# UCDP annual aggregate context Palimpsest publishes a reviewed, receipt-bound annual aggregate derived from three registry-pinned Uppsala Conflict Data Program (UCDP) version 26.1 bulk archives. The exact public artifact is [`readings/ucdp-aggregate-latest.json`](../readings/ucdp-aggregate-latest.json), and its deterministic release decision is [`readings/ucdp-aggregate-release-receipt.json`](../readings/ucdp-aggregate-release-receipt.json). This is an aggregate historical-context surface for Balochistan and Myanmar. It is not an event feed, actor dossier, route monitor, early-warning system, or attribution engine. ## Rights and citation gate Every acquisition captures the exact UCDP Download Center page and a canonical receipt containing its URL, byte count, SHA-256, observation time, and transport policy. An independent Palimpsest review must then bind that page snapshot, its receipt, a review date, an expiry date, the three input archives, their receipts, the actor-registry count and digest, both exact public-array digests, and all three required citations into [`config/ucdp_acquisition_lock.json`](../config/ucdp_acquisition_lock.json). The lock is closed by [`protocol/ucdp-reviewed-acquisition-lock-v1.schema.json`](../protocol/ucdp-reviewed-acquisition-lock-v1.schema.json) and independently revalidated by the strict typed parser. The approved lock SHA-256 is `5cdce957a4be6eb66f22634cfdc3c59ac7ac7882e03e186be0bb61ae4d4c8dba`. The reviewed Git lock is Palimpsest's explicit approval anchor. It is not a cryptographic signature by UCDP, proof that UCDP approved the acquisition, or a substitute for checking the captured rights page. The lock permits only annual aggregate context with attribution, and its exact SHA-256 travels in the public bundle. The required references are: - rights and citation index: - license identified by the reviewed decision: - Davies, Pettersson and Oberg, “Organized violence 1989-2025, and violent political protests”: - Gleditsch et al., “Armed Conflict 1946-2001: A New Dataset”: - Sundberg and Melander, “Introducing the UCDP Georeferenced Event Dataset”: ## Evidence boundary The 26 September 2026 review captured the current Download Center rights page with SHA-256 `6739bd1dcc85b5e3467342c0da4e38577375eba8ac94470e6a78f49c77d97f2a`. It still states CC BY 4.0 with the dataset-specific citations. All three newly retrieved archives and decompressed members match the previously reviewed bytes; all 331 conflict-year and 74 country-year observations are unchanged. Only acquisition identities, review clocks, and their derived bindings changed. The decision expires at `2026-10-26T06:24:19Z`; arrange the next source review by 19 October. Extending an expiry without reviewing a fresh rights snapshot is not a renewal. The private acquisition layer accepts exactly these versioned archives: | Input | Exact member | Encoding | Public use | |---|---|---|---| | UCDP/PRIO Armed Conflict 26.1 | `UcdpPrioConflict_v26_1.csv` | UTF-8 | annual conflict, territory, and source-side actor IDs | | UCDP Actor 26.1 | `Actor_v26_1.csv` | Latin-1 | referential-integrity validation; names never cross the public boundary | | Organized Violence Country-Year 26.1 | `OrganizedViolenceCYDataSet26_1.csv` | UTF-8 | Pakistan and Myanmar annual low/best/high fatality bounds | Each acquisition receipt binds the exact URL, ZIP byte count and SHA-256, exact member name, decompressed byte count and SHA-256, HTTP `Last-Modified`, retrieval time, version, transport policy, and byte ceilings. These Palimpsest receipts prove captured-byte identity and replay consistency; they do not authorize publication. Redirects are disabled. TLS certificate and hostname verification are required. The safe-fetch layer pins validated public IP addresses, while the ZIP layer independently enforces one flat member, CRC validity, supported compression, decompressed-size caps, and a compression-ratio ceiling. The public verifier hashes each complete public acquisition receipt and requires an exact match to its reviewed lock pin, then independently rechecks its URL, member, byte ceilings, archive/member identities, transport policy, and fixed receipt order against the registry and lock. Publication uses separate current, publication, retrieval, source, rights observation, review, and expiry clocks. The gate requires: - an explicit publication clock no earlier than the rights review and no more than 300 seconds ahead of the current build clock; - both the publication and current clocks to be within the rights decision's validity window; - rights observation, review, and expiry clocks in order, with future clocks rejected beyond the same 300-second allowance; - the latest input retrieval no later than the rights-page observation, and the observation no later than review and publication; - all three retrieval clocks within 900 seconds of one another; and - each input's `Last-Modified` and retrieval clock no more than 550 days old at both publication and build time, and not future-dated beyond 300 seconds. A missing or duplicate `Last-Modified`, unknown actor ID, changed CSV header, changed version, cumulative-total disagreement, incomplete country-year matrix, extra ZIP member, expired decision, stale clock, or receipt/hash/lock mismatch fails closed. Failure never produces zero fatalities, an empty “healthy” history, or