Global Hydropower Database#
Latest reviewed coverage (11 November 2026): 7,454 records / 128 countries, including 18 records in the Democratic Republic of the Congo.
The initial research batch contains 7,156 historical WRI plant records in 128 countries, imported from the complete primary-fuel Hydro partition. This is a discovery baseline, not 7,156 independently reverified operating plants. The WRI repository stopped maintenance in early 2022. The initial rows retain historical capacity; 110 Norwegian identities now have NVE-backed current status and technical metadata. Aggregation scope remains unresolved for all records.
Data attribution: World Resources Institute, Global Power Plant Database v1.3.0
repository snapshot, CC BY 4.0. Source commit, checksum, licence and retrieval
time are in sources.json; per-field provenance and original upstream links
are retained on every row. Upstream plant links have not been independently
reverified. Name matching never proves identity or pumped-storage technology.
Reproduce#
uv run python scripts/stage_hydro_gppd.py /path/to/global_power_plant_database.csv /tmp/hydro-stage --retrieved-at YOUR_UTC_TIMESTAMP
uv run python -m energy_data_retrieval.hydro_catalogue validate
uv run python -m energy_data_retrieval.hydro_catalogue build --output artifacts/hydro-v1
uv run python examples/database/plot_hydro_catalogue.py
Reviewable JSONL is canonical. The shared snapshot writer builds deterministic, typed Zstandard Parquet; generated binaries stay outside Git and publish with Pages. The shared browser viewer offers a clustered world map, category/numeric filters, all schema fields, evidence details and CSV/Parquet downloads.
Scientific boundaries#
Historical capacity is separate from installed, operational and planned MW.
Pumping electrical input is separate from generating output. Usable storage MWh
is never inferred from reservoir volume. Missing data is not zero. Reported and
modeled annual generation observations remain separate, with periods and units.
When a source omits capacity year, 2021 is only a dataset-vintage marker; the
actual observation year remains unknown in extensions_json.
Only reviewed independent/child scopes without unresolved overlap enter capacity summaries. Summaries separate lifecycle and technology, and historical totals also separate vintage. Parent complexes are excluded even after filtering. All launch records are excluded from current capacity totals, which display Unknown. Country charts show record counts, not national market shares or capacity ranks.
Continuation#
Use the persistent research/backlog.json, coverage manifest and decision ledger.
Reconcile registry IDs and dated operator/regulator evidence before enriching
status, technology, hydraulic metadata or scope. Preserve creation dates and
use the existing conflict-quarantining upsert API. Reuse the open hydro PR; refresh
remote refs before writing and never overwrite concurrent PV/BESS/Wind work.
See the published snapshot schema for every field and unit.
Norway official registry staging (5 November 2026)#
The NVE all-assets API was processed completely: 2,010 source rows yielded
1,977 generating candidates and 33 pump-only exclusions, with no parsing
quarantines. 110 candidates now enrich existing WRI identities; none add new plant IDs.
The existing 291 Norwegian WRI identities require an explicit registry crosswalk;
name equality is not proof of identity. Candidate facts and decisions persist in
catalogues/hydro/research/nve-20261105/, alongside the original-response checksum.
Attribution: Norwegian Water Resources and Energy Directorate (NVE), NLOD;
NVE source and field documentation.
uv run python scripts/stage_hydro_nve.py nve-response.json /tmp/nve-stage --retrieved-at 2026-11-05T08:45:58Z
The importer stages generating MW, owner/operator, basin, lifecycle, gross head (at NVE’s 67% filling reference), maximum turbine flow and normal-period 1991–2020 annual generation. The latter is modeled normal production, not a measured annual observation. Price area is not a physical grid connection. Commissioning strings remain raw metadata until date precision is reviewed. Pumped capability does not establish pure/mixed subtype or pumping input MW. Coordinates, reservoir energy and precise dates are not inferred.
research/market-ranking-202603.json records GEM’s March 2026 top-ten operating
hydropower capacity ranking and a separate pumped-storage ranking. The named
metric is operating generating nameplate MW for facilities of at least 30 MW;
it is not an all-size national inventory. Pumped capacity is a subset, never an
addition to the hydropower ranking. Source sheet URLs and row locators accompany
the factual summary. GEM bulk-file access and reuse review remain outstanding.
Reviewed Norwegian identity enrichment#
On 5 November 2026, 110 existing WRI records were linked to NVE registry IDs
through an explicitly reviewed crosswalk: unique exact names, NVE API/map ID
agreement, points within 250 metres and matching current API/map capacity and
operating state. No fuzzy names or spatial proximity alone trigger a merge.
The crosswalk and selected official map evidence are retained under research/.
Map data attribution: NVE, NLOD; https://www.nve.no/kart/kartdata/.
Enrichment preserves UUIDs, creation dates, historical capacities and generation observations. Current generating MW, status, owners/operators, basin, head, maximum turbine flow and normal-period generation retain separate provenance. Differences between historical and current ratings are not attributed to uprating. All 110 remain excluded from capacity sums pending scope review. The remaining 1,867 NVE candidates are not additional accepted plants.
Replay uv run python scripts/enrich_hydro_nve.py from the repository root.
Reapplying the same evidence leaves update timestamps and JSONL bytes unchanged.
US EIA annual generator inventory#
The final 2025 EIA-860 plant-code partition 1–500 enriches 173 existing plant identities with 369 distinct generating-unit observations. The reported nameplate sum is 13,153 MW at year-end 2025, not live 2026 capacity. All accepted plants have only HY/OP generator rows in the Operable sheet. The other 43 hydro-containing plants are held: 37 mixed technology/state, four missing existing identities and two name discrepancies. No fuzzy merges or inferred new plants are accepted.
The WRI upstream importer explicitly encodes EIA plant codes in USA-prefixed
identities; exact names and coordinates independently confirm this partition.
Existing UUIDs, creation dates, historical capacity and generation are preserved.
Generator IDs, statuses, nameplate and seasonal MW, operating month/year and
reported unit metadata remain in extensions_json.eia_hydro_annual_inventory
with workbook/sheet/row provenance. Seasonal capacity is not nameplate power.
Generator count is not turbine count. The HY code does not establish reservoir,
run-of-river or diversion subtype. Pumping MW and energy storage remain unknown.
All aggregation roles remain unresolved, including split reporting facilities;
these annual observations therefore do not enter global capacity sums.
Source: U.S. Energy Information Administration, final 2025 EIA-860 release,
retrieved 5 November 2026. Public-domain reuse policy and original ZIP/workbook
checksums are in the source registry. The importer is
energy_data_retrieval.hydro_catalogue.eia.parse_partition; replay the committed
bounded evidence using uv run python scripts/enrich_hydro_eia.py.
The next reviewed partition, plant codes 501–1000, adds 78 enrichments and
179 generator observations, with 3,462.0 MW of year-end 2025 nameplate capacity.
Its 31 holds comprise 14 coordinate discrepancies, 11 mixed technology/state
cases, four name discrepancies and two missing historical identities. These
holds remain outside accepted records. Combined EIA coverage is 251 identities
and 548 generators; catalogue coverage remains 7,156 records / 128 countries.
Replay the second partition with:
uv run python scripts/enrich_hydro_eia.py --batch catalogues/hydro/research/eia-2025-0501-1000.json
Next partition: EIA plant codes 1001–1500, then rotate to Brazil ANEEL SIGA. The source backlog retains Norway, other regions and all unresolved identities.
