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 44.6, 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 space revenue in the company’s own SEC annual report (its latest 10-K, 20-F or 40-F on its own CIK, or before its first annual report a Form 10, S-1 or F-1 with audited annual statements, for a fiscal year that ended within the last 2 calendar years): a full-year revenue fact tagged with a business line (not a customer, counterparty, legal entity, acquired business or related party) whose label names space and nothing else (it contains the word space, or satellite, spacecraft, launch, orbital or lunar, and none of defense, aerospace, aviation, aircraft, airborne, missile, munition, security, cyber, military, naval, government, intelligence or industrial), taking the largest such line, of at least $100M, or at least $10M that is also at least 1% of the company’s total revenue, or at least $250K that is at least 10% of it, counted only after a second check that the stored figure matches the filing; or, for a company we have searched on USAspending, space-coded federal work counting verified subcontracts: its space-coded prime obligations plus subcontracts reported under space-coded prime awards to one of its registered UEIs (every reported action summed, exact duplicates dropped, a subcontract reported above its prime award’s total obligation discarded), against all its prime obligations plus all its subcontracts over three years, passing the same three thresholds, counted only after a second check confirms the subcontract totals, and only while both the confirmed totals and today’s totals pass; or a space-segment profile without evidence to the contrary. A space-segment profile is one whose sector is a space segment (launch, satellite, ground segment, infrastructure, analytics, manufacturing, components, exploration and the like); a company our record marks as cross-domain, meaning a defense, defence, aerospace, government or military sector, passes only if its subsector names space work (it contains space, orbit, orbital, satellite, launch, lunar or spacecraft), and the subsector names the company’s main line of business; a diversified aerospace group, whose space business is one line among many, passes only if its subsector names space work and that space line’s revenue is cited from the company’s own annual report or results at $100M or more, counted only after a second check. 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. Space revenue, subcontracts and launches are positive evidence that the test of evidence to the contrary never overrides; that test reads prime awards only, and it overrides every profile label. A revenue line that combines space with defense (such as “Defense and Space”) is never evidence of space activity. 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

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 (487 tracked today, 357 scored, 93 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.

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 6 · 2026-10-07 · The profile test reads the subsector, and diversified groups need a cited space line

   • Space-activity gate, profile test: a company with no measured space evidence passes on its profile only when its own record says its business is space. Until now any sector except general defense passed, so a diversified group filed under manufacturing passed on that label while a defense-sector company whose whole business is space did not. The test is now the one the funding editions already use: a space-segment sector passes; a company our record marks as cross-domain, meaning a defense, defence, aerospace, government or military sector, passes only if its subsector names space work (it contains space, orbit, orbital, satellite, launch, lunar or spacecraft); and the subsector names the company’s main line of business, not a side line.
   • A diversified aerospace group, whose space business is one line among many (the classification the space stocks page already uses for HEICO, Airbus, Safran and others), is treated as cross-domain whatever its sector. It passes on its profile only if its subsector names space work and that space line’s revenue is cited from the company’s own annual report or results at $100M or more, entered with its sources and counted only after a second check. Every cited line, confirmed or pending, is listed in the evidence register on this page.
   • The test of evidence to the contrary is unchanged and still overrides every profile label. A revenue line that combines space with defense, such as “Defense and Space”, is never evidence of space activity; where a cross-domain company reports space only inside such a line, its unranked reason now says that space revenue is not reported separately from defense.

