1 · Climate trends & projections IN CR

Serves CN C.1 · FP B.1. Every GCF proposal opens with the climate context — a defensible account of how the climate has already shifted in the project area and how it is projected to change, from historical timeseries and future projections at admin level. The datasets must answer: what is the temperature and rainfall baseline and trajectory here, by season and scenario?

Verdict. Core capability exists and is the notebook’s strongest section. Gap is breadth of SSP/period coverage in the exposed outputs, not the underlying data.

2 · Extreme events IN CR

Serves CN C.1 · FP B.1, D.1. A proposal must show the acute hazard is real and rising — that damaging extremes (drought, heat, heavy rain) already occur and intensify under future scenarios. The datasets must answer: how often do unusual / extreme events hit, historically versus projected, and is the frequency climbing?

Verdict. Matches the Togo SAT z-score approach (Fig. 5 rainfall anomalies). Note the Togo document’s own Z=1 / Z=2 labelling is internally inconsistent with its annex — resolve the threshold labels when this feeds a template.

3 · Crop & livestock hazard exposure PARTIAL

Serves CN C.1, C.2 · FP B.1, D.1. The crux of the adaptation rationale: what is at stake — the value of agricultural production and livelihoods exposed to the hazard, valued in USD (in the style of the Togo SAT Table 5 example). The datasets must answer: how much crop and livestock production value is exposed to compound hazards, where, and under which scenario?

Present in the notebook (verified against the Atlas CR data audit, 2026-07-09): MapSPAM 2020 v1r2 production value × GLW4-2020 livestock (a combined crop + livestock exposure parquet — livestock is present, not partial) × the hazard indices; Ecocrop-default thresholds. Scope: Sub-Saharan Africa only, adm1 only (the champion's "biggest limitation"); fisheries / aquaculture absent. Only the severe severity tier is live (moderate / extreme pending). Valued in nominal 2021 USD, not constant dollars.

Table 3 — Theme 3 datasets: crop & livestock hazard exposure. Source: HAVE in CDH/Atlas · CACC1 Cesare's memo · CACC2 added by this review. Priority: P1 do first · P2 next · P3 later (rubric above).

Verdict. Highest-value theme — the hazard-exposure matrix in the style of the Togo SAT Table 5 example (see Example target output). Priorities: water-stress additions and coverage. We already hold the open GLW4 gridded animal-numbers layer, so GLEAM isn't needed to count livestock; pair Aqueduct bws with GMIA to resolve stress onto irrigated cropland. The open coverage call: served data is SSA + adm1 only (Atlas audit 2026-07-09), but the active GCF pipeline spans non-African countries (Syria, Iraq, Sri Lanka, Egypt) — so the team must decide the minimum v1 coverage, and whether to run global MapSPAM or signpost an external global product where GCF prefers one.

Overlaps → GLW4 (open animal numbers, held) supersedes GLEAM for counting livestock — GLEAM only adds emission intensity, so it belongs in Theme 6, not here. Sea Around Us overlaps FishStat — add only if sub-national/marine detail is needed beyond FishStat's national totals.

4 · Vulnerability & socioeconomic context PARTIAL

Serves CN Exec Summary · FP D.4. The executive summary and needs case must establish who is vulnerable and why concessional GCF finance is warranted — poverty, food insecurity, water access, gender. The datasets must answer: how vulnerable are the people in the project area, on composite indices and sectoral measures?

Present in the notebook: World Bank indicators (poverty rates, GDP by sector, land use) in Key Facts — national only.

The crux of this theme: most memo datasets are national-only (admin0) — they can only shade a whole country. Genuine subnational granularity comes from DHS, IPC/CH, INFORM-Subnational (regional coverage only), and the CACC2 additions (GDL SHDI, Meta RWI, FEWS NET).

Table 4 — Theme 4 datasets: vulnerability & socioeconomic context. Source: HAVE in CDH/Atlas · CACC1 Cesare's memo · CACC2 added by this review. Priority: P1 do first · P2 next · P3 later (rubric above).

Verdict. Additions are mostly public, low-risk integrations (memo action a) — good early Hub wins. The real design decision is subnational coverage: add the CACC2 trio (GDL SHDI, Meta RWI, FEWS NET) so admin1/2 mode is more than a single country colour. Non-commercial flags on IPC, JMP, GDL are fine for GCF/non-profit use but must ride through to any LLM-skill citation.