EIA continuation and Brazil source rotation#
The 5 November 2026 continuation covers plant codes 1001–1500 and 1501–2000: 84 existing plants enriched, 370 generator observations, and 2,397.6 MW at year-end 2025. Thirteen plants remain held (nine mixed technology/state and four missing existing identities); none are automatically created. Combined EIA coverage is 335 identities and 918 generators, with 87 holds. UUIDs, creation dates and historical observations remain unchanged; all capacity scopes remain unresolved and excluded from global totals.
uv run python scripts/enrich_hydro_eia.py --batch catalogues/hydro/research/eia-2025-1001-1500.json
uv run python scripts/enrich_hydro_eia.py --batch catalogues/hydro/research/eia-2025-1501-2000.json
Next regional work is Brazil: SIGA metadata is accessible under ODbL but its CSV and official datastore API returned HTTP 503. The independent ONS capacity inventory reports generating-unit nominal MW for ONS-dispatched operation classes, not a census of every small plant. Its publisher identifies CC-BY reuse with attribution. Identity, unit/plant aggregation and date interpretation must be reviewed before acceptance. EIA resumes at codes 2001–2500 afterward.
Brazil ONS exact registry enrichment#
On 5 November 2026, the first 100 sorted ONS hydro CEG identifiers yielded 78 existing plant enrichments and 289 generating-unit observations. Sixteen name discrepancies and six missing historical identities remain held. WRI’s Brazil importer explicitly derives its numeric plant ID from the CEG code; acceptance additionally requires exact names allowing only case/whitespace normalization. No fuzzy matching or new plant creation is performed.
CEG IDs, reported owners/operators, unit equipment IDs, nominal MW and raw
test/operation/deactivation dates are retained with CSV row locators. The ONS
dictionary defines val_potenciaefetiva as nominal unit MW. These unit facts
do not establish complete plant installed capacity or current lifecycle:
plant status, coordinates, historical capacity and aggregation roles remain
unchanged. Historical owner evidence is retained before updating owner names.
Pumping power, storage energy and hydraulic subtype are not inferred.
Attribution: Operador Nacional do Sistema Elétrico (ONS), CC-BY per publisher;
normalization/grouping and the WRI identity crosswalk are our transformations.
Original CSV, dictionary and WRI importer checksums are in the source registry.
The CSV lacks the optional id_ons dictionary field; no value is invented.
uv run python scripts/enrich_hydro_ons.py
Next: remaining 80 sorted ONS hydro CEGs after the saved partition cursor, followed by the outstanding alias/identity cases. Total coverage stays at 7,156 historical plant records in 128 countries, now with 523 enriched identities.
Completed ONS partition review#
The remaining 80 sorted CEGs add 53 existing Brazilian plant enrichments and 289 unit observations. Twenty-four name discrepancies and three missing historical identities remain held. Across the complete 180-CEG ONS hydro inventory, 131 identities are enriched with 578 unit observations and 49 identities remain held (40 name discrepancies and nine missing identities). The catalogue now contains 576 enriched identities; plant/country counts and capacity aggregation rules are unchanged.
The refreshed CSV checksum differs, but all 819 hydro unit facts match the
previous download when row locators are excluded. The prior partition’s
observations and timestamps are preserved; the new source registry entry
pins the new bytes and row locators. The drift comparison is retained in
research/ons-20261105-source-drift.json.
uv run python scripts/enrich_hydro_ons.py --batch catalogues/hydro/research/ons-20261105-0101-0180.json
Next regional partition: the Indian CEA September 2026 station list. Its official archive metadata is reviewed; original-file acquisition, reuse assessment, identity reconciliation and phase/pumped-storage scope review remain outstanding. No Indian candidates have been accepted in this run.
EIA 2001–3000 continuation and India reuse review#
The next two EIA code partitions enrich 122 existing US plants with 330 generator observations, dated 31 December 2025. The 34 holds comprise 29 mixed technology/state cases, four missing historical identities and one name discrepancy. Full held plant/unit evidence is retained for review; it is not accepted catalogue data. EIA now covers 457 identities and 1,248 generator observations, with 121 cumulative holds. Worldwide coverage stays at 7,156 records in 128 countries, with 698 cumulatively enriched identities. All capacity aggregation scopes remain unresolved.
The offline preparation utility validates the archive against a registered
SHA-256 and writes accepted/held decisions without changing canonical records.
It refuses different existing review files and duplicate registry identities;
existing capacity conflicts are held. Download the official EIA archive
separately, register its checksum/retrieval timestamp, then prepare a bounded
review with explicit --year, --first, --last, --reviewed-at, --sha256,
--archive and --output arguments to scripts/prepare_hydro_eia.py.
After reviewing decisions, reproduce the committed enrichments with:
uv run python scripts/enrich_hydro_eia.py --batch catalogues/hydro/research/eia-2025-2001-2500.json
uv run python scripts/enrich_hydro_eia.py --batch catalogues/hydro/research/eia-2025-2501-3000.json
The Indian CEA September 2026 original PDF was acquired, but its published copyright policy requires prior permission. No CEA plant data is imported or redistributed. The review ledger pins the report and policy checksums. The data.gov.in generation catalogue returned no original resource and displayed a testing notice.
An independent alternative, IHA’s India Hydropower Database, was acquired under its published CC BY 4.0 terms. Its 182 historical rows include 110 matched stations, 40 cascade assets, 20 unmatched rows, seven non-hydro rows and five construction/stalled rows. These are unaccepted candidates, not 182 additional plants. Attribution: International Hydropower Association, 2019 release; published temporal coverage 1917–2015. Do not infer current status from those historical labels or turn reservoir volume into storage energy. The CSV requires Windows-1252 decoding and distinguishes station coordinates from dam/reservoir coordinates.
Next regional partition: the first 100 matched-station IHA rows sorted by
numeric source ID, with explicit identity, phase and coordinate review.
research/india-source-review-20261105.json records acquisition, reuse and
scope findings. EIA resumes at plant codes 3001–3500 afterward.
IHA India historical identity review#
The first 100 matched-station candidates (numeric source-ID order) yield 12 explicitly reviewed enrichments and 88 identity/phase/location holds. Exact names, historical capacity and station coordinates within 0.001 degree per axis support the accepted crosswalks; dam coordinates are never substituted. The complete 182-row original also includes 72 rows in other asset classes, including seven non-hydro rows. No new plants are created.
Added IHA IDs, river metadata and immutable historical raw observations with CSV row locators and source references. Owner, lifecycle, generating capacity, coordinates and aggregation roles remain unchanged. Reservoir volume is not converted to storage energy. IHA 2019 labels do not establish current operation. CC BY 4.0 attribution: International Hydropower Association, India Hydropower Database, 2019 release; crosswalk review and normalization are our changes.
uv run python scripts/enrich_hydro_iha.py
Coverage remains 7,156 records / 128 countries, with 710 enriched identities. Next: the ten remaining matched-station rows, then Africa rotation; 88 held identities retain complete evidence for later review.
India completion, EIA continuation and Africa discovery#
The last ten IHA matched-station candidates yield one accepted Koldam enrichment and nine holds. All 110 matched rows are now reviewed: 13 enriched and 97 held. Lower Periyar’s coordinates disagree substantially and are not overwritten. Historical river and raw evidence do not establish current status.
EIA codes 3001–3500 add 77 existing identities and 260 generator observations, with 14 holds (13 mixed technology/state and one name discrepancy). EIA totals are 534 identities / 1,508 generators / 135 holds. Overall coverage remains 7,156 records in 128 countries, with 788 enriched identities.
uv run python scripts/enrich_hydro_iha.py --batch catalogues/hydro/research/iha-india-2019-0101-0110.json
uv run python scripts/enrich_hydro_eia.py --batch catalogues/hydro/research/eia-2025-3001-3500.json
African discovery acquired the original RePP Africa workbook. Its hosting metadata says CC0 but its Info sheet says CC BY-SA 4.0; reuse remains under review and no African records are accepted or redistributed. The source review ledger records checksums, structural counts and access attempts. An independent African Hydropower Atlas v2.1 successor is the next target. Its spreadsheet, field units, original evidence and identity boundaries need review before import; modeled generation must stay separate from measurements.
