v2.1 ALPHA · 2026-08 · Edition month 2026-08
Methodology
Michelin rates restaurants with anonymous inspectors and confidential criteria; we do the exact opposite — every rule is published, and a score can only be computed from rules + data. There is no hand-editable score field anywhere in the system. You are free to disagree with our weights, but you can always see exactly where a score came from.
Independence declaration · Never pay-to-play
Listing and ranking are never paid, never purchasable, never negotiable. Companies can correct facts (welcome — we label them with the ◐ source badge), but they cannot negotiate scores: scores move only with data and published rules. There is no sponsored slot, paid placement, or any pay-to-play path in the codebase. Founders in the community play by the same rules — their company page gets a 👋 badge, but their votes carry no extra weight.
How the score is composed
Career Score is the serious, research-driven score — everything on this page. Vibe Check is the community's fun vote (does the website slap, does the founder have aura, is the funding story dramatic). Until voting opens, Overall equals Career. The two are always displayed separately and never blurred together.
Momentum 势头
16
How fast the company is moving right now: revenue growth (compared within stage cohorts) plus recent capital-market validation — including when that validation went the wrong way.
Efficiency 经营效率
20
Revenue scale, and how efficiently that revenue is produced: revenue per employee and revenue per dollar raised. Gross margin and NRR aren't public for private companies — these are the closest proxies public data can actually support.
Backing 靠山
8
Who's backing it and how deep the war chest is. Down-weighted: nearly every company on this list has top-tier backing, so it barely discriminates.
Market 赛道
11
Market tailwinds versus the risk of being squeezed by incumbents or model vendors. Editorial judgment (✎) scored from public research.
Product 产品力
11
Product reputation, customer quality, and competitive position. Editorial judgment (✎) scored from public research.
Upside 个人回报
16
The option math as an employee: how far the entry valuation is from the ceiling (growth-adjusted), whether the upside is already priced in, and the liquidity track record.
Missing data ≠ zero
Private-company data is incomplete by nature. When a metric has no reliable data, it drops out of the denominator: Career Score = points earned ÷ points assessable × 100, with coverage shown alongside (“scored on 78% of the rubric”). Companies with fewer than 55 assessable points get no overall score — only their known items. So low coverage is a signal of an information gap, not of a bad company — and it doubles as our recruiting pitch to insiders: work at this company? Those “unknowns” are one message away from filled. (v2.0 raised that floor from 30 to 55: renormalising assumes missing data is missing at random, and it isn't — companies disclose flattering numbers and withhold the rest, so a half-covered score reads high rather than merely uncertain.)
Data provenance tiers
Every fact that enters a score carries a provenance label and an as-of date — there are no naked numbers in this interface. Provenance also carries weight, but as confidence rather than punishment: a weakly-sourced metric is pulled toward the neutral middle of its own band table by the factor below, so a shaky high score comes down and a shaky low score comes up by the same amount. (v1.1 multiplied points instead, which dragged every unverified fact toward the worst possible value and quietly penalised companies whose researchers were honest enough to label an estimate.) Each company also shows a verification rate — the share of its scored metrics backed by a ✓ public fact with a source link.
Independently verifiable sources — press coverage, regulatory filings, official databases — with source links attached.
×1.00What the company or its founders say about themselves, not independently verified — treated as biased but informative.
×0.85Provided by identity-verified community members who work there; insiders can only submit facts about their own company.
×0.85Estimates from third-party research firms such as Sacra, with sources attached — used as directional reference.
×0.70Our own research judgment — reasoning published in full, rebuttals welcome.
×1.00Stage cohorts
Tripling from $4M ARR and tripling from $300M ARR are not the same feat, so revenue growth applies different bands per stage cohort — early-stage companies are measured against early-stage peers, late against late. Everything else is graded on one absolute scale, and scale-based metrics mean late-stage companies still average roughly 16 points higher. Read a score against its own cohort; comparing a 76 in Early with a 76 in Late is not a like-for-like comparison.
