# The Public-Signal Momentum Score for Early-Stage Triage

*You will score any early-stage company on five public momentum dimensions and reach a defensible pursue-now, watch, or pass call with a written rationale.*

- Canonical URL: https://www.refolk.ai/guides/public-signal-momentum-score
- Pillar: Investing and deal sourcing
- Format: Framework
- Published: 2026-08-02
- Last reviewed: 2026-08-02
- Reading time: 14 min
- Keywords: how to evaluate startup momentum from public signals, score a startup before a first meeting, spot breakout startups early, distinguish real traction from vanity metrics, startup hiring velocity signal, early stage deal triage

## Key takeaways

- Investors spend zero to fifteen minutes screening a company, yet GitHub engineering acceleration leads a fundraise announcement by three to six weeks, so a pre-built score lets you act inside a window the crowd has not priced.
- Headcount alone is a weak fundraise predictor: a 2023 analysis found no statistically significant headcount-growth difference between top-quartile and bottom-quartile fundraisers, so weight hiring below engineering velocity and retention.
- Panel traffic estimates overreport small sites by roughly 94 percent and lose accuracy below about 5,000 monthly users, which is exactly the pre-seed range where you most want the number.
- The one durable test is whether the metric grows when the company stops pushing on it, which isolates demand pull from spend the same way a burn multiple under 1.5x does.
- In Refolk's index, 4,387 profiles hold a Founding Engineer title at US software companies versus 604 in the UK, a 7.3x gap that makes the same job posting mean more urgency outside the US.
- A crude 1 to 5 scorecard wins not by being smarter but by homogenizing dozens of weekly triage calls into comparable numbers that minimize emotional bias.

You spotted a name. A scout mentioned it, a product launched, or a search surfaced it, and now you have to decide whether it is worth a warm intro before you know anything private. This guide gives early-stage investors, fund platform and talent partners, and angels a fixed set of public momentum dimensions, a threshold for each, and a way to combine them into a defensible pursue-now, watch, or pass call with a written rationale. It is the triage judgement you make dozens of times a week, reduced to a score instead of a gut feel.

This is not the sourcing process that finds the name, and it is not the founder grade. It is the company-level call you make in the gap between spotting a startup and spending scarce social capital on it, using only what is public.

## Why a fixed score beats a gut call at triage

A fixed score wins because it makes dozens of weekly triage calls comparable, and comparability is the whole job at this stage. The documented practice is a crude weighted scorecard: rate founder, market, traction, economics, and exit on a 1 to 5 scale. As one angel training source puts it, a scorecard minimizes emotional bias and promotes deal homogenization. The value is not that the scorecard is smarter than you. It is that it forces every company through the same filter, so a strong signal in one deal is measured against the same yardstick as the next.

The economics of triage make this urgent. A documented seed funnel screens 2,000 companies to reach 500 partner meetings and 12 investments. You cannot spend real time on 2,000 names. Research puts partner screening research at zero to fifteen minutes per company, a deck scan at two to five minutes, and the raw meet or no-meet reflex at under two minutes. Whatever you use to decide has to run inside that window.

**2,000 to 12 - Companies screened per seed investment, per a documented VC funnel**

You cannot afford full attention on the top of that funnel, only a fast, repeatable filter.

Here is the edge. The triage window is short, but the best public signal has a long lead time. GitHub engineering acceleration precedes a fundraise announcement by three to six weeks. So a pre-built score lets you act inside a window the crowd has not priced yet, on evidence that is public but not obvious. That is the case for building the score once and running it fast, rather than re-deriving your opinion from scratch each time.

> **Rule:** Triage is not diligence
>
> Screening is a zero-to-fifteen-minute filter to decide whether to meet. Durable conviction needs a separate 40-hour diligence effort. Never confuse a high triage score with a decision to invest.

## The five momentum dimensions and what each proves

Score five public dimensions on a 1 to 5 scale: founder, market, traction, hiring velocity, and engineering velocity. Each proves something specific, and each has a way it lies. Naming the lie up front is what keeps the score honest.

**Founder.** What it proves: pattern-matched execution odds. Serial entrepreneurs show roughly 30 percent success versus 18 percent for first-timers per Gompers and colleagues, so prior founding is a real, if blunt, signal. How it lies: a strong resume with no evidence of shipping. Look for public work, not just titles.

**Market.** What it proves: whether the demand is real and large. How it lies: a founder-supplied total addressable market number in a deck. At triage you have no deck, so infer the market from third-party evidence, not projections.

**Traction.** What it proves: demand that persists. How it lies: a launch-day spike. The clean test is whether the number grows when the company stops pushing. Retention is the durable form: median net revenue retention for venture-backed B2B SaaS is 106 percent, and under 100 percent stalls most Series A talks.

