The Comparable-Round Set: Sanity-Checking a Valuation Ask
You will be able to build a defensible comp set of recent rounds from public signals, adjust for differences, and turn it into a valuation range you can hold in an IC.
You have a valuation ask on the table and need to decide whether the number is fair before you take a view to your investment committee. This guide is for early-stage investors, platform and talent partners at funds, and angels who want to build a set of genuinely comparable recent rounds from public signals, adjust for how the target differs, and turn that set into a defensible valuation range and a negotiation stance. The founder-facing best answers explain valuation methods to people setting their own ask. This flips it to your side and treats comp-building as what it actually is: a search task, then a filtering task, then a small piece of arithmetic you can defend line by line.
What makes a round set defensible rather than decorative
A defensible comp set is 5 to 8 recent rounds, sourced from filings rather than press, converted to one valuation basis, and traceable deal by deal so it survives someone else's counter-comps. The test is not whether your number is right in some absolute sense. It is whether the set holds up when a skeptical partner tries to break it.
Comparable company analysis for a startup valuation borrows its discipline from three adjacent worlds that have argued these questions in front of judges and regulators. Real-estate appraisal under Fannie Mae requires a minimum of three closed comparable sales within the last 12 months. M&A practice caps precedent comps at 24 months for serious work and insists the trend be analysed, not just the level. Delaware appraisal litigation, in cases like DFC Global Corp. v. Muirfield Value Partners, rejects comp sets with fewer than five deals and sets where the largest deal runs more than 5x the median. None of these are venture rules. There is no securities regulator or venture body that sets a binding recency window or minimum set size for private funding-round comps. Treat that borrowed standard as load-bearing, and flag it as borrowed when you present it.
The comps do not have to be perfect. They have to be defensible against the assessor's counter-comps.
The reason this matters more in venture than in real estate is that private markets re-rate fast and hard. A comp that felt current a year ago can be stale by a full turn today, and staleness cuts both ways depending on the cycle. Get the set right and the arithmetic that follows is almost mechanical. Get the set wrong and no adjustment method will save the number.
The public signals that pin stage, sector, geography and traction
The primary legal signal is SEC Form D. When a company raises a US venture or private-equity round, it must file a Form D with the SEC within 15 days of the first sale of securities, and each filing carries the offering amount, equity type, number of investors, date of first sale, and total sold to date. First-time filer status is often a clean proxy for a first institutional round.
Form D matters because press does not cover the market you are pricing against. Roughly 90% of seed rounds never receive top-tier press coverage. If you build your comp set from TechCrunch or LinkedIn announcements, you are systematically blind to the majority of the market, and the fraction you can see is self-selecting toward the larger, hotter, announced rounds. That skews your baseline up before you have adjusted anything. Filings appear on average 7 to 14 days before any press, so the filing is both more complete and earlier.
Signal layers for a comp set, most authoritative first
- SEC Form DLegal filing within 15 days of first sale; carries amount, equity type, investor count, date
- Round databasesSector, stage, geography, revenue and investor filters over the filing base
- Company web and product signalsTraction proxies like headcount, hiring and product maturity
- Press and newslettersCovers under 10% of seed rounds; skewed toward larger announced deals
Commercial round databases sit on top of the filing base and add the filters you need to build a candidate list: industry, location, stage, revenue, valuation and investors. The most common are the general-purpose round databases; use them to enrich and cross-check, not as your only source. The discipline is to anchor on the filing, then let the database tell you sector and traction. Where you are trying to find the founders behind a filtered slice of the market fast, plain-language search over an index of professional profiles collapses hours of database filtering into one query. Refolk is built for exactly that kind of ask.
The current baselines you are pricing against
Anchor to the Carta stage and sector median, and carry the exact figure with its period and basis so nobody in the room can dispute what you started from. The published venture benchmarks below are the market baseline before any adjustment for how your target differs.
