Cohort Histogram

# Blind Machine curated protocol bundle — cohort_histogram.## This manifest is the ONLY file the Rails platform parses. The author's pure# functions in server.py / local_project_owner.py / local_data_owner.py own the# computation; the numbered stage files are kit-owned shims that call them (see# docs/rfcs/0002). This file owns the governance + review surface (display# metadata + the gate values that let compute be authorized without unpacking the# tarball). See docs/protocol_structure.md for the schema and# docs/protocol_catalog.md §3 for this protocol's definition.## cohort_histogram is deliberately NON-GENOMIC: it aggregates a bucketed count# over an ordered, published bucket definition (demographic age bands in the v1# demo). It reuses the flagship's additive minimal-params BFV loop verbatim to# prove the additive primitive is generic beyond genotypes — genomics is the# demo domain, not the only possible one.name: cohort_histogramversion: 1.0.0builder:  type: uv  project: envcrypto: tenseal-bfv             # display hint; real deps live in env/uv.lockcomputation: additive_bfv       # pure coordinate-wise homomorphic vector addmin_contributors: 25            # catalog §3 (higher than the flagship's 20)# The published bucket definition. Every contributor MUST encode their bucket# membership against this identical ordered definition or the blind histogram is# meaningless — that invariant is what folding this manifest into the bundle# SHA-256 enforces. `length` is the number of buckets B; each contribution is a# one-hot vector h in {0,1}^B with sum(h)=1. The v1 demo uses ordered categories# (demographic age bands); a numeric protocol would instead publish `edges`.input:  type: onehot_vector  length: 10                    # B = number of buckets  value_domain: [0, 1]          # one-hot membership; exactly one coordinate is 1  constraint: one_hot           # sum(h) = 1 per contributor  buckets:    kind: ordered_categories    # or `ordered_edges` with `edges: [...]` for numeric bins    seed: blind-v1-histogram-buckets    labels:                     # the published, ordered bucket definition      - "0-9"      - "10-19"      - "20-29"      - "30-39"      - "40-49"      - "50-59"      - "60-69"      - "70-79"      - "80-89"      - "90+"output:  shape: aggregate_count_vector # counts in Z^B (+ append-1 sentinel -> N)  exactness: exact              # BFV is exact in Z_t; one-hot sums never overflow  tolerance: 0release_policy:  aggregate_only: true          # only the summed histogram is ever decrypted  allowed_runs_per_project: 1   # run cap (differencing mitigation, not a defense)resources:  max_memory_mb: 8192  max_wall_seconds: 3600  max_ciphertext_mb: 512# Collaboration scenario (docs/rfcs/0002 §4.1): who runs which role's functions.# The author writes pure functions grouped BY ROLE into the files below; the# numbered stage shims that call them are kit-owned framework code, materialized# at run time (never signed into the bundle). `aggregate` = 1 project owner + N# data owners.scenario:  name: aggregate  roles:    project_owner: { file: local_project_owner.py, functions: [ keygen, decrypt, decode ] }    data_owner:    { file: local_data_owner.py,    functions: [ encode, encrypt ] }    server:        { file: server.py,              functions: [ compute ] }# Content-addressing digests. Computed at registry ingest — `bundle` by# protocols/hash_bundle.py, `env_lock` over env/uv.lock + env/.python-version +# pinned runner metadata. Left null in the committed bundle ON PURPOSE: the# canonical bundle digest is computed over the raw bytes of every signed file# INCLUDING this manifest, so a computed `bundle` value here would be# self-referential (its own input). The authoritative digest instead NAMES the# bundle (protocol-<name>@<digest>) and is recorded out-of-band# (.blind-signature / DB), never inside the file it would have to hash. See# protocols/README.md § "Why the manifest ships null hash placeholders".hashes:  bundle: null  env_lock: null