SMRMap field guide ================= This file explains how to read the public downloads. It does not change the records. Quoted sources remain their owners'. The dataset licence is CC BY 4.0; see LICENSE.txt. Start here ---------- Open reporting-projects.csv for a readable project report: one row per published project. Open this field guide for the column rules. Open the files under "Complete structured data" on the Downloads page when you need the machine tables or the full JSON bundle (smrmap-dataset.json). In smrmap-dataset.json, projects may carry purpose (power or research_test). It is required at operating, commissioning and construction and is not part of the fixed projects.csv header. The map and search JSON files are application files for the website. They are not the report. All data and matching projects ------------------------------ reporting-projects.csv and the machine files are the full published set. The project list and Who is involved can download the same reporting columns for projects that match the filters. That matching file includes every matching project on every page, in the current sort order. It does not stop at the 25 rows on screen, and projects ticked for comparison do not change it. The full reporting file is sorted by project_id. A matching download follows the sort you chose on the page. Reporting columns ----------------- project_id, project_name, project_url, country_code, country, stage, stage_label, design_id, design_name, design_url, site_id, site_name, capacity_summary, parties, parties_limit, party_question, party_name, party_id, party_role, party_note, party_page, party_citation_url, party_citation_status, listed_source_names, listed_source_urls, dataset_fingerprint, build_date, capacity_review, summary, record_reviewed, evidence_dates, question_reviews, evidence_dates_limit project_id, design_id, site_id and party_id are stable identifiers for joins. The name columns beside them are for reading. project_url, design_url and party_page open SMRMap pages. They are not source citations. country_code and stage are the stable codes. country and stage_label are the readable words. capacity_summary uses the same wording as the project list. It keeps electrical and thermal output, gross or net basis, module or plant scope, approximate and up-to wording, peak output recorded separately, derived figures and conflicts. "Not recorded" is not zero. capacity_summary is one cell. A semicolon inside it marks a peak output recorded separately from the nameplate figure. That semicolon is not a break between party columns. capacity_review keeps the full capacity explanation alongside that short summary, including source disagreements, ranges and ceilings that cannot be reduced to a single rating. summary is the public project description, including historical-proposal warnings and permission limits. record_reviewed is the record's last_checked date; it is separate from build_date. evidence_dates gives readable latest dated listed publication(s), recorded actual event(s) and missing-publication-date counts. Multiple dates at overlapping precision are retained rather than arbitrarily choosing one. Dates do not establish current activity; undated sources may be newer. question_reviews preserves the five commercial questions' review states, dispositions, dates, full notes, reconciliation flags and question-review source URLs. These references do not automatically prove a named party's relationship. An unanswered, conflicting, unavailable or unresearched question retains its own meaning. evidence_dates_limit reminds readers that a build date is not a research date, and announcements, physical events, targets and permissions are different things. The six qualification columns are appended after the original reporting columns. parties is a readable sentence for each recorded organization, policy or programme, separated by a semicolon. parties_limit is the same sentence on every row: a missing role is not a finding that none exists, and a blank citation is not a negative finding or a zero. party_question, party_name, party_id, party_role, party_note, party_page, party_citation_url and party_citation_status are one list split across columns. Split each of those cells on a semicolon and read them in the same position. party_role keeps the recorded qualifier, including intended licensed owner, planned construction contractor, prospective customer, operator and vendor wording. Those are not collapsed into one role. party_note is the source-stated note stored on that relationship, when one exists. A commercial review note stays in projects.csv. Design fuel form and enrichment are design statements, not parties. They stay in project-relationships.csv and on the design record. They are not copied into the party columns. listed_source_names and listed_source_urls are the project's listed sources, in the same order. A URL in that list is not a citation of each party. Joins and several values ------------------------ Machine files still join by identifier. projects.csv stores vendor_ids, operator_ids, participant_roles and sources as JSON arrays. programmes.csv separates several values with a semicolon, and a semicolon inside a value is written as a comma. The reporting party columns use that same semicolon rule. A semicolon inside a name, note or URL is written as a comma so the positions stay aligned. The exact characters remain in smrmap-dataset.json. A direct citation URL that itself contains a semicolon is also exact in project-relationships.csv citation_url, which holds one URL, not a joined list. A programme or policy role is a type and a status joined by a semicolon in the project-relationships.csv qualifier column. In the reporting party_role cell that semicolon is written as a comma. The qualifier column keeps the semicolon. Empty positions are kept. "a;;b" has three positions. Do not drop empty positions when you split. The number of positions is the number of names in party_name. A completely blank party column means each of those positions is