From raw candidates to a defensible set
- 15Raw candidates from filings plus databases
EDGAR full-text plus one round database
- 10Same stage, sector and geography
First hard screen
- 7Within recency window
12 to 18 months, 24-month cap
- 6Survives remove-one sensitivity
No single deal moves the range materially
Table A. Carta median valuations by stage, 2025.
| Stage | Median pre-money | Basis and period |
|---|---|---|
| Seed | $16M | Q3 2025, pre-money |
| Series A | $49.3M | Q3 2025, pre-money |
| Seed | $20M | 2025, post-money |
| Seed 95th percentile | $80.5M | 2025, post-money |
The direction of travel is up. Median seed pre-money on Carta reached a new high of $16M in Q3 2025, up 14% year over year, and median Series A pre-money hit an all-time high of $49.3M in the same quarter, having already reached $48M in Q1. On a post-money basis the median seed valuation is $20M, and the 95th percentile hit $80.5M in 2025, nearly 3x the $28.5M 95th percentile in 2019. Startups on Carta raised $27.3B in Q3 2025, the highest quarterly sum in three years.
Two structural facts change how you read a raise amount. Median Series A dilution fell from 20.9% to 17.9% in a year, so the same check now buys less equity at a higher valuation, which means comparing raise amounts without valuations misreads the stage. And the down-round rate fell from roughly 19% in Q1 2025 to under 12% by early 2026, which tells you the cycle is rising, not falling, when you decide which direction to time-adjust an old comp.
Segment AI from non-AI before you touch the median
The single biggest distortion in a current comp set is the AI tag, not fundamentals. A comp misclassified as AI moves your Series A baseline by roughly 38% and your late-stage baseline by up to 193%, so segment AI and non-AI comps into separate columns before you compute anything.
Table B. AI vs non-AI valuation premium, 2025.
| Stage | AI premium over non-AI | AI share of cash raised |
|---|---|---|
| Series A | +38% | not stated |
| Series D | not stated | 58% |
| Series E+ | +193% | not stated |
The mechanism is capital concentration. At every stage from Series A onward, AI startups raised larger rounds at higher valuations than non-AI peers in 2025. At Series D, 58% of all cash raised went to AI startups, and more than 60% of all venture on Carta in Q1 2026 went to AI companies. The blended market median is being pulled by a distinct sub-market. If your target is not an AI company, an AI-heavy comp set will hand you a baseline a full stage-premium too high, and you will walk into your IC defending a number the market never priced for a company like this one.
Whether a defensible set is even reachable in your sector
Before you spend two hours pulling candidates, check whether the pool is deep enough to reach five survivors, because comp pool depth varies by roughly 3.5x across sectors in the same country. A thin sector forces you toward the five-deal floor where Delaware-style rejection risk is highest.
Table C. Founder supply by market and sector, from Refolk's index.
| Segment | Founders in index | Derived ratio |
|---|---|---|
| US AI founders | 389 | 1.0x base |
| UK AI founders | 130 | 0.33x of US |
| US fintech founders | 1,344 | 3.45x the US AI pool |
In Refolk's index of professional profiles, a US fintech seed has 1,344 candidate founders behind it versus 389 for a US AI company, so your fintech comp pool is roughly 3.5x deeper. That difference decides your risk posture. A deep pool lets you build 6 to 8 tight comps and discard the borderline ones. A thin pool leaves you at the floor, where every included and excluded deal has to be footnoted.
Cross-border sets are structurally thinner still. The UK AI founder pool is 130 in the index, roughly one-third the US depth, which means UK comp sets breach the five-deal threshold more often and frequently need US comps time-adjusted and geography-adjusted in. Know this before you promise anyone a clean domestic set.
The procedure, start to finish
Run these seven steps in order. Analyst time is noted at each stage; the whole pass is a focused half-day for one person, plus an hour of partner time on the adjustment.
Building the comp set and deriving a range
- Fix the subject profilePin stage, sector, geography, round size and traction from a Form D filing plus one database record. Done looks like a one-line spec, for example US seed, vertical SaaS, NYC, roughly $3M raise, $40k MRR. Analyst, about 30 minutes.
- Pull the candidate setQuery Form D through EDGAR full-text search plus a round database for matching raises. Done looks like 8 to 15 raw candidates, allowing for EDGAR's overnight publish lag. Analyst, 1 to 2 hours.
- Filter to true comparablesScreen on industry, size, growth, business model, margin, geography and recency, and enforce the 12 to 18 month window with a 24-month hard cap. Done looks like 5 to 8 survivors. Analyst, about 1 hour.
- Normalise valuationsConvert every comp to one basis, since post-money equals pre-money plus amount raised. Done looks like a single comparable column with no pre-and-post mixing. Analyst, about 45 minutes.
- Set the market baselineAnchor to the Carta stage and sector median, keeping AI and non-AI comps in separate columns. Done looks like one baseline number carrying its source and date. Analyst, about 30 minutes.