The IHA parser now rejects shifted/truncated CSV rows. EIA preparation parses the exact bytes whose checksum was verified, avoiding a second file read.
The AHA v2.1 spreadsheet has now been acquired and checksummed under its CC BY 4.0 resource terms. Its Overview label says v1.0 despite its filename; a v2.2 successor is linked by the repository. Compare these releases before reviewing the first 100 spatial/technical rows. No African plant data is accepted in this batch.
African Hydropower Atlas v2.3 historical enrichment#
Compared v2.1, v2.2 and the latest linked v2.3 successor (file modified 2024-02-14). The full 670-row spatial/technical sheet yields 26 explicitly reviewed existing identities across 18 African countries; 644 candidates remain held for identity, scope or source review. No new plants are created. Names, country, historical capacity and coordinates within 0.01 degree per axis support the crosswalks; shared-plant flags exclude automatic acceptance. Moukoukoulou’s duplicate COD/COG identities and Imboulou’s country discrepancy remain held. Matching rules never establish aggregate/phase independence.
River and basin metadata and immutable historical observations are added. Capacity, coordinates, lifecycle and aggregation roles remain unchanged. Model outputs (including energy per unit volume and storage available) are excluded. Retained raw flow, filling time and availability values may be modeled and are not promoted to canonical measurements. Reservoir volume is never converted into energy. Model channel/reservoir IDs are not official plant registry IDs. Unknown canonical measurements remain null.
Attribution: Sebastian Sterl and Charles J. Chawanda, African Hydropower Atlas
v2.3, CC BY 4.0. Crosswalk review and normalization are our changes. The source
registry pins the resource URL, retrieval time, release date and checksum;
research/aha-v23-reviewed.json preserves row locators and original source
columns, while aha-version-review.json records version comparisons.
The workbook Overview label is stale; the resource and checksum identify the
release. Historical status labels do not establish present-day operation.
uv run python scripts/enrich_hydro_aha.py
Coverage remains 7,156 historical records / 128 countries, now with 814 enriched identities. The gallery reports the AHA count separately. All plant aggregation scopes remain unresolved, so these records do not contribute to a verified worldwide generating-MW total. Next: EIA final-2025 plant codes 3501–4000; targeted African alias/scope review and Oceania discovery follow.
EIA final-2025 codes 3501–4000#
Reacquired the official ZIP on 2026-11-05T20:40:13Z; its checksum matches the previously reviewed final-2025 release. Reviewed 109 hydro-bearing plant IDs: 92 existing identities enriched, 17 held, no new plants. Retained 280 generator observations with official IDs, operators, nameplate/seasonal MW, reported commissioning dates and year-end-2025 status. Holds: {‘mixed_technology_or_state’: 15, ‘coordinate_review’: 1, ‘missing_existing_id’: 1}. Pumped/mixed and non-OP units remain unaccepted by this conventional-hydro partition. Historical status is not promoted to current operation; capacity aggregation remains unresolved and excluded from worldwide totals.
uv run python scripts/enrich_hydro_eia.py --batch catalogues/hydro/research/eia-2025-3501-4000.json
The replay command also accepts --root for isolated catalogue copies.
Coverage: 7,156 records / 128 countries / 906 enriched identities; EIA totals
626 plants / 1788 units / 152 holds. Created timestamps and prior
observations are preserved. Snapshot checksums and completeness are recorded
in research/eia-2025-3501-4000-validation.json.
Next regional partition is New Zealand’s official dispatched-generation fleet. The old EMI/new EA portals show differing CSV dates, and the listing includes formerly dispatched plants. Verify the original and reuse terms before accepting any records. EIA resumes at codes 4001–4500 afterward.
New Zealand dispatch review and EIA 4001–4500#
Acquired Electricity Authority’s 2 July 2026 dispatched-generation CSV under CC BY 4.0 (attribution: Electricity Authority, New Zealand; extraction is our change). The listing prose has a stale filename. All 157 rows were read; 68 hydro observations for 42 names remain held, with no NZ canonical additions. Historical effective intervals and multiple connection points must not become duplicate plants or summed generating MW. Benmore’s 540 MW historical and split 270 MW entries illustrate this risk. Sentinel dates 1899/9999 are retained raw, never interpreted as commissioning/retirement. The parser rejects duplicate intervals, malformed rows and invalid ratings.
uv run python scripts/stage_hydro_nz.py /path/to/20260702_DispatchedGenerationPlant.csv --output /tmp/nz-candidates.json
uv run python scripts/enrich_hydro_eia.py --batch catalogues/hydro/research/eia-2025-4001-4500.json
EIA codes 4001–4500 add 44 existing US identities and 129 generator observations; two candidates remain held. Fresh ZIP checksum matches the reviewed final-2025 release. Dates, operators, generating/seasonal ratings and registry IDs remain dated observations, not live operation. Coverage is 7,156 records / 128 countries / 950 enriched identities. EIA totals: 670 plants / 1917 unit observations / 154 holds. Aggregation scopes remain unresolved and excluded from totals. Next: explicit NZ identity/interval review; EIA resumes at 4501–5000. Source, candidate and validation ledgers retain resumable decisions and deterministic snapshot hashes.
Reviewed NZ identities and EIA continuation#
Explicit owner/WRI/EA identity reviews accept Aratiatia and Aviemore, retaining CSV rows 14 and 15 and their dispatch connection identifiers. Mercury’s station page corroborates Aratiatia’s 78 MW; Meridian’s station page corroborates Aviemore’s 220 MW. Each has a single EA row and matching historical WRI rating. Owner pages are used only for limited factual corroboration, with no page text or imagery redistributed; EA observations retain CC BY 4.0 attribution. Scalar lifecycle, capacity, coordinates and aggregation roles remain unchanged. Dispatch UnitCode is not treated as a physical generator ID. The 1899/9999 sentinels remain raw. Sixty-six observations for 40 other names remain held; Clyde’s differing rating bases and Benmore’s split connections need review.
uv run python scripts/enrich_hydro_nz.py
uv run python scripts/enrich_hydro_eia.py --batch catalogues/hydro/research/eia-2025-5501-10000.json
EIA ranges 4501–5000 and 5001–5500 have no hydro-bearing operable plants. The next populated range, 5501–10000, yields 158 existing identities enriched, 388 generator observations and 44 held plants. Empty ranges are persisted as completed cursors. All observations remain dated to year-end 2025, not live operation. Coverage is 7,156 records / 128 countries / 1,110 enriched identities. EIA totals: 828 identities / 2305 generator observations / 198 holds. No generating-MW total is promoted while aggregation scopes remain unresolved. Next: remaining NZ identity/interval review; EIA resumes at 10001–15000.
EIA 10001–15000 and Tuai owner review#
EIA final-2025 codes 10001–15000 enrich 99 existing US identities with 212 generator observations. Fifteen candidates remain held: {‘missing_existing_id’: 4, ‘mixed_technology_or_state’: 10, ‘coordinate_review’: 1}. Namespaced IDs, operators, generating/seasonal MW and unit dates retain original workbook row locators. Reused source bytes were checksum-verified and reparsed. No new plants, inferred pumping ratings or current-operation claims are added. EIA cumulative coverage is 927 identities / 2517 generators / 213 holds.