**Hiring velocity.** What it proves: intent and capacity to scale, but only weakly on its own. How it lies: a live hiring page on a stalled company. A 2023 analysis found no statistically significant headcount-growth difference between top-quartile and bottom-quartile fundraisers, so weight this below velocity and retention.

**Engineering velocity.** What it proves: real building. How it lies: commit noise in AI-only repos and star counts that create outdated zombie leads. Read contributor growth and deploy cadence, not stars. A contributor count jump of 50 percent or more usually means a round just closed.

#### The five momentum dimensions, most durable at the base

1. **Founder** - Prior founding and public work, pattern-matches execution odds
2. **Engineering velocity** - Contributor growth and deploy cadence, hard to fake
3. **Traction** - Retention and repeat behavior, grows when unpushed
4. **Market** - Third-party demand evidence, never founder-supplied TAM
5. **Hiring velocity** - Visible but low-power alone, credit only in lockstep

*Weight the durable, hard-to-fake signals below the visible, easy-to-fake ones.*

## Growth thresholds by stage, and where they break

Score traction against the stage-appropriate month-over-month bar, but treat any rate off a tiny base as directional only. The strong bar at seed is 15 to 20 percent or higher month-over-month in the key metric. Below a real base, a single month tells you almost nothing.

| Stage | Strong MoM in key metric | Source basis |
|---|---|---|
| Pre-seed | 15-20%+, erratic, treat as directional | venture benchmark data |
| Seed | 15-20%+, one dataset says 15-25% | seed investor and SaaS data |
| Series A | 10-15% | 2025 SaaS dataset |
| YC weekly target | 5-7% week-over-week | accelerator benchmark |

The pre-seed row carries a warning. As one seed fund puts it, your month-over-month growth can look erratic: you might jump 300 percent after landing one customer, and then flatline the next month. A rate needs a base of hundreds of users before it means anything. This is the base-rate mirage, and it is the single most common way a triage score inflates.

The deeper test outlasts every fixed threshold. Does the number grow when the company stops pushing on it, or only when it pushes? That distinction, demand pull versus spend, is the same one captured by the burn multiple: under 1.5x signals demand pull, above 5x signals push. You will rarely see a burn multiple in public data, but you can often see whether growth continues through a quiet week with no launch, no ad push, and no press.

> The metric that grows when nobody is pushing on it is the only one worth chasing a warm intro for.

## How much attention a company earns at each triage step

Match your effort to the decision. The whole point of a score is to spend the least attention that still produces a defensible call, and the data shows how little that is at each step.

| Step | Time budget | Basis |
|---|---|---|
| Meet or no-meet reflex | Under 2 min | partner behavior study |
| Deck or intro scan | 2-5 min | associate screening report |
| Screening research | 0-15 min | VC screening research |
| Full diligence for conviction | 40+ hours | Wiltbank angel study |

The gap between the last two rows is not a contradiction, it is the shape of the pipeline. Screening is the fast filter this guide serves. Full diligence is what you do after a meeting, and it is where thoroughness pays: angels who spend fewer than 20 hours on due diligence see 2.5x higher failure rates than those who invest 40 or more. Roughly 80 percent of deal flow still arrives through network referrals, so a warm intro is the scarce resource this score is protecting.

#### The seed screening funnel

| Stage | Figure | Note |
| --- | --- | --- |
| Screened | 2,000 | zero-to-fifteen-minute triage each |
| Partner meetings | 500 | warm intro spent here |
| Investments | 12 | full 40-hour diligence |

*A documented seed funnel narrows 2,000 screened companies to 12 investments.*

## Score a company: the procedure

Run these seven steps in order. The whole pass fits inside the fifteen-minute screening window, and it ends with a number, a bucket, and a one-line rationale you can defend later.

#### The public-signal momentum score

1. **Trigger and de-dupe** - Open the company site, GitHub org, and LinkedIn page. Confirm the name is not already logged. Done when all three URLs are in front of you.
2. **Pull public signals** - Gather one dated data point per dimension: a MoM demand proxy, the GitHub commit and contributor trend, headcount trend, and job-posting clusters. Done when each dimension has one observation with a date.
3. **Score each dimension against thresholds** - Rate founder, market, traction, hiring, and engineering velocity on a 1 to 5 scale using the fixed bars. Done when all five carry a number and a one-word reason.
4. **Discount the distortion-prone signals** - Down-weight panel traffic below 5,000 monthly users and any rate off a base under a few hundred users. Done when every fragile signal is annotated low confidence.
5. **Cross-check velocity in lockstep** - Confirm product and hiring velocity move together. One alone is a warning, not a green light. Done when you have written lockstep confirmed or flagged the mismatch.
6. **Weight and total** - Combine the scores with fixed weights into one number and map it to pursue-now, watch, or pass. Done when you have a total and a one-line rationale.
7. **Decide the spend** - Pursue-now spends a warm intro, watch sets a signal alert, pass logs and moves on. Done when the decision and its next action are recorded.