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 51 companies we score, 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 20 companies: AAC Clyde Space, Airbus Defence & Space, APT Satellite, Creotech Instruments, Dassault Aviation, Elbit Systems, Eutelsat, Gilat Satellite Networks, GomSpace, Hanwha Systems, ispace inc., Korea Aerospace Industries, MDA Space, NorthStar Earth & Space, OHB SE, Safran, SES, 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, AADYAH Aerospace, Adaspace, Advanced Navigation, Aerospacelab, Airbus Defence & Space, AIRMO, Amazon Leo, Antaris, Applied Atomics, APT Satellite, Archangel Lightworks, ArcSpace, Arinna, ArkEdge Space, Arkisys, Aspect Aerospace, Astronstone, AstroX, Aule Space, BAAS Technologies, Bellatrix Aerospace, BioOrbit, Blackswan Space, Canada Rocket Company, Capsule Corporation, Carbon Mapper, Cascade Space, ClearSpace SA, COMSPOC, Constellation Space, Constellr, cosmobloom, Cosmoleap, Creotech Instruments, DCUBED, Deep Space Energy, deltaVision, Diffraqtion, Dispatch, Dream Aerospace, ElevationSpace, ENPULSION, EtherealX, Eutelsat, Flexell Space, Flexitech Aerospace, Fortastra, FOSSA Systems, Foundational, Galactic Energy, GalaxEye, Galaxia, GHGSat, Gilat Satellite Networks, GITAI, Golbriak Space, GomSpace, HEO, Hydrosat, HyImpulse, Insera Solution, InspeCity, InterGravity Technologies, iSpace (China), ispace inc., ISPTech, JAOPS, Jingwei Rocket, Kepler Aerospace, Kepler Communications, Khrunichev State Research and Production Space Center, Kirkhill, Latitude, Leaf Space, LEAP Space, LegendSpace, Letara, Lonestar Data Holdings, Longshot Space, MaiaSpace, Manastu Space, MDA Space, MID, MJOLNIR SPACEWORKS, Morpheus Space, NanoAvionics, Nebex, Neuraspace, New Space Intelligence, NewOrbit, Nobility Space, NorthStar Earth & Space, OffWorld, OHB SE, OQ Technology, Orbit Fab, OrbitAID Aerospace, Orienspace, ORiS, Outlier Space, Oxford Space Systems, Pale Blue, Pangea Aerospace, PAVE Space, Perigee Aerospace, Pixxel, Pustar Aerospace, QOSMIC, Quadsat, Ramon.Space, Reaction Dynamics, Reditus Space, Reflex Aerospace, RUAG Space, Sateliot, Satlabs Space Systems, Satlyt, SatSure, SatVu, SES, Sift, SkyFi, Skykraft, Skyrora, Southern Launch, Space BD, Space One, Space Tech Accelerator, Spaceflux, SpaceNav, SPEX, SSC, Star Signal Solutions, Starlab Space, Stellar Alpina, STEPLab, SUiCTE, SWISSto12, Synspective, TakeMe2Space, Telebras, Telesat, Telespazio, TerraByte AI, Thales Alenia Space, The Compression Company, TransAstra, Trusted Space, TUSK IC, Ubotica, Unastella, Unmanned Exploration Laboratory, Vantor, Virgin Galactic, Virtus Solis, Warpspace, Wyvern, Xoople, Xovian Aerospace; with under 50% of their applicable weight backed by data: Aalyria, Aethero, Agile Space Industries, Albedo, Antares Industries, Apogee Semiconductor, Array Labs, Astro Digital, AstroForge, ATLAS Space Operations, Besxar, Beyond Reach Labs, Busek Co., Catalyx Space, CesiumAstro, Charter Space, CisLunar Industries, Citra Space, Cowboy Space Corp, E-Space, Elveo Mobile, Endeavor Optical Networks, Epsilon3, Fenix Space, GeoOptics, Hubble Network, i-Space, Juno Propulsion, Kapta Space, Katalyst Space, Kayhan Space, LeoLabs, Lux Aeterna, Magdrive, Mangata Networks, Max Space, mPower Technology, Northwood Space, NovaWurks, Orbital, Orbital Sidekick, Orbite, Outpost Technologies, Overview Energy, Phantom Space, PlanetiQ, Privateer, Quindar, Ravee Optics, RBC Signals, Reflect Orbital, Relativity Space, Samara Aerospace, Scout Space, Skybound Medtech, Skylo Technologies, Slingshot Aerospace, Sophia Space, Space Pioneer (Tianbing Technology), SpiderOak, SpinLaunch, Star Catcher Industries, ThinkOrbital, Tomorrow.io, Umbra, Ursa Space Systems, Varda Space Industries, Xona Space Systems, Zeno Power; with no measured pillar at all, so “Not scored”: Aavuus, Another Earth, Arabsat, Arkadia Space, Asteroid Mining Corporation, Astradyne, AstraMatrix, Astrolab, ATMOS Space Cargo, AZSPACE, Basalt, Beichen Hangtian, Biyi Jiguang, BlueStar Optical Domain, Bradford Space, Chang Guang Satellite Technology, Chuanyuezhe, Circuits Integrated Hellas, Copenhagen Suborbitals, Deep Blue Aerospace, Deep Space Matrix, Dongsheng Yuhang, EDGX, Emposat, EXA Tech, ExLabs, Exolaunch, GalaxySpace, Geespace, Guangyou Xingkong, Hangyu Servo, Helogen, HeXing Optical Link, Hongqing Technology, IENAI Space, Indra Space, InfinAstro, Innovative Space Carrier, InterCosmos, ISISPACE, JG Starcom, Jianqi, Jitian Xingzhou, Jiuzhou Yunjian, JP Donovan, Juneng Tianqing, Kick Space Technologies, LC60, LingLume, Lingque Weixing, LiveEO, Mantis Space, Marlan Space, Maverick Space Systems, MinoSpace, Mission Control Space Services, Miwei Technology, Nayuta Space, Neo Space Group, NordSpace, Optera, Orbital Chenguang, Orbital Lasers, Orbital Matter, Orbital Sentry, PierSight Space, Pramatra Space, Project-S, Pulsar Fusion, Qingbo Aerospace, Ripple Aerospace, Rivada Space Networks, Rocket Factory Augsburg, RSC Energia, SaeFarm, Sanyark Space, Satleo Labs, Satrec Initiative, SEOPS, Shuowei Optoelectronics, Sichuang Aerospace, Solafune, Space Epoch, Space42, Spacebackend, SpaceIL, Spacelab, SpaceLocker, SpacePharma, SpaceSail (Yuanxin Satellite), Spacety, Spark Space, StarRF, Starry Development, Stellar Power, Stellaria, Sybilla Technologies, Taikong Qiyuan, TelePIX, TerraSpark, Tianxing Tansuo, Tianyu Xingkong, TiSPACE, Tuqiang Yuhang, UniStarCom, Venturi Space, Weifen Hangyu, Weiguang Qihang, Weiyi Xinkong, Xiandao Jixing, Xingce Weilai, Xingchen Daohe, Xinghan Juneng, Xingji Kaifa, Xingji Yuandian, Xinglian Tianshu, Xingweixun, Xingyao Guangyu, Xingyi Lianxin, Xingyi Space, Xingyi Xinneng, Xingyuan Zongheng, Xiphera, Yiwei Yuhang, Yunyao Aerospace, Yusuan Xinghe, Z-Trak Space, Zenk Space, Zenno Astronautics, Zero 2 Infinity. Each one’s breakdown says which inputs are missing and why.