Tuai is explicitly linked to EA CSV row 140 using Genesis’s FY2024 integrated report, printed page 41. It reports 60 MW, consistent with the historical WRI record and EA dispatch observation. The potential 6 MW efficiency benefit is qualified by footnote 1: available only below the station’s 60 MW maximum. It must not become an inferred 66 MW rating. Only the dispatch identity and immutable observation are accepted. The downloaded report checksum, title, version, locator and limited-factual-reuse assessment are retained; report text and images are not redistributed. NZ now has three enriched identities; 65 observations covering 39 other names remain held. Benmore’s six 93 MW units, 552 MW maximum output and historical 540/270 MW dispatch rows have different bases and remain unresolved; no sum or scalar capacity promotion is made.
uv run python scripts/enrich_hydro_eia.py --batch catalogues/hydro/research/eia-2025-10001-15000.json
uv run python scripts/enrich_hydro_nz.py --batch catalogues/hydro/research/nz-tuai-20261105.json
Both replay commands now reject duplicate/missing reviewed identities and validate every transformation before writing any shard. Conflicts preserve all files; successful replay remains idempotent. This is not an I/O transaction. Coverage: 7,156 historical records / 128 countries / 1,210 enriched identities. All unresolved scopes remain excluded from generating-MW totals. Next regional batch: first 50 remaining Norwegian WRI identities sorted by WRI ID, with explicit NVE/owner corroboration; EIA resumes at 15001–20000.
Norwegian review and EIA 50001–55000 (6 November 2026)#
The first 50 unresolved Norwegian WRI identities (WRI1002991–WRI1003149, sorted by WRI ID) remain held. Fresh NVE registry and map responses retain checksums, exact-name candidate rows and geometry. Names are not identity proof: duplicated names, spatial offsets and different station/development ratings need explicit corroboration. Statkraft’s Norwegian Adamselv and Aura pages confirm ratings of 50 and 290 MW, respectively; WRI reports historical 60 and 352 MW. These comparisons do not resolve physical scope or justify overwriting coordinates. No Norwegian canonical facts were accepted in this partition. NVE data retain NLOD attribution; owner checks use limited facts only.
The new scripts/screen_hydro_nve.py stages bounded review partitions and
rejects truncated map responses and replacement of differing review files.
It never accepts or fuzzy-merges candidates. Resume with --offset 50 --limit 50
and original --api, --gis, --reviewed-at and --output arguments.
EIA codes 15001–20000 and 20001–50000 contain no hydro-bearing operable plants; these empty partitions are complete. Codes 50001–55000 enrich 171 existing US plants with 396 generator observations; 41 candidates remain held. Evidence includes official IDs, operators, generating and seasonal MW, unit dates and year-end 2025 status. This is not live operation. Original archive checksum and parsed workbook rows were verified. Replay with:
uv run python scripts/enrich_hydro_eia.py --batch catalogues/hydro/research/eia-2025-50001-55000.json
Coverage is 7,156 historical records / 128 countries / 1,381 enriched identities. EIA totals: 1,098 identities / 2913 units / 254 holds. Aggregation scopes remain unresolved and excluded from capacity totals. Next: Norwegian rows 51–100 for explicit review; EIA resumes at 55001–60000.
Stable Norwegian review cursor and EIA 55001–60000#
Norwegian identities 51–100, WRI1003150–WRI1003313, were screened against the checksum-verified 6 November NVE API/map captures. All 50 remain held; the first 100 screened identities are not accepted enrichments. Statkraft’s Norwegian pages report Grytten at 144 MW and Kvilldal at 1,240 MW. Historical WRI ratings and Grytten’s two potentially related names require explicit alias/scope review. Owner facts remain review evidence only, with URLs, titles, retrieval dates, locators and limited-factual-reuse assessment; no copied text or imagery.
The screener now supports an exclusive stable --after-wri-id cursor. An
accepted earlier identity can disappear from the unresolved list without
shifting the continuation boundary. Offsets remain supported but cannot be
combined with a stable cursor. Regression tests verify this case.
uv run python scripts/screen_hydro_nve.py --api /path/to/nve.json --gis /path/to/map.json --after-wri-id WRI1003313 --limit 50 --reviewed-at YOUR_UTC_TIMESTAMP --output /tmp/next-nve-review.json
uv run python scripts/enrich_hydro_eia.py --batch catalogues/hydro/research/eia-2025-55001-60000.json
EIA codes 55001–60000 enrich 61 existing US identities with 119 generator observations; 11 plants remain held. Added IDs, operators, generating/seasonal MW and unit dates retain year-end 2025 status, not live operation. The original archive checksum and source rows were verified; no new archive retrieval is claimed. Coverage: 7,156 historical records / 128 countries / 1,442 enriched identities. EIA totals: 1,159 identities / 3032 generators / 265 holds. No unresolved capacity enters global totals. Next: Norwegian identities after WRI1003313 (101–150), followed by EIA 60001–65000.
EIA tail and Norwegian pumping-power review — 6 November 2026#
EIA codes 60001–65000 enrich 16 existing US identities with 27 generator observations. Seven plants remain held; codes 65001–99999 contain ten further held plants and no accepted matches. The code scan now reaches 99999, which is not a claim of complete US plant coverage. Unmatched official IDs and mixed or pumped facilities still need explicit identity and scope review. Registry IDs, operators, generating/seasonal MW and unit dates retain year-end 2025 semantics. Coverage is 7,156 historical records / 128 countries / 1,458 enriched identities; EIA contributes 1,175 identities, 3059 units and 282 held plants.
Norwegian candidates 101–150 (after WRI1003313 through WRI1003515) remain held. The first 150 screened candidates are separate from the 110 previously accepted NVE enrichments. Statkraft reports Saurdal at 640 MW generating capacity with four units, two reversible. Its 14 March 2025 filing reports 320 MW existing pumping capacity and proposes 640 MW. The proposal does not establish present operation. Mauranger’s owner rating is 250 MW, compared with 307 MW in the historical record. Dated owner facts, source locators and reuse assessments are retained in the review ledger; canonical values are unchanged.
The NVE screener optionally accepts --api-source-id and --gis-source-id
together. It verifies both captures against registered SHA-256 hashes before
screening and records the source IDs. Existing invocations remain supported;
registry verification is recommended for published research. Tests cover
changed API and GIS captures and ensure rejection leaves no output ledger.
uv run python scripts/enrich_hydro_eia.py --batch catalogues/hydro/research/eia-2025-60001-65000.json
uv run python scripts/screen_hydro_nve.py --api /path/to/nve.json --gis /path/to/map.json --api-source-id nve-hydro-20261106 --gis-source-id nve-gis-20261106 --after-wri-id WRI1003515 --limit 50 --reviewed-at YOUR_UTC_TIMESTAMP --output /tmp/next-nve-review.json
Next: Norwegian candidates after WRI1003515 and remaining Brazilian ONS inventory; retain EIA holds for explicit identity/pumped-storage review.
ONS state and dispatch metadata — 6 November 2026#
All 180 ONS hydro CEGs had already been screened. A fresh CSV download confirms
no changed unit facts for the 131 accepted plants (row positions are not facts).
The retained evidence now supplies structured extensions_json.ons_dispatch_metadata
for state code/name, electrical subsystem code/name and dispatch modality.
Each value retains source ID and CSV row; multiple values remain explicit sets.
Unknown fields are empty sets. A subsystem is not an exact grid connection.
Original unit observations, lifecycle, capacities and aggregation roles are unchanged.
This enriches 131 existing records in Brazil; it adds no newly enriched identities:
coverage remains 7,156 historical records / 128 countries / 1,458 enriched identities.
The ONS source page describes nominal generating-unit MW and CC-BY reuse;
val_potenciaefetiva must be interpreted using that dictionary, not its name alone.
uv run python scripts/enrich_hydro_ons_dispatch.py --batch catalogues/hydro/research/ons-dispatch-20261106.json
The next 50 Norwegian candidates were screened with registered source hashes;
all remain held. Continuation is after WRI1003631 (201–250). The Brazilian backlog
is corrected: 49 retained holds require explicit identity/scope review; there is
no unprocessed partition in this 180-CEG capacity inventory.
