How the Space Score is computed

1. The formula

1. For every company, each metric below is read from our tables and classified as measured, missing (it applies, but we do not hold the data) or not applicable (it does not fit the business).
2. Each measured value becomes a percentile rank (0–100) among the companies measured on that metric: the lowest value scores 0, the highest 100, ties share the average of their positions. Because ranks ignore scale, no log transform or outlier cap is needed, and a single giant cannot compress everyone else. Two adjustments follow. Genuine zeros: on the metrics where zero means none at all (space-coded federal obligations and their share, equity raised in the window, launch attempts in the last 12 months), a zero scores 0 rather than sharing a tied middle position with every other zero, and space-coded obligations under the $250K de minimis floor count as zero. Small pools: when fewer than 30 companies are measured on a metric and they are under half of the companies it applies to, each percentile p is pulled toward the middle as (n × p + 10 × 50) ÷ (n + 10), where n is the number measured, so a thin sample cannot hand out a 0 or a 100. A pool that covers nearly every company the metric applies to (the launch record covers every launch provider) is a census, not a sample, and is not shrunk. A genuine zero stays at 0.
3. A pillar score is the average of its measured metrics. One exception: reported revenue can raise the Scale pillar but never lower it. Scale is the larger of the average of headcount and market value and the average of all three, because only about 30 companies report revenue and a penalty for reporting it would favour companies that disclose nothing.
4. The total is the weighted average of the pillars that apply to the company, times 10. Not-applicable pillars are dropped and the remaining weights re-normalised. An applicable pillar with no data counts at the roster median for that pillar: the median score of every scored company measured on it (today: Scale 43.8, Government space demand 0.0, Capital access 50.0, Momentum 50.0, Flight record 43.6). A gap is scored as the typical measured company: it helps a company that would have scored below the median on that pillar and costs one that would have scored above it. The median, not the mean, because Government space demand is split in two (most measured companies score 0, the rest 66 or more), so its mean describes no company. Ignoring the gap instead would credit the company with its own average, rewarding strength on the pillars we can see. A company with no measured pillar is not scored.
5. A company is ranked only with data in at least 2 pillars and at least 50% of its applicable weight backed by data. Otherwise it is scored and shown here, but left out of the leaderboard and the Top 25. Confidence is High at 90% coverage or more with three or more pillars, Medium for other ranked companies, and Insufficient data for unranked ones.
6. A company is ranked only when our data shows measurable space activity: space-coded federal obligations over three years of at least $100M; or at least $10M that is also at least 1% of all its federal obligations; or at least $250K (the de minimis floor) that is at least 10% of its federal obligations; or an orbital launch attempt; or a space-segment profile sector (every sector except general defense) without evidence to the contrary. Evidence to the contrary is $25M or more of federal obligations whose codes can tell space from other work, under 10% of it space-coded, and no launch: a company with that record fails the test whatever its sector. Research and development awards coded to national defense and bought by the Department of the Air Force (which includes the Space Force), or coded to general science and technology and bought by NASA, are left out of that test, because those codes do not say whether the work is space. Companies that fail are scored and listed, but not ranked.
7. Ranking order: total, high first. Equal totals share a rank (shown as “=8”, and the next company takes the following position: 1, =2, =2, 4). Inside a tie, rows are listed by coverage (high first) and then name (A–Z); that orders the list, it is not a ranking.

Pillar weights sum to 100%. There are no other weights, bonuses or adjustments.

Because each pillar contributes 10 × weight × pillar score (or, with no data, its roster median) ÷ the applicable weight, a change in the total splits exactly into per-pillar changes (plus under a point of rounding). The weekly movers edition uses that split, and the stored input values, to say why each company moved.

Rank range. On the current Top 25, on company profiles and on /compare, a ranked company with a pillar we could not measure also shows the ranks that gap likely allows, for example “Rank 12 (likely between 9 and 15 given the pillars we have not measured)”. We recompute its total with each such pillar at the 10th, then the 90th, percentile of the companies measured on it, hold every other company at its published total, and report the ranks those two totals would take. Those are percentile cases with every other company held fixed, not hard bounds, so the rank is likely, not certain, to fall inside them. It is display only: the score, the rank and this methodology do not change, and archived editions show no range.