The signal pull in step two is where a good sourcing tool removes the most friction. Instead of hand-checking whether a company is actually staffing up in the roles that signal momentum, you can ask for the pattern directly across many companies at once.

I ran this search: `US seed-stage software startups that posted 3 or more founding-engineer roles in the last 60 days` - [see the full result list](https://www.refolk.ai/s/wqevkwawd9).

*Returns companies staffing the earliest, hardest-to-fill roles, which is a stronger intent signal than raw headcount.*

I built [Refolk](/) to answer that kind of question in plain English across GitHub, LinkedIn, and the open web, so the ten minutes you have goes to judgement instead of tab-hopping. When a name surfaces, Refolk lets you check the founding-engineer and Head of Growth hiring pattern that separates a company shipping from one that only announced.

## How this score goes wrong

Every failure mode here produces a false positive: a company that scores well and should not. Knowing the seven ways the score lies is more valuable than the score itself, because a confident wrong call costs you a warm intro you cannot get back.

**Base-rate mirage.** A 300 percent month-over-month jump off 10 users reads as breakout and means nothing. Check the denominator, and require a base of hundreds before you trust any rate.

**Panel-traffic inflation.** Estimation tools overreport small sites by roughly 94 percent, and accuracy drops sharply below about 5,000 monthly users. A surging competitor may simply be below the reliability band. Confirm the site is above roughly 5,000 monthly users before you quote the number at all.

**Vanity spike.** Downloads, press, and sign-ups jump on a launch and decay. Apply the test: does it still grow when the company stops pushing.

**Headcount as proxy.** A funded company that has stalled still shows open roles. Open roles without commit and product velocity are a false positive. Cross-check that engineering and hiring move in lockstep.

**Single-signal GitHub read.** Commit noise is high in AI-only repos, and star counts create outdated zombie leads. Read contributor growth and deploy cadence, never stars alone.

**Checklist theater.** A fully filled scorecard built on founder-supplied projections looks rigorous and proves nothing. Validate market claims against third-party data, and remember founder quality stays the highest-signal input.

**Funding-announcement anchoring.** Not all funded companies execute. Some raise and stall. Validate a round with post-round hiring and shipping velocity, not the headline itself.

> **Watch out:** One velocity without the other is a red flag
>
> Product velocity and hiring velocity need to move in near lockstep. One moving alone is not a partial green light, it is a potential warning signal that hiring is running ahead of what the product can support, or that the team is shipping without the intent to scale.

## The founding-engineer signal and where it is scarce

The founding-engineer title is a scarce, geography-concentrated signal, so the same job posting means different things in different markets. In Refolk's index, 4,387 profiles hold a Founding Engineer title at US software companies versus 604 in the UK, a 7.3x gap.

| Segment | Count in Refolk's index | Derived |
|---|---|---|
| Founding Engineer, US software | 4,387 | baseline |
| Founding Engineer, UK software | 604 | 7.3x fewer than US |
| Head of Growth, US | 2,118 | 0.48x the US founding-engineer pool |

Read the gap this way. A UK startup posting for founding engineers is drawing from a pool 7.3x thinner than a US peer, so the same posting implies more intent and more urgency there. The company is fishing in a small pond on purpose, which is a stronger signal of conviction than the identical role in San Francisco, where the pool is deep. Among US founding-engineer holders in Refolk's index, San Francisco and New York are the two densest metros, and named current employers include Navier AI, Erebor, and Stellar Sleep.

**7.3x - More US Founding Engineer profiles than UK, in Refolk's index**

4,387 in the US against 604 in the UK, which changes what an identical job posting signals by geography.

The Head of Growth title is the demand-side companion signal. There are 2,118 in the US in Refolk's index, concentrated in New York, Austin, Seattle, and Los Angeles. A company hiring both a founding engineer and a Head of Growth in the same window is one where product and go-to-market velocity are moving together, which is exactly the lockstep the score is looking for.

## A copy-paste scoring rubric and weights

Use this rubric verbatim, then adjust the weights to your thesis. The weighting reflects what the evidence supports: engineering velocity and traction carry the most, hiring the least, because headcount alone does not separate winners.