Explicit Brazilian registry-alias review — 6 November 2026#
Thirty-five prior ONS holds are resolved through recorded, individually reviewed CEG/WRI registry crosswalks and single-name historical, abbreviated or state-qualified aliases. The importer still rejects name differences by default. An explicit review must match record UUID, original name, CEG and every source unit name; no fuzzy matching is used. The canonical name remains unchanged and the ONS name becomes an alias. Original hold ledgers remain immutable; the new resolution ledger supersedes only these 35 decisions. Five multi-powerhouse groups and nine missing identities remain held. Copel’s 18 November 2024 primary-source description also corroborates Foz do Areia, Segredo and Salto Caxias aliases; proposed expansions are not accepted as current operating capacity.
The batch adds 150 dated unit observations, CEG IDs, owners/operators, aliases and state/subsystem/dispatch metadata. Plant capacity, lifecycle and aggregation scope are not inferred. Coverage is 7,156 records / 128 countries / 1,493 enriched identities; ONS contributes 166 identities and 728 unit observations. Source reuse and row locators remain those of the original checksum-pinned ONS captures.
uv run python scripts/enrich_hydro_ons_aliases.py --batch catalogues/hydro/research/ons-aliases-20261106.json
The Norwegian continuation after WRI1003631 returns zero rows: the remaining-identity
scan is exhausted. The 200 screened holds remain unresolved, separate from 110
accepted NVE identities. Next: five ONS multi-powerhouse groups, then nine missing
identities; do not repeat the exhausted scan.
Reviewed powerhouse relationships — 6 November 2026#
Belo Monte/Pimental, Henry Borden and Itiquira now retain six stable nested powerhouse identities linked to three existing plant UUIDs and 42 ONS generating unit IDs. These are evidence components, not additional catalogue plants or additive capacities. Component names are not plant aliases. The explicit review must cover every source name and exact registry identity; duplicate equipment and changed immutable observations are rejected before any shard is written.
MME’s 5 April 2016 report (updated 12 April) establishes Belo Monte’s principal and complementary powerhouses; its commissioning projections are not current operation. EMAE identifies Henry Borden’s external/underground powerhouses and reports Pelton turbines, head/flow and a pumping constraint. Unqualified head is retained as reported technical context, not promoted to gross/net head. ONS’s historical Itiquira report identifies two sequential powerhouses and Francis units. Source dates, URLs, publishers, locators, access/reuse assessments and original ONS CSV checksums/rows are retained. Missing publication dates stay unknown.
Simplício/Anta remains held: the 305.7 MW historical row differs in scope from AXIA’s 333.7 MW complex including two 14 MW Anta units. Itaipu’s 20-unit binational group is not assigned to the historical 7,000 MW Brazilian-share record; frequency is not an ownership allocation. Nine missing identities also remain held.
Coverage: 7,156 historical records / 128 countries / 1,496 enriched identities. ONS contributes 169 identities / 770 unit observations; 11 groups remain held. All unresolved capacity scopes stay excluded from totals.
uv run python scripts/enrich_hydro_components.py --batch catalogues/hydro/research/ons-components-20261106.json
Next: nine missing ONS identities, starting São Manoel and Baixo Iguaçu, with explicit duplicate/registry review; retain the two scope conflicts above.
New Brazilian registry identities — 6 November 2026#
Eight reviewed CEG identities absent from the historical baseline are now accepted: Areado, Jesuíta, Colíder, Sinop, Juruena, São Manoel, Baixo Iguaçu and São Roque. The first two are PCH small-hydro facilities. Canonical UUIDs use the full ANEEL CEG, never an invented WRI ID. The importer checks every Brazilian name/alias and all external IDs before writing, rejects collisions and changed replays, and leaves existing records unchanged. Bandeirante remains held because its normalized name collides with a different historical registry identity. Simplício/Anta and Itaipu retain the earlier scope holds.
All eight have ONS owner/operator, unit power/date and dispatch observations, retaining original CSV checksums and row locators. Eight primary plant sources add supported facts: EDP São Manoel, ENERGO-PRO Baixo Iguaçu, EDF Sinop, Nova Participações São Roque, Atiaia Areado, AXIA Colíder, MME Juruena and the Bom Futuro PCH Cidezal PAE’s Jesuíta section. The original 2024 PAE was downloaded after web extraction failed; its SHA256 and page locator are retained. Only bounded facts are redistributed from owner pages; no bulk licence is assumed.
Five records carry reported installed generating MW, not operating MW. Areado’s rating is explicitly dated December 2024; undated portfolio ratings retain their retrieval basis. São Manoel supplies run-of-river type, four Kaplan turbines, May 2018 operation start and 577.22 hm³ total reservoir volume, converted to 577,220,000 m³; usable water volume and storage MWh remain unknown. Baixo Iguaçu and Jesuíta also have turbine descriptors. Rivers and Sinop’s 2019 start year retain source provenance. AXIA’s Colíder safety classification is attributed to the owner and is not a lifecycle determination. Jesuíta’s differing plant/unit ratings remain an explicit unresolved capacity review; Juruena’s old permit is used for geography only. No generating MW are summed from ONS unit ratings.
Coverage is 7,164 records / 128 countries, comprising 7,156 historical WRI records (1,496 enriched existing identities) and eight new CEG records. Brazil has 709 records; ONS evidence covers 177 identities and 791 units. All eight new records retain unknown current status, absent coordinates and unresolved aggregation scope. Existing map and KPI components consume the same typed snapshot and exclude these unresolved ratings from capacity totals.
uv run python scripts/accept_hydro_ons_new.py --batch catalogues/hydro/research/ons-new-20261106.json --output catalogues/hydro/records/projects/ons-new-20261106.jsonl
uv run python examples/database/plot_hydro_catalogue.py
Replay is idempotent. The next regional partition is the first 100 retained AHA Existing holds, beginning Excel row 4, using primary alias/phase/location evidence.
African evidence bases — 6 November 2026#
Twenty-six previously accepted AHA identities now expose structured per-column
provenance under extensions_json.aha_field_provenance: original value, unit,
Excel row, source ID, upstream citation, fallback reference and uncertainty.
These are provenance enrichments, not newly accepted plants or measurements.
Reservoir size is expressed in million m³, not MWh. Mean annual river discharge
is separate from design discharge; neither is a measured operating flow.
First Year is the earliest expected entry year, not verified commissioning.
Availability selects a model profile, not current plant availability or status.
Upstream references remain explicitly unverified; missing citations remain null.
The first 100 retained AHA Existing holds (Excel rows 4–242) have reproducible
name-candidate evidence in research/aha-provenance-20261106.json. None is accepted
by name matching. Congo country/duplicate and phase boundaries remain held.
E2C’s Imboulou page was suspended and its Moukoukoulou page returned 403 during
review. The Republic of Congo government’s 2 July 2023 concession article
corroborates national context but does not resolve duplicate catalogue records
or present operation. Original hold ledgers and source checksum remain intact.
uv run python scripts/enrich_hydro_aha_provenance.py --batch catalogues/hydro/research/aha-provenance-20261106.json
Global coverage remains 7,164 records / 128 countries / 1,496 enriched historical
identities plus eight new registry identities. Resolve the retained 100 holds;
the next unscreened Existing partition starts after Excel row 242.
KenGen’s 2025 owner statistics add 160 dated capacity and energy observations to eight existing Kenyan plants. Installed versus effective MW and generated versus sold GWh remain separate, with fiscal periods and cell-level provenance. Five additional named stations await registry identity review. The catalogue contains 7,164 records in 128 countries; these observations do not establish current operation or change additive capacity KPIs.
The five retained KenGen identity candidates are now accepted from owner-report and portfolio corroboration. Total coverage is 7,169 records in 128 countries. Their 60 dated entries include two missing Wanjii 2021 energy cells; missing is not zero. EPRA licence IDs remain unknown following a blocked download.