2. Pillars and metrics

Thresholds used above: hiring change needs 28 days between snapshots and 5+ open roles; the federal trend uses a $250K floor; space-coded obligations under $250K count as zero; metrics measured for fewer than 30 companies, and for under half of those they apply to, are shrunk toward 50 with a weight of 10.

3. Tier labels

• Elite (900–1000): Averages at or above the 90th percentile of the roster across the pillars that apply to it (a pillar without data counts at the roster median).
• Leader (750–899): Averages between the 75th and 90th percentile of the roster on the measured inputs.
• Contender (600–749): Averages between the 60th and 75th percentile of the roster on the measured inputs.
• Emerging (400–599): Averages between the 40th and 60th percentile of the roster: the middle of the pack.
• Early Stage (200–399): Averages between the 20th and 40th percentile of the roster on the measured inputs.
• Developing (0–199): Averages below the 20th percentile of the roster on the measured inputs, usually small or very early companies.

4. Who is scored

Every company profile we track with status active or pre-revenue (386 scored today, 146 ranked). Government agencies (profile sector: agency) are not companies and are not scored.

Held back from scoring

• ball-aerospace: Acquired by BAE Systems in February 2024 and now operates as BAE Systems Space & Mission Systems. Its profile and federal-award search still use the retired name, so its inputs no longer describe the business.
• aerojet-rocketdyne: A subsidiary of L3Harris Technologies, which acquired it in 2023. L3Harris is ranked, so ranking Aerojet Rocketdyne separately would count the same business twice.
• terran-orbital: A subsidiary of Lockheed Martin, which acquired it in 2024. Lockheed Martin is ranked, so ranking Terran Orbital separately would count the same business twice.
• spaceflight-inc: Acquired by Firefly Aerospace in 2023. Its profile still describes the independent company, so its inputs no longer describe the business.
• lumen-orbit: Lumen Orbit renamed itself Starcloud. Starcloud has its own profile, so this one is a duplicate and is not ranked.
• astra-space: Taken private in 2024. The valuation on file dates from its time as a listed company and no longer describes the business, so it is held out of the ranking until its record is refreshed.
• outpost-space: A duplicate of the Outpost Technologies profile: the same company, with the same website (outpost.space). Outpost Technologies holds its funding and federal-award records, so this second profile is not ranked.
• roscosmos: Russia’s state space agency: a state corporation that holds the government’s authority and budget for the national space programme. Government agencies are not ranked (DLR and MBRSC are held out for the same reason), and its own enterprises, such as RSC Energia and Khrunichev, are scored on their own profiles.

Display-name corrections (label only; the score is unaffected)

• amazon-kuiper shown as Amazon Leo: Amazon renamed Project Kuiper to Amazon Leo in 2025.

Launch-provider attribution aliases

Launch Library 2 agency names that differ from our profile names. Every other provider is matched on its exact name.

• “China Aerospace Science and Technology Corporation” → casc: LL2 long name for CASC.
• “China Rocket Co. Ltd.” → casc: CASC commercial-launch subsidiary (Jielong); attributed to CASC.
• “ExPace” → casic: CASIC commercial-launch subsidiary (Kuaizhou); attributed to CASIC.
• “Russian Federal Space Agency (ROSCOSMOS)” → roscosmos: LL2 name for Roscosmos.
• “Northrop Grumman Space Systems” → northrop-grumman: Northrop Grumman launch business (Minotaur, Antares).
• “Orienspace Technology” → orienspace: LL2 long name.
• “Space Pioneer” → space-pioneer: Profile name adds the legal name (Tianbing Technology).

5. Update cadence and editions

• Inputs are measured and scores computed at most once a day (a scheduled run at 02:10 UTC). Every page shows when the inputs were measured, never when the page was rendered.
• A daily reading of every company is stored with the methodology version, forming the Space Score history.
• The quarterly Top 25 edition is named for the quarter in which its inputs were measured. It updates daily during the quarter and freezes when the quarter ends. Rank movement is shown only against a previous edition computed under the same methodology version.
• Any change to a metric, weight, window or rule bumps the methodology version and is recorded in the change log below. Since version 4 a test enforces this: it fingerprints the rulebook and fails if the rules change while the version number does not.