Early 早期
ARR < $20M (or unknown with rounds ≤ Series B)
Growth 成长期
ARR $20M – $150M (or Series C/D)
Late 后期
ARR > $150M (or ≥ Series E / Pre-IPO)
The full rubric · 12 metrics
This is the entire rulebook — the band tables behind every clickable score on the rankings are exactly the ones below.
Momentum 势头 Weight 16/100
Revenue growth 收入增速
Max 12
Input: arr_growth from the structured data (YoY multiple); bands vary by stage cohort
Early cohort · ARR < $20M (or unknown with rounds ≤ Series B)
Growth cohort · ARR $20M – $150M (or Series C/D)
Late cohort · ARR > $150M (or ≥ Series E / Pre-IPO)
Capital validation 资本验证
Max 4
Input: Months between the latest round and 2026-08, combined with the direction of the valuation — a down round or discounted secondary mark overrides recency
Note: v1.1 scored recency alone: a down round three months ago earned full marks while a company that hadn't needed money in two years scored lowest. A down mark now overrides recency, and the weight drops from 8 to 4 — 'raised recently' is a weak proxy for 'doing well'.
Efficiency 经营效率 Weight 20/100
ARR scale 收入体量
Max 6
Input: ARR from the structured data ($M, incl. third-party estimates; provenance shown with the score)
Revenue per employee 人均收入
Max 8
Input: ARR ÷ headcount. Gross margin isn't public for private companies; this is the closest public proxy for how efficient the business actually is
Note: Deliberately lean teams (turbopuffer at ~$4.5M/head) rank far ahead here — exactly the profile that war chest and funding recency underrated in v1.1. Not cohort-banded: a lean early company can and does beat a bloated late one.
Capital efficiency 资本效率
Max 6
Input: ARR ÷ total raised — the higher the ratio, the less dilution was burned for the same revenue. Also the best public proxy for burn discipline at a private company
Note: Weight raised from 3 to 6: it's the best burn-discipline signal public data supports, and it's the half of 'war chest' worth keeping.
Backing 靠山 Weight 8/100
Investor quality 机构层级
Max 5
Input: Lead/participating investors on record, matched against the published Tier-1/Tier-2 lists (see methodology)
Note: Weight cut from 9 to 5. Across 21 companies this averaged 8.71/9 with three distinct values — a near-constant eating a tenth of v1.1's weight. It's also circular: the index selects for tier-1-backed companies, then rewards them for being tier-1-backed.
War chest 弹药储备
Max 3
Input: Publicly reported total raised ($M)
Note: Weight cut from 6 to 3: 'raised a lot' is simultaneously validation and dilution, so the sign is ambiguous. Capital efficiency answers the question this was reaching for.
Market 赛道 Weight 11/100
Market (editorial) 赛道判断
Max 11
Input: Editorial 0–10 score from public research (✎), scaled proportionally; the rationale is shown with the score
Scaled directly: points = score ÷ 10 × 11. Anchors below describe the scale — they don't bucket it.
Note: v1.1 re-banded the 0–10 score into five buckets, collapsing 21 companies into just two distinct values — the bands were coarser than the input, discarding information the editor had already provided. Now scaled directly.
Product 产品力 Weight 11/100
Product (editorial) 产品力判断
Max 11
Input: Editorial 0–10 score on public reputation, customer quality and competitive position (✎), scaled proportionally; rationale shown with the score
Scaled directly: points = score ÷ 10 × 11. Anchors below describe the scale — they don't bucket it.
Upside 个人回报 Weight 16/100
Equity upside 期权上行空间
Max 8
Input: Current valuation ($M), growth-adjusted: fast growth discounts the effective valuation, because it takes a smaller multiple to reach the same ceiling
Note: Growth adjustment: effective valuation = valuation ÷ growth multiple (growth clamped to 1–3×). v1.1 used raw valuation, which treats 'big' as 'no upside' — but a $30B company growing 3× has more expected upside than a flat $400M one.