Evidence register

The space-activity gate’s SEC, subcontract, cited-line and profile-label evidence, entry by entry. Space revenue is read from the XBRL data of each company’s own annual report by the same scan for every company with a CIK; an entry counts only once a second check has confirmed that the stored figure matches the filing. Subcontracts are read from USAspending.gov by registered UEI only, and a company’s subcontract totals count only once a second check has confirmed them, and only while they still match. Nothing here changes a score: it decides only whether a company can be ranked.

Space revenue in SEC annual reports

• Astra Space (Pending confirmation, not counted)

  Business line “Space Products”: space revenue $3.9M of $3.9M total revenue, fiscal year ending 2023-12-31. Meets the gate’s thresholds.

  Form 10-K, CIK 0001814329, accession 0000950170-24-045209; fact us-gaap:RevenueFromContractWithCustomerExcludingAssessedTax with srt:ProductOrServiceAxis = astr:SpaceProductsMember.

• Comtech Telecommunications (Pending confirmation, not counted)

  Business line “Satellite And Space Segment Communications”: space revenue $269.3M of $499.5M total revenue, fiscal year ending 2025-07-31. Meets the gate’s thresholds.

  Form 10-K, CIK 0000023197, accession 0000023197-25-000080; fact us-gaap:Revenues with us-gaap:StatementBusinessSegmentsAxis = cmtl:SatelliteAndSpaceSegmentCommunicationsMember.

• EchoStar (Pending confirmation, not counted)

  Business line “Broadband and Satellite Services”: space revenue $1,456.1M of $15,005.0M total revenue, fiscal year ending 2025-12-31. Meets the gate’s thresholds.

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.subsector, c.tags, 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

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.