Overlaps → GDL SHDI is the subnational version of UNDP HDI (promote GDL for admin1/2, keep HDI for the national headline); FEWS NET and IPC/Cadre Harmonisé both classify acute food insecurity (FEWS adds livelihood zones + open licence + reaches gaps IPC misses); ND-GAIN and INFORM Risk are both national composite indices — one may suffice.

5 · NDC & NAP alignment DEPRIORITISED

Serves CN A.16 · FP D.5. A proposal must show it aligns with the country's own climate commitments — NDC targets and NAP priorities — to demonstrate country ownership. The datasets must answer: what has the country pledged for agriculture, land use and water, and does this project deliver on it?

Deprioritised — but unpacked for the dev team. The memo named this the highest-priority new section, but on 2026-04-29 the champion withdrew the automation ask (the NAP/NDC corpus — roughly a dozen documents a year — is too small to justify text-extraction automation; at most a metadata page that links out). We record the conflict rather than silently picking one. Still, if it is ever revisited, here is how each source would connect: three of the four are PDF document registries that need NLP text-extraction; only Climate Watch exposes structured NDC content via an API, so it is the pragmatic connect route.

Table 5 — Theme 5 datasets: NDC & NAP alignment. Source: HAVE in CDH/Atlas · CACC1 Cesare's memo · CACC2 added by this review. Priority: P1 do first · P2 next · P3 later (rubric above).

Verdict. Stays deprioritised (champion withdrew automation; corpus is small). If revisited, start with Climate Watch's API — it already structures NDC targets, so most of the NDC side needs no text-extraction; reserve NLP for the NAP PDFs. A metadata/signpost page linking out is the low-effort interim.

6 · Impact potential & beneficiary estimates NEW

Serves CN A.6–A.7 · FP D.1, E.3. A proposal must estimate its impact — how many beneficiaries are reached and, for mitigation, how much GHG is avoided. The datasets and tools must answer: how many people benefit, and what is the tCO₂e mitigation impact?

Present in the notebook: none (WorldPop population comes via Section 4).

Read this theme as two layers: EX-ACT and the IPCC EFDB are accounting machinery, not spatial datasets — other layers supply activity data per admin unit (land-cover areas, livestock heads, holdings), which the factors/tools convert to tCO₂e for the CN A.6 mitigation claim.

Table 6 — Theme 6 datasets: impact potential & beneficiary estimates. Source: HAVE in CDH/Atlas · CACC1 Cesare's memo · CACC2 added by this review. Priority: P1 do first · P2 next · P3 later (rubric above).

Verdict. Mitigation accounting (EX-ACT, IPCC EFs) is new territory — CGIAR's prior GCF work is adaptation-only, so the L2 mitigation focal point must be engaged. Build EX-ACT/EFDB as a computation layer, not a tile service. Prefer the open CACC2 trio (GLW4, ESA WorldCover, EDGAR/FAOSTAT — all CC BY 4.0) for activity data; note GYGA and GAEZ v4 are both non-commercial (derive-and-cite only). National census is the recurring manual cost — budget it.

Overlaps → GYGA and GAEZ v4 both give attainable yield / yield-gap and are both non-commercial — pick one, don't ingest both. GLW4 (held) and national census both supply livestock/holding activity data — census only where finer ground-truth is essential.

7 · Theory of change & GCF portfolio evidence NEW

Serves FP B.2, D.2 (memo action (b)). A proposal must justify its theory of change with evidence and show it is not duplicating what other funders already back. The datasets must answer: what interventions are shown to work here, and what comparable food / land / water projects have the funds already financed?

Present: the CACC1 MCF project-level dataset (5,115 GEF/GCF/AF records) is the seed/spine — see the Notebook tab.

Geometry note (all rows): these carry country as a text/ISO attribute only — no sub-national geometry, many regional/global records. For the geoselector they are admin0-after-ISO3-join; this theme is better served by the LLM-skill / citable-output route than by mapping. Keep two functions separate: comparable-projects (funds DBs) vs evidence (EGM/GAMI, feeding FP B.2/D.2 narrative).

Table 7 — Theme 7 datasets: theory of change & GCF portfolio. Source: HAVE in CDH/Atlas · CACC1 Cesare's memo · CACC2 added by this review. Priority: P1 do first · P2 next · P3 later (rubric above).