a fresh timestamp. ## Public contract The public JSON schema is [`protocol/ucdp-aggregate-v1.schema.json`](../protocol/ucdp-aggregate-v1.schema.json). The checked-in version 26.1 aggregate contains 331 conflict-year records, 74 country-year records, and referential-integrity coverage for 1,928 UCDP actor IDs over 1948–2025. Its exact SHA-256 is `07e8ed2b9861ef54dcebf9c7fe0a4efe3f23b2742231d5dbe5173f14699701b1`. It permits only: - Balochistan conflict-year rows where UCDP location is Pakistan and territory is exactly Balochistan; - Myanmar conflict-year rows where the reviewed UCDP location token is present; - Myanmar aggregate territory labels only from the reviewed version 26.1 vocabulary (a new label fails closed for review); - annual conflict ID, aggregate territory name, and no more than 64 distinct source-side actor IDs per side; - Pakistan and Myanmar country-year state-based, non-state, and one-sided uncertainty bounds; and - an explicit total that must agree with both the three category bounds and the UCDP cumulative total fields. The Pakistan country-year bounds describe all organized violence within Pakistan's borders. They are **not** Balochistan-only counts. The Balochistan surface contains conflict-year identifiers but does not derive provincial death totals from the country-wide table. The contract and schema prohibit event coordinates, village names, event dates, event narratives, source articles, person records, live or tactical fields, and drug-actor inference. No UCDP row is joined to NarcoScope by actor, route, project, guilt, or causality. A downstream product may use only coarse geography and annual historical time as contextual facets, and must preserve each product's separate provenance and rights. ## Baloch movement semantics “Baloch movement” is an umbrella research concept, not one modeled actor. The bundle carries five non-interchangeable lanes: 1. civic society — unavailable from these UCDP inputs; 2. electoral and political activity — unavailable from these UCDP inputs; 3. armed-conflict organizations — distinct UCDP side-B actor IDs by conflict-year; 4. state authorities — distinct UCDP side-A actor IDs by conflict-year; and 5. human-rights documentation — unavailable from these UCDP inputs. Unavailable lanes require separate rights-reviewed evidence. Political parties, peaceful advocacy, human-rights allegations, administrative designations, and armed organizations must never be merged merely because they share a geography or a broad political theme. ## Private acquisition and offline replay Network acquisition is a private-only step. It writes eight immutable files and does not build a public artifact: ```bash mkdir -m 700 /private/evidence/ucdp-26.1 python3 -m scripts.ucdp_bulk_pull acquire \ --fetch \ --evidence-output-dir /private/evidence/ucdp-26.1 ``` The evidence directory contains: ```text armed_conflict.zip armed_conflict.receipt.json actor_registry.zip actor_registry.receipt.json organized_country_year.zip organized_country_year.receipt.json rights-page.snapshot.html rights-page.receipt.json ``` `archive-check` verifies private byte, ZIP, receipt, and rights-snapshot self-consistency. It deliberately does not consult the reviewed lock and cannot authorize a public build: ```bash python3 -m scripts.ucdp_bulk_pull archive-check \ --input-dir /private/evidence/ucdp-26.1 ``` After an independent reviewer has inspected the exact evidence and committed an approved lock, `check` can evaluate publication eligibility without writing an artifact: ```bash python3 -m scripts.ucdp_bulk_pull check \ --input-dir /private/evidence/ucdp-26.1 \ --publication-at 2026-08-26T19:28:50Z ``` With the same reviewed evidence, build the public review artifact outside the private evidence directory: ```bash python3 -m scripts.ucdp_bulk_pull build \ --input-dir /private/evidence/ucdp-26.1 \ --publication-at 2026-08-26T19:28:50Z \ --output /review/ucdp-aggregate-v1.json ``` The same ZIPs, receipts, rights snapshot, approved lock, and publication clock produce identical JSON bytes and the same bundle ID. `generated_at` is the explicit publication clock; `latest_retrieved_at` separately preserves the latest receipt-bound input retrieval time. `check` and `build` deliberately do not accept a caller-selected lock path. Publication authority comes only from the repository-controlled `config/ucdp_acquisition_lock.json`; accepting a temporary lock would let self-issued bytes approve themselves. Running `check` with no evidence only validates the adapter configuration and reports the lock state. Public-release verification is a separate, private-data- free gate: ```bash python3 -m scripts.verify_ucdp_public_release check \ --current-at 2026-08-26T19:34:49Z ``` That verifier requires exact canonical aggregate bytes, the closed aggregate and receipt schemas, the approved lock, current rights and evidence clocks, the registry and all semantic scrubs. The release receipt makes the external trust precondition explicit: approval comes from a protected reviewed Git revision plus its exact release manifest. The lock binds the reviewed evidence identity; it is not an origin signature. The eight private acquisition files are neither tracked nor served. This repository promotion does not itself acquire data, deploy Railway, or prove that any host is serving the candidate revision.