- Adjust for differencesRun the Bill Payne Scorecard weights against the baseline for team, market, product and traction. Done looks like an adjusted point estimate you can trace back to each factor score. Analyst and partner, about 1 hour.
- Build the range and stress-testBracket the estimate with percentiles, then remove each comp in turn to test whether the conclusion depends on one uncertain deal. Done looks like a defensible range plus a clear stance against the ask. Analyst, about 45 minutes.
One note on ordering. Some founder guides run the Scorecard adjustment before fixing a set, starting from a regional average, because the founder is pricing their own company. Investor practice inverts this: fix the comp set first, establish the baseline from it, then adjust. That keeps your number traceable to specific deals rather than an abstract regional figure, and it is what survives an IC challenge.
The Scorecard adjustment, in numbers
The documented adjustment route is the Bill Payne Scorecard method. It starts from the average pre-money valuation of comparable companies in the region, sector and stage, then adjusts for how the target compares on weighted factors.
| Factor | Weight |
|---|---|
| Entrepreneur, team, board | 30% |
| Size of opportunity | 25% |
| Product and technology | 15% |
| Sales and marketing | 10% |
| Need for more financing | 5% |
| Other | 5% |
Mechanically, you score the target on each factor and multiply the weighted total by the peer average. If the average pre-money is $10M and the target scores 80% on team, 70% on market, 90% on product, 60% on traction and 50% on risk, the weighted result lands at roughly $7.15M pre-money. The output is a point estimate; the range comes from the stress-test in the final step.
Company | Stage | Sector | AI? (Y/N) | Geography | Round date | Amount raised | Pre-money | Post-money | Source (Form D / DB) | Include? | Reason if excluded
One row per candidate. Fill every field or footnote why it is blank; blanks are where counter-comps get in.
How this goes wrong
Most bad comp sets fail in one of seven predictable ways. Each has a false positive it produces and a specific check that catches it. This is the part of the work worth slowing down for, because a clean-looking number resting on a broken set is worse than no number at all.
| Failure mode | False positive it produces | Check that catches it |
|---|---|---|
| Stale comp passed off as current | Range one turn too high | Enforce 12-to-18-month window; re-rate anything older |
| AI-halo contamination | Non-AI target priced on an AI baseline | Segment AI and non-AI into separate columns |
| Thin set dressed as robust | Confident number resting on 2-3 deals | Delaware under-5 test plus remove-one sensitivity |
| Outlier dominance | Range pulled by a single mega-round | Flag any comp over 5x the set median; cap its weight |
| Basis mismatch | Apparent 20-25% gap that is just round size | Convert all comps to one basis first |
| Press-only sourcing | Baseline inflated by announced rounds | Source from Form D and EDGAR, not newsletters |
| Cherry-picked highs or lows | Set built to fit a predetermined stance | Document every included and excluded deal |
A few of these deserve extra weight. Stale comps are the most common single error and the most dangerous right now because the market is rising. With seed medians up 14% year over year, an old comp understates as often as it overstates, which is the opposite of the 2022-to-2023 down cycle. That makes upward time-adjustment defensible today rather than a thumb on the scale. A 2021 comp can sit two to three turns above today's reality if you leave it unadjusted, so re-rate it explicitly and show your work.
The thin-set failure hides behind confident presentation. A number computed from three rounds looks identical on a slide to one computed from eight. The remove-one sensitivity test is what exposes it: pull each comp out in turn and recompute. If the indicated value changes materially when any single deal leaves, disclose that sensitivity, because your conclusion depends on one uncertain sale. Delaware rejects sets under five deals for exactly this reason.
Basis mismatch is the quiet one. Mixing pre-money and post-money figures manufactures an apparent 20 to 25% valuation gap that is nothing more than the round size. Post-money equals pre-money plus the amount raised, so convert everything to one column before you compare a single pair.
What good looks like before you take it to IC
Run this checklist before you call the set finished. Each item is a thing to verify, not a topic to think about. If any fails, you have more work before the number is defensible.
Comp-set readiness check
- The set contains at least 5 comps, ideally 6 to 8, after filtering.
- Every comp raised within 18 months, and anything near 24 months is time-adjusted with a note.
- AI and non-AI comps sit in separate columns and the target is priced from the matching one.
