Skip to Content
FounderGrade
Request a Pilot
The Score and Methodology

The Eight Dimensions, and Who Decides Them.

Every FounderGrade score comes from eight published dimensions with fixed weights, awarded by named human reviewers. This page states the rubric in full, then states exactly where automated assistance stops and human judgement begins.

Read the Eight Dimensions Trust and Security
A reviewer annotating printed financial pages beside a laptop showing a highlighted document

The Score

How the Hundred Points Are Split

Eight dimensions, a fixed published weight for each. They sum to 100 and they do not change between companies, sectors or reviewers.

Team18Market13Product13Traction17Financials12Scalability10Risk10Fundability7

Dimensions

What Each Dimension Asks, and What It Requires

A reviewer cannot score a dimension without the evidence listed under it. Missing evidence is recorded as missing, never assumed.

01

Team

Weight 18

Can these specific people build and sell this specific thing?

Evidence Required

  • Founder biographies with verifiable roles and dates
  • Evidence of relevant domain or operating experience
  • Equity split, vesting and any founder departures to date
  • Named key hires and the gaps still open

Scores Well

Direct operating experience in the problem space, a complete founding team, and a clean cap table.

Scores Poorly

Unverifiable claims, a single founder covering every function, or unexplained co-founder churn.

02

Market

Weight 13

Is the market real, reachable and large enough to matter?

Evidence Required

  • Bottom-up market sizing with the arithmetic shown
  • Named competitors and an honest account of how they win
  • Regulatory or structural conditions that gate entry
  • Evidence that the buyer has budget today

Scores Well

Bottom-up sizing tied to a real buyer, with competitors named rather than dismissed.

Scores Poorly

Top-down sizing borrowed from an analyst report, or a claim of no competition.

03

Product

Weight 13

Does the product exist, work, and solve the stated problem?

Evidence Required

  • A working demonstration or recorded walkthrough
  • Architecture summary and current technical debt
  • Usage data from real users, not internal accounts
  • Roadmap with what has shipped versus what is planned

Scores Well

A live product in the hands of users who are not friends of the founders.

Scores Poorly

Design mockups presented as product, or a roadmap with no shipped history behind it.

04

Traction

Weight 17

Is anything actually growing, and is the growth durable?

Evidence Required

  • Revenue or usage series by month, exported from source systems
  • Retention and churn cohorts
  • Pipeline with named stages and conversion rates
  • Signed contracts, letters of intent or pilot agreements

Scores Well

Month over month growth in a metric tied to revenue, with retention that holds up in cohorts.

Scores Poorly

Cumulative charts, vanity registrations, or a single anchor customer presented as a trend.

05

Financials

Weight 12

Do the numbers hold together and does the runway support the plan?

Evidence Required

  • Historic profit and loss, and a current balance sheet
  • Monthly burn, cash position and runway in months
  • Unit economics with the assumptions stated
  • Use of proceeds tied to the milestones being raised against

Scores Well

Unit economics that survive a change in one assumption, and runway that reaches the next milestone.

Scores Poorly

A forecast that hockey-sticks with no stated driver, or use of proceeds that funds the round rather than the plan.

06

Scalability

Weight 10

What breaks at ten times the current volume?

Evidence Required

  • Cost to serve per unit and how it moves with scale
  • Operational dependencies that are currently manual
  • Infrastructure and vendor concentration
  • Hiring plan against the growth being forecast

Scores Well

A cost to serve that falls with volume, and named plans for the parts that are still manual.

Scores Poorly

Growth forecasts that assume headcount scales linearly with revenue forever.

07

Risk

Weight 10

What could end this company, and does the team see it?

Evidence Required

  • Regulatory and licensing exposure by market
  • Customer, supplier and platform concentration
  • Legal disputes, IP position and key-person dependency
  • The team's own stated top three risks

Scores Well

Risks named by the founders before a reviewer finds them, with mitigations that cost something.

Scores Poorly

A risk section that lists only competition and hiring.