Change log

1. Version 4 · 2026-10-01 · Missing data, zeros, small pools and the share price

   • A pillar that applies but has no data counts at the roster median of that pillar (the median over every scored company measured on it), not at a fixed 50. Most searched companies hold no space-coded federal awards, so the typical measured company sat well below 50 and a gap was a reward: on the last version 2 board, companies we had never searched on USAspending averaged 525 against 481 for searched ones. We use the median, not the mean, because the Government pillar is split in two: most measured companies score exactly 0 and the rest score 66 or more, so its mean (about 25) describes no company. Its median is 0, so a company we could not search is treated like the typical searched company, which holds no federal space work.
   • Reported revenue can raise the Scale pillar but never lower it. Only about 30 companies report revenue, mostly large listed groups, so a smaller company’s revenue ranked low among them: on the last version 2 board, 28 of the 30 scored lower for reporting it (AST SpaceMobile by 98 points, Planet by 60, Rocket Lab by 46) than they would have by publishing nothing.
   • Space-coded federal obligations under $250K over three years count as zero, for both the dollar total and the space share, and a genuine zero (no space-coded awards, no equity raised in the window, no launch in the last 12 months) takes percentile 0 instead of a tied middle position. Before, $1 of space-coded awards lifted a company from the 33rd to the 66th percentile, and a $5K all-space award book read as a 100% space share.
   • A metric measured for fewer than 30 companies, when those are under half of the companies it applies to, has each percentile p shrunk toward 50, as (n × p + 10 × 50) ÷ (n + 10), so a pool of 14 tracked job boards can no longer hand out a 0 or a 100. The launch metrics cover every launch provider, so they are not shrunk.
   • The 90-day share-price return is no longer an input. It measured investor sentiment rather than the company’s activity, and for many listed companies it was the only Momentum input. A company with no other Momentum data now has the pillar missing, counted at the roster median. A consequence: 11 listed companies, mostly outside the U.S., no longer have enough measured pillars to rank (on the 30 September inputs: Safran, Korea Aerospace Industries, MDA Space, SES, Eutelsat, Telesat, OHB, Synspective, AAC Clyde Space, Gilat Satellite Networks and Virgin Galactic). They are still scored and listed, with the reason shown.
   • A company with no measured pillar has no score at all and shows “Not scored”, rather than a score of 0 with the lowest tier.
   • An equity round in the last 24 months whose size was not disclosed is no longer read as $0 raised. When a company raised in the window but no round there states an amount, Capital raised is missing; when some rounds state an amount, only those are added, and the reading says so.
   • Space-activity gate: a company now needs space-coded federal obligations of $100M at any share of its federal work, or $10M that is also at least 1% of its federal obligations, or at least $250K (the de minimis floor) that is also at least 10% of them (before, $10M was enough at any share). Research and development awards coded AC (national defense) and bought by the Department of the Air Force, which includes the Space Force, or coded AJ (general science and technology) and bought by NASA, say nothing about whether the work is space, so they are no longer counted as evidence against space activity. The $250K de minimis floor applies to the space-coded figures.
   • Market value: subsidiaries and joint ventures have no market value of their own, and neither does a listed profile that names a parent company, because the ticker on it is the parent’s: such a profile is not scored on that ticker (Hanwha Systems, whose own shares trade separately from its parent’s, is the one listed exception). When two profiles that each claim a ticker as their own share it, neither is scored on it, so a ticker never counts twice. Market caps of companies listed outside the U.S. are converted to U.S. dollars at that day’s exchange rate. Undated valuations, and valuations from rounds more than 36 months old, are not used.
   • Reported revenue: one figure per fiscal year, taking a figure backed by a filing first, so duplicate or conflicting rows can no longer decide the value by the order the database returns them.
   • Roster: the Outpost Space profile (a duplicate of Outpost Technologies) and Roscosmos (Russia’s state space agency) are not ranked, each with the reason shown.
   • The 12-month federal space trend is not applicable when neither 12-month period reaches the $250K de minimis floor, the same threshold the Government pillar uses: a first $203K award against nothing no longer reads as strong growth while the Government pillar counts it as zero.
   • Equal totals share a rank (shown as =8) everywhere, including the weekly edition, the quarterly Top 25 and this page; coverage and then name only order the rows inside a tie.
   • Headcount from an open-ended size band (such as 10,000+) is the band’s lower bound, and the reading says so.
   • A test now fingerprints the whole rulebook (metrics, weights, rules, thresholds, code lists, roster exclusions, aliases, input rules, queries and the scorer’s output on a fixed test roster) and fails unless a change ships with a new version number and an entry on this list.