- Every valuation is on one basis, with no pre-and-post-money mixing.
- No single comp exceeds 5x the set median without a footnoted justification.
- The set was sourced from Form D filings, not press announcements alone.
- Every included and excluded deal is documented with a reason.
- Removing any one comp does not move the range materially, or the sensitivity is disclosed.
The output you carry into the room is three things: a range, not a point; a stance versus the ask stated plainly, such as "the ask sits above our set's 75th percentile and we would push toward the median"; and the record table itself, so anyone can trace the range back to named deals. A range without the underlying rows is an opinion. A range with the rows is an argument.
Keeping the set current
A comp set has a short shelf life because the medians it rests on move every quarter. Re-anchor the baseline against the latest Carta State of Private Markets release before you reuse an old set, and re-check the down-round rate and dilution figures, since both shift the direction and size of your time-adjustment. A set built when the down-round rate was 19% carries different assumptions than one built when it is under 12%.
The mechanism to watch, rather than any single number, is capital concentration in AI. As long as more than half of venture dollars flow to AI companies, the blended market median will keep drifting away from the median a non-AI company actually faces. Re-check the AI share each quarter and keep your segmentation honest. When you need to rebuild the candidate pool quickly for a fresh set, searching an index of professional profiles in plain English gets you from a subject profile to named recent-raise founders faster than rebuilding database filters from scratch, which is where most of the half-day actually goes.
Questions practitioners ask
How recent does a comparable funding round have to be?
Default to a 12 to 18 month window and treat 24 months as a hard cap for serious work. No venture body codifies this, so it is a borrowed standard from adjacent practice: Fannie Mae requires closed comps within 12 months and M&A convention caps precedent comps at 24 months. Markets re-rate, so a 2021 comp can sit two to three turns above today's reality. Time-adjust anything older rather than dropping it silently.
How many comparable rounds do I need for a defensible set?
Aim for 5 to 8 survivors, pulling 8 to 15 raw candidates first and narrowing on similarity and recency. The floor is 5. Delaware appraisal precedent rejects comp sets with fewer than 5 deals, sets where the largest deal exceeds 5x the median, and sets that exclude obviously relevant deals without a footnoted reason. Below 5 deals your number rests on 2 or 3 rounds and will not survive counter-comps.
Where do I find rounds that never got press coverage?
Use SEC Form D through EDGAR full-text search. Companies raising a US venture round must file within 15 days of the first sale, and each filing carries the offering amount, equity type, number of investors and date of first sale. Around 90% of seed rounds never get top-tier press, so a newsletter-built set is systematically blind to most of the market and skews toward larger, hotter announced rounds.
How do I stop AI comps from inflating my baseline?
Segment AI and non-AI comps into separate columns and never apply an AI-premium baseline to a non-AI target. The distortion is large: at Series A the median AI valuation ran 38% above non-AI in 2025, and at Series E and beyond the premium reached 193%. The mechanism is capital concentration, with more than 60% of Q1 2026 venture on Carta going to AI companies, so the blended median reflects a distinct sub-market.
Should I adjust for differences before or after building the comp set?
Build the comp set first, then adjust. Some founder guides run the Scorecard adjustment first, starting from a regional average, because the founder is setting their own ask. Investor practice inverts this: fix a filtered set of genuinely comparable recent rounds, establish a baseline from it, then run the Scorecard weights against that baseline. This keeps the number traceable to specific defensible deals rather than an abstract regional figure.
Try it on the search you came here for
Stop building boolean strings. Just describe the person.
Type one sentence. I plan the search, read GitHub, public LinkedIn and Crunchbase records, and the open web as it is right now, and hand back a ranked list with the reason next to every name.
01Describe them
One plain sentence. Role, city, stack, stage, whatever matters to you.
02I read the web live
GitHub, public LinkedIn and Crunchbase records, the open web. Not a database that went stale last quarter.
03You read the shortlist
Ranked, with the reasoning under every name. Open a profile, ask a follow-up, narrow it down.
- Staff backend engineers in NYC who shipped Rust in production
- Series A fintechs in SF under 50 people, growing headcount this year
- Maintainers of fast-growing Rust web frameworks on GitHub
- No boolean, no filters, no seat to buy. One box.
- Read at search time, so a profile updated yesterday counts today.
- Every step visible as it runs, every name with its reason.
500 free credits on sign-up. No card, no demo call. See real searches.