08

Fundability

Weight 7

Is this raise structured in a way an investor can actually take?

Evidence Required

  • Instrument, amount, valuation or cap, and prior rounds
  • Cap table and any structural overhangs
  • Milestone plan the tranches would be released against
  • Data room completeness

Scores Well

A clean instrument, a defensible price, and milestones that a third party could verify.

Scores Poorly

Terms that change between conversations, or a data room assembled after the first investor asks.

Bands

What a Number Actually Means

Bands are descriptive, not promotional. A low score is a statement about available evidence, not about the founder.

85 to 100

Exceptional

Evidence is complete and independently checkable across every dimension. Rare, and reviewers are asked to justify it in writing.

70 to 84

Strong

Material strengths with documented evidence, and any weaknesses are named and being worked.

55 to 69

Developing

A credible business with real gaps. The reviewer notes say which gaps and what would close them.

40 to 54

Early

Direction is visible, evidence is thin. Usually pre-traction or pre-product.

Below 40

Not Yet Ready

Not a judgement on the people. It means the evidence needed to score higher does not exist yet.

Disagreement

What Happens When Two Experts Disagree

Averaging away a disagreement hides the most useful information in the review. FounderGrade escalates it instead.

  1. 1

    Two Independent Reviews

    Two calibrated experts score every dimension separately. Neither sees the other's scores, notes or identity, and neither sees the founder's identity.

  2. 2

    Divergence Check

    Scores are compared dimension by dimension. Where the two reviewers differ materially on any single dimension, that dimension is flagged rather than averaged away.

  3. 3

    Third Reviewer on Flagged Dimensions

    A third expert scores only the flagged dimensions, with the same blind conditions. The reconciled score and the reason for the divergence are recorded.

  4. 4

    Founder Appeal

    A founder can appeal any dimension once per review cycle by submitting evidence that was not in the original file. Appeals are heard by a reviewer who did not score the original.

  5. 5

    Published Reasoning

    The score a founder publishes carries the per-dimension reasoning with it. A score without its reasoning is not a FounderGrade score.

The Rubric Is Versioned

Every published score records the rubric version it was produced under. When the rubric changes, older scores keep their original version rather than being silently restated. Changes are listed in the changelog.

See the Changelog See How a Review Runs

The Rule

No automated output is a score, a verification or a release decision. It is an input to a judgement made by a person whose name is attached to it.

Machine extraction is a drafting aid. It has no authority over a score, a verification outcome or a disbursement. A named human must accept, reject or amend every item before anything can be submitted.

Division of Labour

Step by Step, Who Does What

Read this as a contract. If a row says the human decides, there is no path in the product that skips them.

Reading the Data Room

Machine

Locates candidate passages against each rubric dimension and cites the document and location.

Named Human

Accepts, rejects or amends every candidate before it can inform a score.

Scoring a Dimension

Machine

Nothing. No automated system produces or suggests a score.

Named Human

Sets the score and writes a rationale for each of the eight dimensions.

Checking a Milestone Claim

Machine

Compares the claim against available sources and marks each source corroborated, contradicted or unable to verify.

Named Human

Reads the checks, requests what is missing and records the verification decision under their own name.

Releasing Money

Machine

Nothing. No automated system can release a tranche.

Named Human

A named verifier releases or holds, with a written reason attached to the disbursement record.

Raising an Integrity Concern

Machine

Surfaces an observation and lists the innocent explanations for it.

Named Human

Decides whether to dismiss it, ask for clarification or hold a disbursement, with a recorded reason.

Managing Reviewer Quality

Machine

Measures each reviewer against the pool: mean, spread, drift, disagreement.

Named Human

A lead reviewer decides on weighting, tier and recalibration, and records the justification.

Automated checks never approve a milestone and never release funds. They tell the verifier where to look. The release decision is recorded against a named person.

Limits

What the Assistance Cannot Do

Every claim below is a limitation we would rather publish than have discovered.