2. Version 3 · 2026-09-30 · Audit corrections to the inputs

   • Launch records count orbital attempts only: New Shepard, HASTE (Rocket Lab’s suborbital Electron variant), SpaceShipTwo and Starship test flights before Flight 14 are suborbital and no longer count.
   • Capital raised counts equity rounds only, using the funding tracker’s own test: debt and credit facilities, grants, owner funding, acquisitions, secondaries and public-market deals are excluded, and only an equity round on record makes the Capital pillar apply.
   • Self-reported funding rounds count only after an editor approves them, and paid or user-posted job listings never reach the hiring input.
   • The government rule recognises the service forms of the space-vehicle product codes (W018, J018, K018, L018, N018, H118, H218, H318, H918).
   • Subsidiaries ranked beside their parents, and duplicate or stale profiles, are held out of the ranking with the reason shown.

3. Version 2 · 2026-09-23 · Computed from sourced data

   • The hand-maintained table was replaced by a computed composite: five weighted pillars, each metric a percentile rank among the companies measured on it, an applicable pillar with no data counted at 50, ranking only with data in at least 2 pillars and half the applicable weight, and a space-activity gate.
   • Changed on 2026-09-25 without a version bump, which our own rule forbids: a partial launch failure stopped counting as a success, and the hiring window started restarting on the day the way a company’s roles are counted changed. Readings before and after 25 September are both labelled version 2. The rulebook test added in version 4 exists to stop this happening again.

4. Version 1 · until 2026-09-23 · Hand-maintained table (retired)

   • Scores were a table of hand-set numbers. Readings made under it carry methodology version 1 in our history and are never compared with later readings.

