subjects were involved. The authors declare no competing interests.

References

  1. In vitro scratch assay: a method for the analysis of cell migration. Nat Protoc (2007).
  2. Target expression after the same treatment.
§ 6·Pricing

Simple, credit-based pricing

  • Free TrialFor a first look
    $0
    250 credits flat, never renews
  • CarbonFor getting started
    $49/ month
    2,000 monthly credits
  • GrapheneFor your core work
    $99/ month
    4,000 monthly credits
  • DiamondFor research at scale
    $249/ month
    10,000 monthly credits

Buying for a team or institution? We offer custom contracts — contact us at sales@symbyai.com.

What we do | SymbyAI