Owner technical evidence now covers twelve Kenyan stations, with replacement unit groups, generator/turbine descriptions, grid connections, watercourses, reservoir volumes and dated commissioning assertions. Conflicting dates and unspecified head remain qualified; neither MVA nor reservoir volume is converted to generating MW or stored energy. The copied Turkwel description is excluded.
The Uganda ERA batch adds 32 station/phase identities, including three hospital/community schemes, bringing the catalogue to 7,201 records across 128 countries (38 in Uganda). The 39-row hydro workbook partition is fully reviewed: six possible matches to existing records and Swam remain held. Accepted records contain 265 reported MW values and 599 missing annual cells across 2000–2026; 2026 means September. Current status, physical aggregation scope and map coordinates remain unknown. The shared viewer includes these records in country and plant counts, while unresolved capacity stays excluded from MW KPIs.
The primary sources are ERA’s Installed Generation Capacity as at September
2026 (uganda_era_capacity_2026) and Electricity Supply Industry State of the
Environment Report 2023, Annex 3 (uganda_era_environment_2023). See the formal
references. Run scripts/accept_hydro_era.py for idempotent
replay; add --workbook PATH to verify the downloaded workbook against every
retained source cell before acceptance. The offline gallery
examples/database/plot_hydro_catalogue.py rebuilds the typed Zstandard snapshot,
shows reported locations and country counts, and reports the Uganda partition.
ERA Annex 3 enrichment retains 128 historical entries for 32 Ugandan stations
in extensions_json.era_2023_assertions: commissioning precision, capacity MW,
grid connection and lifecycle group. The report’s partial commissioning caveat
is retained, and two construction/licensed records have unknown commissioning.
Historical assertions never promote current operating MW or lifecycle. Kuluva’s
intra-report commissioning discrepancy remains unresolved. The report is already
registered as uganda_era_environment_2023; replay uses
scripts/enrich_hydro_era_report.py. Twenty-seven other Annex rows remain held.
The ERA construction/licensed acceptance batch adds 21 historical project
identities, for 7,222 records across 128 countries (59 in Uganda). Its 21 project
MW observations and seven dated UECCC procurement kW observations remain distinct
from installed/operational MW. The joint lifecycle group is not split or promoted
to current status. Unknown locations are omitted from the map; counts include
the identities, while unresolved physical capacities remain excluded from KPIs.
The source ueccc_orio_tender_2020 is listed in the formal references. Replay with
scripts/accept_hydro_era_planned.py; optional --text verifies the full parsed
Annex partition. Six older identity matches, Swam, Rushobe and Kisonko remain held.
Nepal DoED construction licences now contribute 82 new reviewed identities and
82 licensed-MW observations, bringing coverage to 7,304 records in 128 countries.
Rows 1–100 were reviewed; 18 remain held. All 262 source rows were parsed for
identity collisions, but later rows are not accepted yet. Full licence IDs,
rivers, promoters, districts, source date strings and four coordinate bounds
are retained. No current operation, installed MW, Gregorian date conversion or
point coordinates are inferred. The shared viewer counts these records but
excludes unresolved capacities and unmapped locations. See
nepal_doed_construction_hydro_2026 in the formal references and replay with
scripts/accept_hydro_doed.py (optional --html source verification).
Nepal continuation: 7 November 2026#
The second DoED partition reviews S No 101–200, accepting 91 licensed project identities and retaining 9 identity/scope holds. Together with the previous partition, 173 DoED identities are accepted and 27 candidates remain held. Coverage is 7,395 records across 128 countries, including 185 Nepal records. The 7,304 prior canonical records and their timestamps are unchanged.
The new capture reproduces all 262 normalized source rows. Its HTML checksum changed, so both capture checksums and retrieval times remain registered. The page explicitly says Updated on — October 15, 2026: new licensed-MW observations use 2026-10-15, with individual measurement vintage unknown. This does not convert the raw licence issue/expiry calendar strings or establish current construction/operation. No licensed MW enters installed-capacity totals.
Reviews compare full licence IDs, names, promoters, rivers, districts and source bounds. Separately named/licensed Daraudi B/C, Hongu and Budhi Gandaki projects retain distinct identities; shared infrastructure and aggregate capacity remain unresolved. Rolwaling’s repeated name under licences 396/516 remains held, as do three explicit cascades and five historical-name/scope candidates. No fuzzy merge, parent link, plant centroid, pumping rating or storage energy is inferred.
uv run python scripts/accept_hydro_doed.py --batch catalogues/hydro/research/doed-20261107-0101-0200.json --html /path/to/doed-20261107.html
Omit --html for offline replay. With HTML, the capture checksum, full normalized
inventory and page date must match. The importer also checks that partition
decisions are exhaustive/disjoint, each decision reproduces the source row, and
accepted names are not duplicated elsewhere in the full inventory. Both previous
and new partitions replay without changing accepted records. Checksums and field
completeness are in research/doed-20261107-0101-0200-validation.json.
Attribution remains Nepal’s Department of Electricity Development,
nepal_doed_construction_hydro_2026. No open bulk-reuse licence is asserted;
source-specific rights are not replaced by the repository software licence.
Next: DoED S No 201–262, retaining all existing holds, then rotate back to
African Hydropower Atlas Existing candidates after Excel row 242.
Nepal final partition: 7 November 2026 evening#
DoED rows 201–262 add 54 licensed project identities, with 8 candidates held: five explicit cascades, the repeated Rolwaling name, historical Marsyangdi scope and reversed Myagdi coordinate endpoints. All 262 source rows are now reviewed: 227 accepted identities, 35 held candidates. Catalogue coverage is 7,449 records / 128 countries, including 239 Nepal records.
The fresh source capture reproduces the full previous normalized inventory. New records retain licence IDs, licensed MW dated to the page update of 15 October 2026, promoters, rivers, districts, raw dates and coordinate bounds. No current operation, exact plant point or additive independent capacity is inferred. Uttarganga’s name does not establish pumped storage or energy MWh. All 7,395 prior records and timestamps are preserved. Source-specific reuse assessment remains in sources.json; no open bulk-reuse licence is asserted.
Replay with scripts/accept_hydro_doed.py --batch catalogues/hydro/research/doed-20261107-0201-0262.json (one command line);
optional --html verifies the original capture checksum and full source table.
Validation hashes/completeness are in the corresponding -validation.json.
The shared viewer and runnable examples/database/plot_hydro_catalogue.py
consume this same canonical data; licensed MW remains excluded from current
generating totals. Next: African Hydropower Atlas Existing candidates after
Excel row 242, while preserving the 35 Nepal holds.
African reviewed aliases — 8 November 2026#
The next 100 previously held AHA Existing rows (243–464) were reviewed against
canonical plant names, countries and station coordinates. Thirty-two explicit
crosswalks add historical evidence in Burkina Faso, Côte d’Ivoire, Madagascar,
Morocco, Namibia, Rwanda, Sudan, Tanzania, Tunisia, Uganda and Zambia. The
catalogue stays at 7,449 plants in 128 countries, with 58 AHA-enriched
identities. These source observations do not establish current operation.
research/aha-crosswalk-20261108.json freezes each accepted canonical identity
and original source row; 68 candidates stay held. Fifteen differing historical
capacity values are explicitly quarantined from scalar rating promotion.
Shared Kariba shares, grouped Mukungwa units and ambiguous Afourer pumped versus
conventional facilities are not accepted by this crosswalk. No capacities are
added to totals and no city-centre coordinates or reservoir energy are inferred.