6. Known limitations

Said plainly, from today’s figures

• Government space demand: the roster median is 0, because most companies we have searched hold no space-coded federal awards. So a U.S. company we have not matched in USAspending (not searched, or similar names our matcher did not attribute) is treated like the typical company: one with no federal space work. If it does hold federal space contracts, its score understates them.
• Momentum, for most companies measured on it, is the 12-month federal space trend alone. We track the job boards of 52 companies, and today 3 of them have a measurable like-for-like 90-day change in open roles (enough history on one counting basis and enough open roles); so of the 40 companies measured on Momentum, 37 are measured on the federal trend alone. One award arriving or ending can swing that trend, so Momentum is noisy.
• A company listed outside the U.S. usually has Government not applicable (no U.S. federal award record) and Capital not applicable (it is listed), so unless a second pillar is measured it is scored but not ranked. Today that is 18 companies: AAC Clyde Space, APT Satellite, Creotech Instruments, Dassault Aviation, Elbit Systems, Eutelsat, Gilat Satellite Networks, Hanwha Systems, ispace inc., Korea Aerospace Industries, MDA Space, OHB SE, Safran, SES, Space42, Synspective, Telebras, Telesat.
• Companies that pass the space-activity gate but are not ranked on thin data, today, with data in fewer than 2 pillars: AAC Clyde Space, Aerospacelab, Antaris, Applied Atomics, APT Satellite, Arabsat, Asteroid Mining Corporation, Astrolab, Bellatrix Aerospace, BioOrbit, Blackswan Space, Bradford Space, Cailabs, Carbon Mapper, Cascade Space, Charter Space, ClearSpace SA, Constellation Space, Constellr, Copenhagen Suborbitals, Cosmoleap, Creotech Instruments, Dassault Aviation, deltaVision, Dhruva Space, Digantara, Dispatch, ElevationSpace, Embratel Star One, ENPULSION, Eutelsat, Exotrail, Flexell Space, Flexitech Aerospace, FOSSA Systems, GHGSat, Gilat Satellite Networks, GITAI, HEO, Infostellar, Insera Solution, InspeCity, Interstellar Technologies, iSpace (China), ispace inc., ISPTech, Jiuzhou Yunjian, Kepler Aerospace, Kirkhill, Korea Aerospace Industries, Lanteris Space Systems, Leanspace, LEAP Space, LegendSpace, Letara, Lonestar Data Holdings, Longshot Space, Manastu Space, MDA Space, NanoAvionics, NewOrbit, OHB SE, Open Cosmos, Optera, OrbitAID Aerospace, Orbital Matter, Orienspace, ORiS, Outlier Space, Oxford Space Systems, Pale Blue, Pangea Aerospace, PAVE Space, Perigee Aerospace, PierSight Space, Pulsar Fusion, QOSMIC, Quadsat, Ramon.Space, Reditus Space, Reflex Aerospace, Ripple Aerospace, Rocket Factory Augsburg, RSC Energia, RUAG Space, Safran, Sateliot, Satrec Initiative, SatSure, SatVu, SES, Skybound Medtech, SkyFi, Skykraft, Skynopy, Skyrora, Southern Launch, Space Pioneer (Tianbing Technology), Space42, Spaceflux, SpaceIL, SpacePharma, Spacety, SSC, Stellar Alpina, SWISSto12, Synspective, TakeMe2Space, TASI Group, Telebras, TelePIX, Telesat, Telespazio, Thales Alenia Space, The Exploration Company, TiSPACE, U-Space, Ubotica, Unastella, UNIVITY, Ursa Space Systems, Virgin Galactic, Vyoma, Wyvern, Xoople, Zenk Space, Zero 2 Infinity; with under 50% of their applicable weight backed by data: Aethero, Agile Space Industries, Antares Industries, Array Labs, Astro Digital, Beyond Reach Labs, Citra Space, Diffraqtion, Endeavor Optical Networks, Infinite Orbits, Juno Propulsion, Kapta Space, Lux Aeterna, Magdrive, mPower Technology, Northwood Space, NovaWurks, Observable Space, Orbital, Orbite, OroraTech, Quindar, Ravee Optics, Samara Aerospace, Sophia Space, Xona Space Systems, Zeno Power; with no measured pillar at all, so “Not scored”: Aldoria, Archangel Lightworks, ArcSpace, ArkEdge Space, Aspect Aerospace, Astradyne, ATMOS Space Cargo, Basalt, Chang Guang Satellite Technology, Chuanyuezhe, Circuits Integrated Hellas, DCUBED, Deep Blue Aerospace, EDGX, EtherealX, ExLabs, GalaxySpace, Geespace, Hongqing Technology, Indra Space, ISISPACE, JP Donovan, LC60, LiveEO, Look Up Space, MaiaSpace, Marlan Space, Maverick Space Systems, MinoSpace, Nayuta Space, Neo Space Group, NordSpace, Reaction Dynamics, Satleo Labs, SEOPS, Space BD, Space Epoch, SpaceSail (Yuanxin Satellite), StarRF, Sybilla Technologies, TerraByte AI, TerraSpark, Trusted Space, Unseenlabs, Venturi Space, Xingce Weilai, Xingchen Daohe, Xingyi Lianxin, Xiphera, Z-Trak Space, Zenno Astronautics. Each one’s breakdown says which inputs are missing and why.