Valuation / ARR 估值倍数
Max 4
Input: Latest round valuation ÷ ARR. The higher the multiple, the more upside is pre-priced and the less room employee options have
Note: Moved from Financials to Upside and cut from 8 to 4. It's driven by the same (unsourced) rumoured valuation as equity upside; in v1.1 the two sat in different pillars and totalled 16 points, penalising one fact twice.
Liquidity track record 流动性记录
Max 4
Input: Publicly recorded employee tenders / secondary liquidity events
Note: v1.1 had only the 'has record' tier, so absence dropped the metric from the denominator — a free boost for the (almost entirely late-stage) companies with tenders. Late-stage companies now score low for having none: at that size, no employee liquidity is itself information. Early/growth companies are still treated as unknown.
Investor lists (for the Backing pillar)
“Investor quality” matches the leads and participants on funding records against the published lists below. The lists themselves are editorial judgment (✎) — debate welcome.
Tier 1
Tier 2
Versions & known limitations
v2.1 ALPHA · 2026-08 — one change, and it moves scores:
- Reality is switched off again (−18 pts), leaving 6 pillars and 12 metrics. The editorial call is that the pillar as a whole isn't yet backed by enough high-confidence data to move a headline number. The cost is explicit: visa sponsorship no longer counts toward the Career Score, even though it resolves for all 21 companies from the ingested LCA/H-1B/PERM data. Sponsorship is still researched, still shown on every company page, and still filterable on the list — it just doesn't score. No company lost its score in the change and the largest movement was 5 points.
v2.0 ALPHA · 2026-08 — rebalanced after auditing v1.1 against every published company. What changed:
- Reality is scored again (18 pts). It was switched off in v1.1 because Glassdoor data was sparse — but sparse metrics already drop out of the denominator on their own, so that also removed visa sponsorship, which resolves for all 21 companies from the ingested LCA/H-1B/PERM data.
- Provenance now shrinks a weak score toward the middle of its band table instead of multiplying it down toward zero. The old behaviour meant a well-sourced mediocre number beat an estimated excellent one, and it taxed the companies whose researchers bothered to label an estimate.
- Editorial Market/Product scores pass through proportionally instead of being re-bucketed. The v1.1 bands were coarser than the 0–10 input and collapsed all 21 companies into two distinct market scores.
- Added revenue per employee, and raised capital efficiency to 6 pts. Gross margin and NRR aren't public for private companies; these are the closest proxies public data supports — and they finally credit deliberately lean teams.
- “Funding recency” became “Capital validation”: a down round or discounted secondary now overrides recency, where v1.1 gave a down round from three months ago full marks.
- Backing cut from 15 to 8 pts, the valuation multiple moved into Upside and halved (it double-counted the same rumoured valuation as equity upside), and equity upside is now growth-adjusted.
Known limitations (come argue in the community):
- “Investor quality” is still close to a constant — nearly everything on this list is tier-1 backed, and there is circularity in selecting for that and then rewarding it.
- Nothing about day-to-day working life currently scores at all. Glassdoor and work-life-balance reached only 7 of 21 companies, and rather than score a pillar half-covered we switched it off entirely in v2.1 — which also took visa sponsorship and remote policy out of the score despite both being near-fully covered. Collecting the missing employee-experience data is the fix, and it's the clearest thing insiders can help with.
- Funding rounds carry no provenance record, so valuations and investor lists are treated as public-equivalent. The honest fix is sourcing rounds properly, not inventing a discount.
- Market and Product are still editorial scores (✎) — reasoning published in full, to be swapped for hard metrics as community and insider data arrives.
- Stage cohorts use three ARR thresholds rather than a continuous curve — we picked the version that is easier to explain.
- The scoring reference month is fixed at 2026-08; updates ship as whole editions, so ranking moves are the news.
Disagree with a rule, found a data error, or want to add facts about your company → Join the community