The existing aha_historical_inventory and aha_field_provenance extension
structures retain river/basin, station coordinates, historical capacity,
reservoir volume, mean annual/design discharge, inventory status and year/profile
inputs where present. Each field retains units, basis, Excel row, source registry
ID and upstream citation with an explicit unverified flag. First Year is not a
verified commissioning date, Availability is a model-profile selection, and
reservoir volume is neither usable storage nor MWh. Filling Time remains raw
model input only and is not promoted as a hydraulic measurement.
Replay with scripts/enrich_hydro_aha_reviewed.py --batch catalogues/hydro/research/aha-crosswalk-20261108.json --workbook <downloaded.xlsx>.
Checksum and row equality checks precede the shared staged replay. Concurrent
identity changes and shared scopes fail closed; identical replay preserves
created_at, updated_at and last_verified_at. The existing shared viewer
exposes these extensions, and examples/database/plot_hydro_catalogue.py maps
the catalogue and reports the expanded AHA evidence count. Parquet snapshots
remain deterministic typed Zstandard outputs outside Git, with hashes committed
in the validation ledger.
Attribution: Sterl and Chawanda (2024), African Hydropower Atlas v2.3, HydroShare,
CC BY 4.0; our changes are explicit crosswalks and provenance normalization.
The 8 November capture matches SHA-256
5a24121837e0b57c226afbcc63a9e3e821b1fe2a3715db2b1a4ee73dca599ae6.
Next regional work is the New Zealand retained owner-identity backlog; the next
AHA partition begins after Excel row 464.
New Zealand owner equipment and rating boundaries — 8 November 2026#
Fifteen Mercury and Meridian station pages were downloaded and checksummed. Twelve existing New Zealand identities receive 59 normalized owner assertions. Nine historical/owner rating comparisons are quarantined from scalar promotion. The catalogue remains 7,449 plants in 128 countries. No current generating total, status, location, or storage-energy estimate changes.
extensions_json.nz_owner_technical_assertions reuses the owner-assertion
structure: source registry ID and facts with key, value, unit, basis, section
locator and uncertainty. The registry holds URL, publisher, title, retrieval
instant, unspecified publication date, HTML checksum and access/reuse assessment.
Only limited factual extracts are retained; owner prose, images and original
HTML are not redistributed. Average annual GWh has no specified averaging
period or gross/net basis and is not a dated measured-generation observation.
Planned upgrade benefits are distinct from completed replacement equipment.
Benmore’s six 93 MW unit ratings remain distinct from its 552 MW station output. Manapōuri’s seven 128 MW units, 850 MW continuous rating and 800 MW consent limit are separate assertions. Maraetai I/II phase ratings are retained without summing canonical totals. Karāpiro’s 2023–2025 refurbishment and September 2025 completion are recorded as owner assertions. Historical conflicting ratings are preserved.
The three Ōhau pages remain candidate evidence: canonical Ohau A has a 688 MW
historical rating, while the owner reports individual A/B/C outputs of
264/212/212 MW. Numerical agreement of a sum does not prove aggregate identity;
no phase records or hierarchy correction is accepted without explicit review.
The EA July 2026 dispatch CSV was downloaded again and its checksum is unchanged;
its 39-name interval/identity backlog is not claimed as resolved by owner facts.
Replay uv run --no-sync python scripts/enrich_hydro_nz_owner.py and run
examples/database/plot_hydro_catalogue.py for statistics and mapping. The
existing shared viewer exposes the extension evidence. The generic owner replay
retains its KenGen defaults and rejects conflicting prior evidence; unrelated
extensions and creation timestamps survive. JSONL remains canonical; deterministic
typed Zstandard snapshots stay outside Git, with hashes and field completeness
in research/nz-owner-20261108-validation.json.
Next regional work: Asian national-inventory discovery. Retain the Ōhau scope review and EA interval reconciliation for the next New Zealand partition.
Historical Africa evidence and India reuse review — 9 November 2026#
This run reviews 145 retained AHA Existing candidates: earlier Excel rows 4–242 and the remaining rows 465–673, omitting already accepted identities. 26 existing plants in 16 countries receive historical field evidence; 119 candidates remain held, and 16 differing capacity assertions are quarantined from scalar promotion. These categories overlap: a matched plant may retain a disputed historical source rating without adopting it as capacity. The catalogue remains 7,449 records in 128 countries, now with 84 AHA identities. All 270 AHA Existing rows have been screened: 83 accepted and 187 held. The 84th accepted AHA identity is historically classified Committed. No new plants, current operating ratings or additive generating totals are claimed.
The batch retains reported river/basin, reservoir volume, mean and design flow,
station coordinates, capacity and inventory/model year/profile fields where
present. Units, uncertainty, Excel row and upstream references remain in
aha_field_provenance; upstream citations have not been independently verified.
No reservoir volume becomes storage MWh, and model years are not commissioning dates.
Name variants include Song-Loulou, Gilgel Gibe I/III, Ayame II and Nangbéto.
Cambambe, Boali, Inga, Awash, Poubara and shared international assets remain held
where phase, national-share or aggregate boundaries need primary evidence.
Replay scripts/enrich_hydro_aha_reviewed.py --batch catalogues/hydro/research/aha-crosswalk-20261109.json --workbook <downloaded.xlsx>.
The frozen partition is checked before writes: duplicate accepted/held source
rows, many-to-one assignments, missing decisions and changed workbook rows fail.
Legacy reviewed batches remain supported. Exact replay preserves timestamps;
unrelated records and scalar capacities remain unchanged. The shared viewer
already exposes this evidence; the runnable mapping/statistics gallery remains
examples/database/plot_hydro_catalogue.py. Snapshot bytes stay outside Git,
with typed Zstandard replay hashes in the validation ledger.
Attribution: Sterl and Chawanda (2024), African Hydropower Atlas v2.3, HydroShare, CC BY 4.0. The original workbook was fetched again on 9 November and matches the previous SHA-256; our changes are explicit crosswalks and provenance normalization.
CEA’s September 2026 inventory PDF was successfully acquired, but its current copyright policy requires prior permission by email. No such permission was established, so zero CEA plant rows are published. The access ledger retains URLs, retrieval date, policy locator and checksums rather than the station table. The next priority is owner evidence for the retained AHA Inga and Poubara phase boundaries; revisit CEA only with permission or an explicitly reusable release.
NVE registry enrichment — 9 November 2026#
110 previously reconciled Norwegian plants receive 880 dated registry assertions: main-owner organization IDs, actor roles and ownership shares, concession IDs/titles, county/municipality codes, waterfall first-use year, oldest generating-part year, and energy equivalent in kWh/m3. Exact registered NVE IDs and names are required; no fuzzy identity matching or new plants are introduced. No current capacity, status, coordinates, commissioning date or storage MWh is inferred or changed. Energy equivalent is not reservoir energy. Concession association is not proof of current permit validity. Dates and uncertainty stay attached to each assertion.
The complete NVE API (2,010 source entries) was downloaded and checksummed;
110 existing identities were selected, with zero identity-drift quarantines.
The catalogue remains 7,449 records across 128 countries. Attribution: NVE, NLOD.
Raw API bytes and generated Parquet remain outside Git. Source registry and the
research ledger retain checksum, retrieval timestamp and field-level locators.
Replay uv run python scripts/enrich_hydro_nve_registry.py; identical facts
preserve timestamps, and changed evidence requires conflict review. The existing
shared viewer exposes nve_registry_assertions through its evidence display.
Run examples/database/plot_hydro_catalogue.py for statistics and mapping.
Next: first 50 retained NVE identity holds, with primary owner/concession evidence.
Inga and Poubara phase boundaries remain held; discovery did not resolve them.