• Scale measures the whole company. A diversified group (Lockheed Martin, Honeywell, Safran) is scored on its full headcount, revenue and market value, not on its space business alone, because none of our sources splits them out. Government space demand counts space-coded awards only, and the space-activity gate keeps companies with no measurable space activity out of the ranking: those are the corrections for this.
• The space-activity gate can only use what we measure. A company in the general defense sector whose space work we cannot see (not searched on USAspending, headquartered outside the U.S., or with its awards under a legal name we do not attribute) is scored but not ranked, even when it is a real space business. The fix is better input data, not an exception for the company.
• A company can pass the space-activity gate and still go unranked on thin data: ranking needs data in at least 2 pillars and at least 50% of its applicable weight backed by data. A joint venture or subsidiary has no market value or venture rounds of its own, and a company headquartered outside the U.S. with no U.S. federal award has Government not applicable, so such a company can be left with data in only one pillar. Which companies this affects today is listed above, from the current figures.
• Government space demand sees U.S. federal prime awards only: no other governments, no subcontracts, and no classified work (much NRO and some Space Force spending never appears on USAspending.gov). Space-coding follows the government’s own codes and is a conservative floor: space work filed under a non-space code (for example the mixed “guided missile and space vehicle” NAICS codes) is not counted.
• Our award matcher prefers a missed award to a wrong one. Companies whose names are too generic to search safely are not searched, and some companies’ federal work sits under legal names we do not yet attribute (for example Raytheon’s and Collins Aerospace’s awards appear under RTX and Rockwell Collins entities). Where the search saw similar-named recipients but attributed none, the metric is treated as missing rather than zero; where it saw none at all, a zero is recorded.
• Headcount and undated profile valuations are maintained on company profiles and can lag the company by months. Reported revenue exists only for companies that publish it, which excludes most private companies.
• The hiring metric covers only companies whose applicant-tracking boards we snapshot daily, and needs 28 days of like-for-like history and 5 or more open roles, so for most companies it is missing (today’s counts are above).
• The launch record begins on 2025-01-01. Launches are attributed to the provider named by Launch Library 2; subsidiary brands are attributed to the parent we profile (listed below). Blank-provider launches are not attributed.
• A subsidiary or joint venture, or a listed profile that names a parent company, has no market value of its own and Capital raised does not apply to it: the ticker on such a profile is the parent’s (Collins Aerospace and Raytheon both carry RTX), so neither is scored on RTX’s market value. Its Scale rests on its own headcount and on revenue it reports itself, never on the parent’s SEC filings.
• A pillar with no data counts at the roster median, which is an estimate, not a measurement: for a company with a gap, part of its total describes the typical measured company rather than the company itself. The gap helps a company that would have scored below the median on that pillar and costs one that would have scored above it. The Data column and every breakdown show how much of each score is measured, and ranking requires at least half of the applicable weight to be measured.
• Reported revenue can only raise the Scale pillar, never lower it. That favours the roughly 30 companies that report revenue over those that do not; the alternative we replaced, a penalty for reporting, was worse.
• Small-pool shrinkage switches off at 30 companies measured, or when they reach half of the companies the metric applies to, so a pool crossing either line loses it at once: the extremes of that metric jump by 500 ÷ (n + 10) percentile points, where n is the number measured (12.8 for a pool growing from 29 to 30, 20.8 for a 14-company pool reaching half coverage, more for smaller pools).
• Every metric is a percentile rank within the tracked roster, so a company’s score can move when other companies’ data changes, and it says nothing about companies we do not profile.
• Not used, because we do not hold them reliably: satellite fleet counts (our satellite-asset table is empty), patents, news mentions (company tagging covers too few companies), and insider or 5%-holder filings (held for listed companies only, and not a measure of the business).
• The score is descriptive. It is not a forecast, a credit rating or investment advice.

