[package]
name = "jouleclaw-corpus"
version.workspace = true
edition.workspace = true
license.workspace = true
repository.workspace = true
authors.workspace = true
description = "Receipts → training corpus. Projects sealed JouleClaw receipts into the labelled (embedding, stage, success, cost_microjoules) rows an EOC router trainer consumes, turning the audit ledger into a supervision signal. Labels are derived from observed cascade closure — never invented; receipts whose closing tier has no EOC analogue are skipped and counted, not guessed."

[dependencies]
jouleclaw-prov   = { path = "../jouleclaw-prov" }
jouleclaw-energy = { path = "../jouleclaw-energy" }
jouleclaw-intent = { path = "../jouleclaw-intent" }
serde            = { workspace = true }
serde_json       = { workspace = true }
thiserror        = { workspace = true }

# Real semantic embedder, opt-in. Pulls candle + tokenizers + hf-hub, so it is
# behind a feature: the default corpus build stays light and offline. See
# `CandleTextEmbedder`. hf-hub + tokio are needed only by the semantic example
# (loading the model is async and hits the hub); they ride the same feature so
# the default build takes none of them.
jouleclaw-embed-candle = { path = "../jouleclaw-embed-candle", optional = true }
jouleclaw-hf-hub       = { path = "../jouleclaw-hf-hub", optional = true }
tokio                  = { workspace = true, optional = true }

[features]
default = []
candle  = ["dep:jouleclaw-embed-candle", "dep:jouleclaw-hf-hub", "dep:tokio"]

[[example]]
name = "emit_corpus_semantic"
required-features = ["candle"]

[dev-dependencies]
tempfile = { workspace = true }