blank. One listed party has no semicolon, so a blank citation beside that one name is that party's blank citation, not a missing party. Two or more blank values keep the semicolons between them (";" is two blank positions). An empty party_name means no organization, policy or programme is listed on that row. Encoding -------- reporting-projects.csv and the matching reporting download are UTF-8 with a byte-order mark, so a spreadsheet can read accents. This field guide is UTF-8 without a byte-order mark. projects.csv is UTF-8 without a byte-order mark. The full JSON bundle is UTF-8. Blank meanings -------------- A blank is not zero, not "no", and not a finding that something does not exist. A blank party list means no organization, policy or programme entry is recorded in that reporting row. In project-relationships.csv, recorded=no means that question has no recorded answer. The note says that this is not a finding that none exists. citation_status on that row is not_applicable. "Not recorded" in a capacity cell means the output is not recorded. It is not 0 MWe. An unknown, a conflict, a source that could not be opened, and a question that is not applicable stay different from each other in the machine review columns. Citations --------- party_citation_status and the appended citation_status column use these words only: - direct: this relationship stores a source id, and citation_url is that source's public URL. - inherited_operator_link: the operator is recorded on the project, and that link does not store its own source id. citation_url is blank. This does not mean the operator is unsupported. - inherited_vendor_link: the same, for the vendor link. - inherited_programme_link: a programme record names the project. The programme link does not store a separate source id here. - inherited_design_statement: a design fuel statement in project-relationships.csv. It is not given a project source URL. - review_sources_only: the relationship has no source of its own. The question review's URLs are in review_source_urls, not source_url or citation_url, and are not a citation of the named party. - unresolved: no direct public URL is stored, and the row is not one of the inherited kinds above. This is a citation gap to review. It is not a finding that the fact is unsupported. - not_applicable: the row has no recorded answer, so there is no relationship to cite. Do not paste a URL from listed_source_urls, or from the project's sources array, onto a party as if that document proved that relationship. project-relationships.csv keeps its existing columns and appends review_source_urls. source_url now contains only the relationship's own stored source URL; it is blank when none is stored. Before 14 November 2026 that column could fall back to question-review URLs. Those URLs are now preserved separately in review_source_urls and support the question's note, not an inherited design statement or a named party. Read relationship_source_id, citation_url and citation_status to distinguish direct citations from inherited links. No source is inferred for a design fuel specification from a project fuel contract. Dates ----- dataset_fingerprint identifies the ten canonical data collections named in build-manifest.json. It is not a checksum of a CSV or of a page. build_date is the generation date of this file. It is not a research date. A record's last_checked is the record review date. A source's accessed date is when that source was opened. A source_date, when present, is the date the source states, with its kind, such as published or page updated. The optional status_checked in the full JSON bundle dates only an evidenced status/stage check; it never advances last_checked, CSV record_reviewed or the whole-record citation date. Status checks do not establish fuel, waste, funding or other unreviewed facts. Record pages may also show “Source re-read, text unchanged” or “Could not be confirmed on ”. These are dated observations about the status value recorded at the time, not confirmations of status or a review of the full record. A could-not-confirm date is the date of that published observation; unpublished private attempts do not establish public continuity. These observations disappear from the current record page when its status value changes. A later status confirmation supersedes an earlier observation. Neither observation changes status_checked or the whole-record review date. An event date is a recorded project event. An announcement date dates the announcement, not the target it describes. A target, planned date or estimated date is not an achieved date. Permission and application dates stay in the regulatory-action records. In regulatory-actions, status completed ends a design-review phase or step and is not an approval, on_hold is a review its regulator lists as paused, and status_note is a short reader phrase. The dataset metadata field generated dates the coverage statement in LICENSE.txt. It is not the build date. Date precision is kept in the structured records. A year, a month and a day are not rewritten as the same kind of date. publication-history.json lists release days (UTC): the day a public build first carried, changed, removed or restored a record (by stable_id) or an Updates entry (by its id on updates.html). A release day is not an event date. History starts with the first public dataset on 1 October 2026; policy and regulatory records count from their own files (3 October), before the full bundle carried them. Changes to review bookkeeping alone are not counted: the fields last_checked, listing_verified_date, next_review, ownership_verified_date, review_state, status_checked, and next_review, next_trigger, owner, reviewed_at inside evidence_review and detail_reviews. A change only to the evidence lists inside evidence_review and detail_reviews (the sources and locators cited) is listed separately, under evidence_only on its release day, and is not counted as a changed record; evidence_state holds the fingerprints they are compared by. Release days recorded before the