7. The exact queries

These are the statements the scorer runs against our PostgreSQL database, verbatim. Funding rounds are read one by one and our code then keeps only equity rounds, using the same test as the funding tracker. Federal awards are read one by one and our code then classifies each as space-coded or not, using the rule in the Government space demand metric above.

select
  c.slug, c.name, c.sector, c.country, c."isPublic" as "isPublic", c.ticker, c.exchange,
  c."ownershipType" as "ownershipType", c."parentCompany" as "parentCompany",
  c."employeeCount"::float8 as "employeeCount", c."employeeRange" as "employeeRange",
  c."marketCap"::float8 as "marketCap", c."lastVerified" as "lastVerified",
  c.valuation::float8 as "profileValuation",
  coalesce(rv.rows, '[]'::json) as "reportedRevenueRows",
  pv.val as "roundValuation", pv.date as "roundValuationDate",
  (fac.status = 'searched') as "federalSearched",
  coalesce(fac."recipientsSeen", 0) as "federalRecipientsSeen",
  coalesce(fac."recipientsMatched", 0) as "federalRecipientsMatched",
  coalesce(fr.kinds, '[]'::json) as "fundingRoundKinds",
  js_start."activeJobs" as "openRolesStart", js_start.date as "openRolesStartDate",
  js_end."activeJobs" as "openRolesLatest", js_end.date as "openRolesLatestDate",
  jrs.since as "openRolesBasisSince"
from "CompanyProfile" c
left join "FederalAwardCoverage" fac on fac."companyId" = c.id
left join lateral (
  -- every company-reported annual figure on record; the loader picks one per
  -- fiscal year with selectReportedRevenue() (src/lib/space-score-data.ts):
  -- duplicates, conflicting figures and rows of unknown currency are settled
  -- there by a published rule, never by the order Postgres returns them in
  select json_agg(json_build_object(
    'id', r.id, 'year', r.year, 'revenue', r.revenue::float8, 'source', r.source,
    'notes', left(coalesce(r.notes, ''), 200), 'createdAt', r."createdAt"
  ) order by r.year desc, r.id) as rows
  from "RevenueEstimate" r
  where r."companyId" = c.id and r.revenue is not null and r.revenue > 0
    and r."confidenceLevel" = 'reported' and r.quarter is null
) rv on true
left join lateral (
  select f."postValuation"::float8 as val, f.date
  from "FundingRound" f
  where f."companyId" = c.id and f."postValuation" is not null and f."postValuation" > 0
    and (f.source is null or f.source <> 'self_reported' or f."reviewStatus" = 'editor-approved') and coalesce(f."reviewStatus", '') not in ('self-reported-pending', 'self-reported-rejected')
  order by f.date desc limit 1
) pv on true
left join lateral (
  -- every published round's kind, at any date: the loader counts the rounds
  -- that pass countsTowardCapitalRaised() (equity only), and that count
  -- decides whether the Capital pillar applies; the capital SUM comes from
  -- SPACE_SCORE_FUNDING_SQL
  select json_agg(json_build_object('roundType', f."roundType", 'seriesLabel', f."seriesLabel")) as kinds
  from "FundingRound" f where f."companyId" = c.id
    and (f.source is null or f.source <> 'self_reported' or f."reviewStatus" = 'editor-approved') and coalesce(f."reviewStatus", '') not in ('self-reported-pending', 'self-reported-rejected')
) fr on true
left join lateral (
  -- the latest change to the basis the company's roles are counted on: the
  -- space-roles rule applied, a new job board joining a series already under
  -- way, or the snapshot's source label changing (HIRING_BASIS_SINCE_SQL in
  -- src/lib/jobs/hiring-basis-sql.ts, the twin of hiringBasisChanges()); the
  -- hiring window restarts on that day, so a board joining is never growth
  select greatest(
    -- 1. rule-changed
    (select max(l."createdAt")::date from "DataRefreshLog" l
      where l.module = 'jobs-coverage'
        and l.details like '%"kind":"relevance-changed"%'
        and l.details like '%"companyProfileSlug":"' || c.slug || '"%'),
    -- 2. board-joined: a board whose first posting came after the series began
    (select max(b.first_seen::date + 1) from (
        select min(j."createdAt") as first_seen from "SpaceJobPosting" j
        where j."companyProfileId" = c.id
          and j.source in ('greenhouse', 'lever', 'ashby', 'workday', 'clearcompany', 'pinpoint', 'bamboohr', 'rippling', 'usajobs')
          and j."paidAt" is null and j."postedByUserId" is null
        group by j.source
      ) b
      where b.first_seen::date > (select min(s0.date) from "CompanyJobSnapshot" s0 where s0."companyProfileId" = c.id)),
    -- 3. source-label-changed
    (select max(x.date) from (
        select s1.date, s1.source,
               lag(s1.source) over (order by s1.date) as prev,
               row_number() over (order by s1.date) as rn
        from "CompanyJobSnapshot" s1 where s1."companyProfileId" = c.id
      ) x
      where x.rn > 1 and x.source is distinct from x.prev)
  ) as since
) jrs on true
left join lateral (
  select s."activeJobs", s.date from "CompanyJobSnapshot" s
  where s."companyProfileId" = c.id and s.date >= (now() - interval '90 days')::date
    and s.date >= coalesce(jrs.since, date '1900-01-01')
  order by s.date asc limit 1
) js_start on true
left join lateral (
  select s."activeJobs", s.date from "CompanyJobSnapshot" s
  where s."companyProfileId" = c.id
  order by s.date desc limit 1
) js_end on true
where c.status in ('active', 'pre-revenue')
order by c.slug