Additional NVE identity review — 9 November 2026 evening#
50 further Norwegian historical identities receive explicit NVE ID crosswalks and 400 hydraulic/registry assertions. Review uses station names, country, source GIS features and API registry IDs; no automatic fuzzy merge is used. The 181 retained remaining identities yield 50 accepted and 131 held decisions. All 50 historical/current rating comparisons remain quarantined from scalar promotion. Current status, capacities and coordinates are unchanged, and aggregation remains unresolved. New facts retain gross-head basis (67 percent reservoir filling), maximum turbine flow, 1991–2020 normal-period generation, owner/actor and concession metadata, and basin names.
The exact canonical identity and original GIS feature are frozen in
research/nve-crosswalk-20261109-evening.json; API and GIS source registry
entries retain capture timestamps, checksums and NLOD attribution to NVE.
Replay scripts/enrich_hydro_nve_crosswalk.py; drift and conflicting IDs fail
closed. 160 records now carry NVE assertion extensions. Total catalogue coverage
remains 7,449 records / 128 countries. Existing viewer and gallery expose these
assertions; no MW totals are promoted. Next review excludes these accepted UUIDs
from the 131 remaining holds and focuses on aliases and phase/replacement scope.
NVE registry supplement — 10 November 2026#
A fresh NVE API capture adds 400 previously absent assertions to 50 existing Norwegian identities: owner organization ID, first waterfall exploitation year, oldest generating-part year, energy equivalent (kWh/m³), county and municipality IDs, REGINE catchment ID and electricity price area. Municipality codes are the source’s two-digit component; price areas are not physical grid connections. Technical age is not whole-plant commissioning, and energy equivalent is not storage energy. Two source identities are pumped-storage facilities; no generating or pumping ratings are inferred or changed.
Exact reviewed canonical/NVE IDs and names are reused; no new plants or fuzzy
merges. All 131 unresolved identity holds and previous capacity conflicts remain
unresolved. Coverage remains 7,449 records / 128 countries. NVE NLOD attribution,
retrieval time, raw checksum and field locators are in the source registry and
research/nve-supplement-20261110.json. The supplemental assertions live in
extensions_json.nve_registry_supplement, preserving earlier evidence unchanged.
Replay with python scripts/enrich_hydro_nve_supplement.py; unchanged replay does
not change timestamps. Next: first 50 retained NVE identity holds, with primary
owner/concession evidence needed to resolve physical scope.
NVE identity holds: owner review, 10 November 2026 evening#
Reviewed the first 50 retained identities: 6 accepted crosswalks, 44 still held. Bogna, Eidsfoss (Nome, NVE 72), Eikelandsosen, Evanger, Fagerli and Fagervollan gain 96 registry assertions and 24 independently attributed owner assertions. Exact reviewed station coordinates, names and registry IDs are frozen; Eidsfoss is explicitly disambiguated from its namesakes. Six historical-versus-registry MW comparisons remain quarantined from scalar promotion. All 7,449 records, 128 countries, scalar capacities, lifecycle, coordinates and creation dates are preserved. There are now 125 unresolved NVE identities, including 44 reviewed here.
Owner fields include turbine types/counts, unqualified head, commissioning-year
precision, mean generation with unspecified averaging period, intake reservoirs
and permitted reservoir level ranges (not volume or storage MWh). These are under
extensions_json.norway_owner_assertions; NVE assertions keep their separate
source and basis. Owner content has no identified open licence: only limited
normalized facts are retained, with URL, capture time and HTML checksum. NVE
uses NLOD. Earlier GIS capture dates remain explicit.
Replay: python scripts/enrich_hydro_nve_owner_review.py. Research decisions and
conflicts: research/nve-held-review-20261110.json. Next: the remaining 81
unprocessed holds after Hjartdola; the 44 reviewed holds need alias or physical
scope evidence. No automatic fuzzy merges or inferred locations are used.
11 November 2026: final retained NVE hold partition#
The frozen review triages the final 81 previously unprocessed holds and accepts 22 existing Norwegian plant identities: 374 NVE assertions and 157 independently attributed owner assertions. No new plants are inserted. The catalogue remains 7,449 records across 128 countries/territories; 188 identities now have NVE links. The other 59 candidates remain held, alongside 44 from the previous review (103 total). Pump-only Hjorteland is explicitly excluded from generating-plant acceptance. Repeated GIS features for Vamma share registry ID 488 and remain evidence for one canonical plant.
Eleven supported technology classifications include Saurdal, Tevla and Nygard as
pumped storage, without inferring pure/mixed subtype or pumping MW. These classifications
feed the existing viewer filters and separate pumped-storage display. All 22 physical
capacity scopes remain unresolved, so no generating/pumping MW is added to KPIs.
Nine owner/NVE generating-rating disagreements and Kobbelv’s conflicting turbine
counts are retained in canonical conflicts_json with null scalar fields. The ledger
also retains 21 historical/registry rating comparisons and Tussa’s inconsistent annual
production statements. These categories overlap and must not be added as plant counts.
Owner evidence adds unit groups, turbine types, reservoir and dam names, explicitly qualified water volumes, heads, grid voltages, commissioning/refurbishment periods, and Skibotn’s minimum-flow obligation. Generator MVA, generating MW and turbine flow remain distinct. Straumsmo’s 2026 works remain a plan; no completion is inferred. Statkraft’s original HTML returned HTTP 403; its primary pages were inspected through web retrieval, with extracted-text hashes distinguished from original-file hashes. Other owner HTML captures have original SHA-256 checksums. Owner dates are unknown unless stated; annual production has no invented measurement year. NVE is attributed under NLOD; owner pages contribute limited factual selections, not reproduced prose.
Replay this batch offline with:
uv run python scripts/enrich_hydro_nve_owner_review.py --batch catalogues/hydro/research/nve-held-review-20261111.json
uv run python -m energy_data_retrieval.hydro_catalogue validate
uv run python -m energy_data_retrieval.hydro_catalogue build --output artifacts/hydro-20261111
uv run python examples/database/plot_hydro_catalogue.py
Canonical JSONL and frozen evidence are committed. Typed Zstandard Parquet is generated reproducibly and published by the existing documentation pipeline, rather than committed as a competing data source. Next regional partition: African Hydropower Atlas Inga I/II and Poubara owner-backed plant/phase boundaries; retained Norwegian holds are persistent.
SNEL historical station review — 11 November 2026#
The complete 12-station hydro table on page 24 of SNEL’s 2015 annual report adds five reviewed historical identities (Inga I, Tshopo, Bendera, Mobayi and Kilubi) and enriches seven existing records. The batch retains 96 reported cells plus 12 exact MWh-to-GWh conversions: commissioning periods, installed generating groups, group availability, installed MW, average available MW, availability percentages and gross generation. Average available MW is not current operational MW. Dates remain reported periods; no exact coordinates, pumping power or storage energy are inferred. All capacity aggregation roles remain unresolved, so no additional MW enter totals.
Page 24’s table uses M’Sha while its chart uses Mwadingusha. Both existing IDs are retained and the pair is flagged for explicit duplicate resolution, never automatically merged. Poubara remains held pending phase and reuse review. The current SNEL website’s inconsistent Nseke/Koni grouped ratings are excluded. ARE transmission-circuit MVA ratings are not interpreted as plant MW.
The original annual-report PDF was downloaded and checksummed, and the table
was visually checked. No open report licence was identified; only normalized
factual selections are retained with attribution, not report prose or images.
Frozen cell strings, decisions, locators, access assessments and checksum are
in research/snel-2015-review-20261111.json and the source registry.
Replay from the repository root:
uv run python scripts/enrich_hydro_snel.py
uv run python -m energy_data_retrieval.hydro_catalogue validate
uv run python -m energy_data_retrieval.hydro_catalogue build --output artifacts/hydro-snel
uv run python examples/database/plot_hydro_catalogue.py
The existing shared viewer and gallery consume these records automatically. Stations without coordinates remain visible in filters/counts but are not plotted at invented map locations. Typed Zstandard snapshots remain generated from canonical JSONL under the existing repository workflow.