Extraction Is Not Comprehension

Automated extraction finds passages that look relevant. It does not know whether a number is credible, whether a contract is enforceable or whether a customer is real. That is the reviewer's work, and it is the part we pay for.

A Corroborated Check Is Not a Verified Milestone

The strongest value an automated check can return is Corroborated, which means one source agreed with one claim. It does not mean the milestone is met. Only a named verifier concludes that.

A Signal Is Not a Finding

Integrity signals describe an observation and the ordinary reasons it might occur. They carry no penalty, they do not change a score, and a team is always able to respond in writing.

Calibration Does Not Make a Score Correct

Measuring reviewers against one another makes scores comparable. It does not make any single score right. We would rather state that plainly than imply a precision we do not have.

Timing Figures Are Assumptions

Both minute figures are stated assumptions used to illustrate the workflow. FounderGrade is pre-launch and has no measured reviewer timing data yet. Published timings will replace these once real engagements exist.

Reviewer Calibration

How We Keep Reviewers Comparable

Two reviewers reading the same rubric should reach similar conclusions. When they do not, we measure it rather than hope.

  • Every reviewer scores against the same published rubric, and every score carries a written rationale per dimension.
  • Reviewer scoring patterns are measured against the pool: the mean they give, the spread across their scores, how far they have drifted, and how often a second reviewer disagreed with them.
  • A reviewer who drifts away from the pool is not removed quietly. They score a set of sealed anchor files, the gaps are discussed, and they return at a reduced weight until they are re-measured.
  • Calibration status determines the reviewer tier, and the tier determines the rate band a reviewer can charge. Consistency is what a reviewer is paid for.
  • None of this makes a single score correct. It makes scores comparable to one another, which is the honest claim.

What Triggers A Flag

Mean Deviation
A reviewer whose mean sits more than five points from the pool mean over twenty reviews enters recalibration.
Spread
A spread below ten points is flagged. A reviewer who scores everything near the middle is not discriminating between companies.
Drift
A change of more than three points in the rolling mean over ten reviews is flagged, in either direction.
Disagreement
A disagreement rate above eighteen percent against second reviewers is flagged, because one of the two is reading the rubric differently.
Sample Size
Under twenty reviews a reviewer is Provisional and carries a reduced weight, because the measures above cannot be read yet.

What Recalibration Involves

  1. 1

    Anchor Set

    The reviewer scores five sealed anchor files that the pool has already scored, without seeing the pool result.

  2. 2

    Variance Review

    Their scores are compared dimension by dimension against the pool, and the largest gaps are discussed with a lead reviewer.

  3. 3

    Written Rubric Note

    The reviewer records where their reading of the rubric differed, which feeds the next rubric version.

  4. 4

    Weighted Return

    The reviewer returns at a reduced weight for ten engagements, then is re-measured against the pool.

Integrity Signals

Signals, Not Accusations

These are signals for human review, not findings and not accusations. A signal describes an observation and the innocent explanations for it. No signal changes a score, blocks an application or withholds money on its own. A person decides, records a reason, and the team can respond.

If A Signal Is Raised On Your Work

  1. 1

    The Team Is Told

    A team is notified whenever a signal on their work is acted on. Signals that are dismissed without action are not raised with the team.

  2. 2

    The Team Sees the Observation

    The notification states what was observed and which of their submissions it concerns. It does not characterise intent.

  3. 3

    The Team Responds

    The team has ten working days to respond in writing and attach evidence. The response is attached to the signal permanently.

  4. 4

    A Second Person Decides

    The response is reviewed by someone other than the person who acted on the signal. Their decision and reason are recorded.

  5. 5

    Held Funds Are Released Or Explained

    If the response resolves the signal, any held disbursement is released and the hold is recorded as resolved. If it does not, the team receives the reason in writing.

Ready When You Are

You Know the Rubric. Now Get Scored Against It.

Eight dimensions, fixed weights, published before anyone is assessed.

Create Your Account Verification Standard

Pursuing DFSA Innovation Testing Licence · DIFC, Dubai