select
  e.agency,
  count(*)::int as "attemptsOnRecord",
  count(*) filter (where e."launchDate" >= now() - interval '365 days')::int as "attempts12m",
  -- Full successes only: a partial failure is stored as completed with the LL2
  -- label in feedStatus, and is an attempt, not a success (the same test as
  -- isPartialFailureLabel in src/lib/launch-outcome.ts; 2026-09-25).
  count(*) filter (where e."launchDate" >= now() - interval '365 days' and e.status = 'completed' and coalesce(e."feedStatus", '') !~* 'partial')::int as "successes12m"
from "SpaceEvent" e
where e.status in ('completed', 'failed')
  and e.rocket is not null
  -- orbital attempts only: no suborbital orbit, no suborbital-only vehicle,
  -- no Starship test flight before the first orbital one
  and coalesce(e."orbitType", '') !~* '^sub'
  and e.rocket !~* '^(new shepard|haste|spaceshiptwo|spaceship two)'
  -- HASTE flies under rocket 'Electron': read the vehicle from the name too
  and trim(split_part(coalesce(e.name, ''), '|', 1)) !~* '^(new shepard|haste|spaceshiptwo|spaceship two)'
  and not (e.rocket ~* '^starship' and e."launchDate" < '2026-09-28')
  and e."launchDate" >= '2025-01-01'
  and e."launchDate" <= now()
group by e.agency

select
  c.slug,
  f.amount::float8 as amount,
  f.date,
  f."roundType" as "roundType",
  f."seriesLabel" as "seriesLabel",
  c."isPublic" as "isPublic"
from "FundingRound" f
join "CompanyProfile" c on c.id = f."companyId"
where (f.source is null or f.source <> 'self_reported' or f."reviewStatus" = 'editor-approved') and coalesce(f."reviewStatus", '') not in ('self-reported-pending', 'self-reported-rejected')
  and f.date >= now() - interval '24 months'
  and c.status in ('active', 'pre-revenue')

select
  c.slug,
  a.amount::float8 as amount,
  a."actionDate" as "actionDate",
  a."pscCode" as "pscCode",
  a."naicsCode" as "naicsCode",
  a."cfdaProgramTitle" as "cfdaProgramTitle",
  a."awardingAgency" as "awardingAgency",
  a."awardingSubAgency" as "awardingSubAgency"
from "FederalAward" a
join "CompanyProfile" c on c.id = a."companyId"
where a."countsTowardTotals"
  and a.amount is not null
  and a."actionDate" >= now() - interval '3 years'
  and a."actionDate" <= now()
  and c.status in ('active', 'pre-revenue')

8. Every company’s breakdown

Pillar values are 0–100; “no data” counts at that pillar’s roster median; “n/a” is dropped; a company with no measured pillar shows “Not scored”. Choose a company for every metric’s raw value, percentile and the reason any input is missing.

The Space Score ranks companies relative to each other on public data we hold. It describes; it does not rate, forecast or recommend. It is not a credit rating and not investment advice. Full disclaimer