file's schema 2 counted such changes as changed records. The RSS pubDate of an entry is noon UTC on its first release day; only the day is recorded. These identifiers do not provide an archive. The public URLs show the current release. Save a copy of the file you cite if you need that version later. A project id is not a way to retrieve an older release. Release identity ---------------- build-manifest.json has: - dataset_fingerprint and dataset_fingerprint_scope, for the canonical collections - built, the generation date - release_files_sha256 and release_file_count, a checksum of the published tree except the manifest itself - download_file_sha256, the sha256 of each file in the Downloads list - projection_file_sha256, the sha256 of data/project-list.json, which the matching button reads A file checksum can change when wording in that file changes even though the dataset fingerprint stays the same. The fingerprint changes only when the canonical collections change. Spreadsheet precaution and its limits -------------------------------------- In reporting-projects.csv, the matching reporting download and projects.csv, a text cell whose first visible character could start a formula, or that begins with a control character, is written with one leading apostrophe. Spaces, control characters and invisible formatting characters are skipped when looking for that first character. The characters = + - @ and their full-width forms are treated as formula starts. In the reporting join, that precaution is also applied to each value before the semicolon, so a later value is marked too. This is a precaution, not a guarantee for every spreadsheet program. Excel's own behavior was not used to certify it. The original columns of project-relationships.csv are written as exact structured values and do not receive that apostrophe. The appended citation columns and review_source_urls do. The exact structured text, without the apostrophe and without semicolon substitution, is in smrmap-dataset.json. Machine compatibility --------------------- projects.csv keeps this fixed header, in this order: stable_id, name, slug, country, stage, design_id, site_id, design_class, electrical_mwe, electrical_basis, capacity_scope, thermal_mwth, thermal_basis, module_count_planned, vendor_ids, operator_ids, participant_roles, sources, confidence, last_checked, inclusion_basis, derived_electrical_mwe, electrical_mwe_basis, summary, dataset_fingerprint, build_date, latest_dated_listed_source_publications, latest_dated_actual_project_events, listed_sources_without_publication_date, listed_source_count, commercial_customer_review_state, commercial_customer_disposition, commercial_customer_note, commercial_customer_reviewed_at, commercial_customer_sources, commercial_customer_reconciliation_needed, commercial_contractor_review_state, commercial_contractor_disposition, commercial_contractor_note, commercial_contractor_reviewed_at, commercial_contractor_sources, commercial_contractor_reconciliation_needed, commercial_funding_review_state, commercial_funding_disposition, commercial_funding_note, commercial_funding_reviewed_at, commercial_funding_sources, commercial_funding_reconciliation_needed, commercial_fuel_review_state, commercial_fuel_disposition, commercial_fuel_note, commercial_fuel_reviewed_at, commercial_fuel_sources, commercial_fuel_reconciliation_needed, commercial_schedule_revision_review_state, commercial_schedule_revision_disposition, commercial_schedule_revision_note, commercial_schedule_revision_reviewed_at, commercial_schedule_revision_sources, commercial_schedule_revision_reconciliation_needed. Nothing in that header was renamed, removed, reordered or extended for the reporting view. reporting-projects.csv is a separate file. project-list.json keeps format smrmap-project-list-1 and its columns, header and rows. A reporting object was added beside them. Readers that already use header and rows still see the previous projection. The matching button uses the reporting object when it is present. Worked example -------------- Project: HTR-PM Shidaowan demonstration (prj-htr-pm-shidaowan) Page: https://smrmap.com/projects/htr-pm-shidaowan/ Design: HTR-PM (dsn-htr-pm) Site: Shidaowan / Shidao Bay (site-shidaowan) Capacity: 210 MWe (gross) · plant · 500 MWth (nameplate; basis unspecified) Parties, in the same order as the joined columns: - Project proposer or applicant: China Huaneng Group — Parent company — direct — https://pris-stats.iaea.org/reactor/reactors-by-code/CN - Operator or intended licensed owner: Huaneng Shandong Shidao Bay Nuclear Power Company, Ltd. — Operator — inherited_operator_link — no direct source URL on this relationship - Operator or intended licensed owner: Huaneng Shandong Shidao Bay Nuclear Power Company, Ltd. — Intended licensed owner — direct — https://www.mee.gov.cn/xxgk2018/xxgk/xxgk09/202108/W020210823629115829690.pdf - Contractor: China Nuclear Engineering Corporation (CNEC) — Construction contractor — direct — https://world-nuclear-news.org/articles/first-vessel-installed-in-china-s-htr-pm-unit - Contractor: Chinergy / CNNC Energy Technology — Construction contractor — direct — https://world-nuclear-news.org/articles/first-vessel-installed-in-china-s-htr-pm-unit - Fuel: CNNC North Nuclear Fuel Element Co., Ltd. — Fuel supplier named for this project — direct — https://www.tsinghua.edu.cn/info/1181/51730.htm The listed_source_urls cell lists sources of the project record. A URL there is not a citation of every party above. Use party_citation_url only where party_citation_status is direct. An inherited operator link can name the operator without a source URL on that link; another role for the same organization can carry its own direct citation. Those are different relationships. Data fingerprint 434a3ad4fbbaaa6e4b79ee0db9d79c7b9b85f663d24417697f7e09f26d609d55; build date 2026-11-17. The fingerprint identifies the canonical data collections. The build date is when this file was generated. Neither is a source-review date, and neither retrieves an older copy of the file.