**Public-signal momentum scorecard**

```
Founder (weight 3): 5 = repeat founder with public shipping evidence; 1 = strong resume, no public work
Engineering velocity (weight 3): 5 = contributor growth 40%+ and steady deploys; 1 = stars only, no commit trend
Traction (weight 3): 5 = grows unpushed, NRR above 106%; 1 = single launch spike, then flat
Market (weight 2): 5 = third-party demand evidence, large; 1 = founder-supplied TAM only
Hiring velocity (weight 1): 5 = founding-eng + growth roles in lockstep with product; 1 = open roles, no product velocity

Total = sum of (score x weight). Max = 60.
45-60 = pursue-now: spend a warm intro
30-44 = watch: set a signal alert, recheck in 4-6 weeks
under 30 = pass: log the name and move on

One-line rationale (required): ___
```

*Score each 1-5, multiply by the weight, sum, then map to a bucket. Adjust weights to your thesis, not the thresholds.*

The buckets map to spend, not to opinion. Pursue-now spends a warm intro. Watch sets an alert, because GitHub acceleration leads announcements by three to six weeks, so a watched company can move to pursue-now on a fresh signal. Pass logs the name so a future signal can resurface it.

#### Before you call the score final

- [ ] Every dimension has one dated public data point, not a guess
- [ ] Any growth rate is computed off a base of at least a few hundred users
- [ ] No panel traffic number below 5,000 monthly users is treated as reliable
- [ ] Product velocity and hiring velocity are confirmed to move in lockstep
- [ ] Market size rests on third-party evidence, not a founder-supplied figure
- [ ] The total maps to a bucket and a written one-line rationale
- [ ] The decision is logged with a next action, even for a pass

## Keeping the score current

Re-run the score on watched companies on the lead-time clock, not on a calendar. Because GitHub engineering acceleration precedes a fundraise by three to six weeks and a contributor jump of 50 percent or more usually means a round just closed, a four-to-six-week recheck on watch-bucket names catches the move before the announcement makes it public. Set the alert on the contributor and deploy trend, since those are the earliest and hardest to fake.

Recalibrate the thresholds themselves as your own outcomes accumulate. Track which pursue-now calls converted to meetings that you were glad you took, and which passes you later regretted. The scorecard earns its keep by homogenizing decisions across dozens of calls, so the weights should shift toward whichever dimension most often predicted the calls you got right. Keep the two velocity signals, engineering and hiring, checked against each other every time, because the lockstep test is the one cross-check that no single-dimension threshold can replace.

## Frequently asked questions

### How long should scoring a startup's momentum before a first meeting actually take?

Under fifteen minutes for the whole triage. Research shows VC partners spend zero to fifteen minutes screening a company, scan a deck in two to five minutes, and often make the meet or no-meet call in under two minutes. This score is built to fit that window: roughly one minute to trigger, five to ten to pull signals, and five to score and total. Full diligence that builds durable conviction is a separate 40-hour job you only start after the meeting.

### What month-over-month growth rate counts as strong momentum at seed?

Strong seed-stage companies often show 15 to 20 percent or higher month-over-month growth in their key metric, with one 2025 SaaS dataset putting the seed band at 15 to 25 percent and Series A at 10 to 15 percent. Y Combinator's weekly equivalent is 5 to 7 percent week-over-week. At pre-seed, treat any single-month rate as directional only, because growth off a tiny base is erratic and one customer can produce a misleading 300 percent jump.

### Why is hiring velocity a weaker signal than it looks?

A 2023 analysis comparing top-quartile fundraisers to the bottom three quartiles found no statistically significant difference in headcount growth, so hiring counts alone do not predict who raises. A funded company that has stalled still keeps a hiring page live. Hiring earns its weight only when it moves in lockstep with product and engineering velocity, which is why the score cross-checks the two before crediting either.

### How do I tell real traction from a vanity spike?

Apply one test: does the number get bigger when the company stops pushing on it, or only when it pushes? Downloads, press mentions, and one-time sign-ups spike on a launch and decay, so they are activity, not traction. Durable traction shows up as retention and repeat behavior. The median net revenue retention for venture-backed B2B SaaS is 106 percent, and anything under 100 percent stalls most Series A conversations.

### Can I trust web traffic estimation tools for a pre-seed company?

No, not at that scale. A study of 1,787 sites found one panel-based tool overreported sessions by about 94 percent, with accuracy improving sharply for larger sites. Estimation quality drops significantly below roughly 5,000 monthly users, and the reliable band sits around 5,000 to 100,000. Since pre-seed lives below that floor, treat any traffic estimate for a tiny site as low confidence and lean on GitHub and hiring signals instead.

---

*From the Refolk guide library. I revise these guides rather than replacing them, so the current version is always at https://www.refolk.ai/guides/public-signal-momentum-score*