Verdict. Strongly overlaps the Notebook tab — the MCF dataset is the natural engine; sequence them together. Licences are not uniformly open: IEG, OECD CRDF, GAMI and the IATI feeds are clean CC BY; GCF (unstated), GEF and AF (all-rights-reserved), IFAD (unstated) and CGIAR 3ie EGMs (NC-SA) need link-out-and-cite or written permission — don't silently derive-and-host. Use OECD CRDF + IATI as the open comparable-projects backbone.

Overlaps → heavy — the held MCF dataset already unifies GEF/GCF/AF projects, and OECD CRDF + IATI supersede pulling each fund's own DB (demote the individual GCF/GEF/AF DBs to source-of-record / drill-down); CGIAR evidence maps and GAMI are both evidence corpora (GAMI is CC-BY, cleaner); WB IEG and IFAD IOE both supply project-evaluation ratings.

8 · Safeguards & gender screening NEW

Serves CN C.4 · FP G.1–G.2. A proposal must pass environmental & social safeguards and gender screening — protected areas, biodiversity, indigenous lands, land tenure, SEAH risk. The datasets must answer: what environmental, social and gender risks does the project zone carry?

Present in the notebook: none.

Two delivery routes here: polygon/raster layers (WDPA, IUCN, KBA, LandMark, GFW) drive project-zone overlays in the notebook; tabular-by-country layers (SIGI, CPIA, FAO 5.a.x, CSPD tier) suit the LLM skill (ISO3 join → cited sentence). Split the portfolio by licence, not theme: rehost the CC-BY ones; derive-then-host the non-commercial ones (host the computed overlay stat, never the raw layer).

Table 8 — Theme 8 datasets: safeguards & gender screening. Source: HAVE in CDH/Atlas · CACC1 Cesare's memo · CACC2 added by this review. Priority: P1 do first · P2 next · P3 later (rubric above).

Verdict. Public global layers, largely project-zone overlay work — good once Sections 3–4 land. Licence split is the rule: rehost the CC-BY set (GFW, CPIA, FAO 5.a.x, SIGI); derive-then-host the non-commercial set (WDPA, IUCN, KBA, CSPD) — computed stat only, federate users to source. LandMark fills the indigenous-territories gap but carries FPIC conditions. ACLED intentionally omitted — its conflict/SEAH signal already rides through the CACC1 CSPDxCF.

Overlaps → WDPA (protected areas) and KBA (biodiversity-significant sites, incl. unprotected) overlap but are complementary — KBA fills what WDPA misses; IUCN Red List ranges feed the same biodiversity screen. Promote the open GFW / CPIA / FAO 5.a.x; the NC layers (WDPA / IUCN / KBA) all share the derive-then-host pattern.

9 · Financial context & justification NEW

Serves CN D.1–D.4 · FP B.5, C.1. A proposal must justify its funding request and additionality — the climate-finance landscape, the adaptation-finance gap, fiscal space and co-financing. The datasets must answer: why is GCF finance needed and additional, and what do comparable approved projects cost?

Present: the CACC1 MCF dataset carries project financing fields (project value, GCF financing, co-financing).

One client covers most of this theme: a single SDMX/Data360 client pulls OECD CRS + CRDF and the IMF/World Bank finance series — all natively country-year, so they drop straight into the geoselector and self-cite (deterministic query URL = citation).

Table 9 — Theme 9 datasets: financial context & justification. Source: HAVE in CDH/Atlas · CACC1 Cesare's memo · CACC2 added by this review. Priority: P1 do first · P2 next · P3 later (rubric above).

Verdict. Bundle with Section 7 (both lean on the MCF dataset). Ingest CRS + CRDF from one OECD SDMX endpoint (CRDF is the climate view of CRS — not a separate pipeline); route IMF through Data360 to stay within terms. CPI is regional, not country — the biggest surprise; keep it as a citable benchmark only and clear its all-rights-reserved terms before republishing any figure. Dropped: IMF Climate Dashboard (repackages CRDF) and Climate Funds Update (overlaps MCF, no API/licence).

Overlaps → OECD CRDF is the climate-tagged subset of OECD CRS — ingest from one SDMX endpoint, don't treat them as two pipelines; the held MCF dataset overlaps the multilateral-fund flows (use it for GCF/GEF/AF, CRS/CRDF for the wider donor picture); CPI is regional so it duplicates nothing but can't serve as a country layer.