25 October 2026. manage.py eligibility_scan (read-only) runs crawl/procurement_signals.restrictions over the 45,554 documents the site counts and keeps restrictions found inside an eligibility clause (申报条件, 支持对象, 资格要求, 投标人须 …). Output: reports/data/eligibility.csv (one row per document and type, with the clause) and reports/eligibility.json (counts and rates, read by the Insights card "Who may apply", Compliance view).
Hand check (local registration or establishment)
Three random samples of 30 hits each, read in full clause context:
| Sample | Correct | Misses |
|---|---|---|
| 1, before guards | 25/30 | central rule banning the practice; fair-competition and negative-behaviour lists; a product registrant; a support line *for* setting up a studio |
| 2, after ban guards | 28/30 | athlete registration; conditions for foreign arbitration bodies |
| 3, new named-place pattern (在武汉市行政区域内注册) | 28/30 | two government-reorganisation clauses (要求我省成立相应机构), now excluded |
Samples 2 and 3 give the precision quoted on the card: 56/60 (93%). A vehicle that must be registered locally to earn a subsidy counts as a local condition.
Recall is not measured. Before the named-place pattern, the screen missed about 200 documents that name the place instead of saying 本市. Clauses in unread attachments are still missed, so the shares are lower bounds.
Other types
capital_threshold: 12/15 correct (misses: criteria for zones and clusters, and a notice *removing* a capital requirement). Size thresholds are not local preference and are not on the card.sme_reserved: 13/20 before the fix; most misses were 本项目非专门面向中小企业 ("*not* reserved"), now excluded. SME set-asides are national procurement policy (财库〔2020〕46号), open to foreign-invested SMEs.foreign_exclusion: market-access law (foreign investment banned in broadcasting, legal services), not conditions on applicants to a programme. Reported in the CSV, not on the card.ownership: 0/3, off the card.domestic_only: 2/2. These are the 2016 rehabilitation-equipment rule "国产产品能够满足要求的原则上须采购国产产品" (buy domestic where domestic products suffice), central and Liaoning.no_consortium: tender boilerplate. Its 2025–26 jump reflects which tender notices were collected.
Results (rates among 4,681 non-tender documents with an eligibility section, 2021–26)
- 5.2% require local registration or establishment (243 documents).
- By level: county 10.8%, province 5.9%, prefecture 3.7%, central 0.9%.
- 2021–24: 5.9%; 2025–26: 4.7%. The intervals overlap, so there is no clear fall after the 2024 Fair Competition Review Regulation, in contrast with the fall in explicit local preference on the fair-competition card.
By province (added 26 October 2026)
eligibility_scan now also writes, for local-registration conditions, the share of held 2021–26 programme documents by the issuer's province (rates.local_presence.by_province_2021_26). The site shows these figures on a tile map with every clause: /provinces/who-may-apply/. The highest shares with at least 30 programmes held are Shanghai (96 of 620, 15%, 95% interval 13–19%), Beijing (13 of 100, 13%) and Sichuan (37 of 451, 8.2%). Guangdong, Chongqing, Guangxi, Shanxi and Tibet show none.
Each figure is a share of the programmes we hold, which depends on what was collected. For example, Shanghai's portal crawl includes many 专项资金申报指南 (special-fund application guidelines), which are the documents that state eligibility. Read the figures as "in the held programmes", not as a ranking of provinces' openness.