899ca0b97c
The open-standards quintet's commercial pitch made real. Before
this commit the README claimed JouleClaw + JCR-1 + OpenPay + WAI
+ ARL + Sandbox + Smart Byte + AAC compose into one stack, but
nothing in the workspace ran a multi-standard pipeline end-to-end.
The new example, openpay_per_joule_billing.rs, runs three
standards in one binary:
1. JouleClaw cascade walk through JouleClawStack::with_defaults
— counts every joule actually spent (one L3 + one L0-cache
hit, ~2 J total).
2. JCR-1 receipt minted over the session via the wave-7
canonical det-CBOR + COSE_Sign1 path (335 B core / 456 B
envelope, signed by the published test steward key).
3. OpenPay settlement Transaction emitted in op-ledger v1's
serde shape (status, ledger_id, entries[] with proper
Money/Currency/Direction encoding, metadata pinning the
JCR-1 receipt id) — consumable by any conforming OpenPay
ledger via serde_json::from_value::<Transaction>.
Plus a binding document linking the receipt id to the Transaction
id by an explicit, verifiable invariant: receipt.joules_uj /
100_000 == settlement.amount_minor_units (1 ¢ per 100 mJ = $0.0001
per J pricing). An auditor verifies the JCR-1 signature, runs the
binding invariant, and the OpenPay Transaction is provably grounded
in measured energy.
Why no op-ledger import: OpenPay lives in a sibling Cargo
workspace; cross-workspace path deps don't carry workspace.package
inheritance, so direct cross-linking fails. The DEEPER reason for
NOT importing is that this proves the right kind of composition —
standards compose at the WIRE LAYER (serde_json), not at the
crate-graph layer. Standards in different language ecosystems
compose this way by construction.
Three artifacts written to /tmp/:
openpay-per-joule-receipt.json — JCR-1 signed receipt
openpay-per-joule-entry.json — OpenPay Transaction
openpay-per-joule-binding.json — receipt ↔ entry binding
Self-checks the invariant on exit; aborts if it doesn't hold.
WAI / ARL / Sandbox compositions are roadmap, not part of this
commit — same pattern (wire-layer serde_json), different
standards. Adding them is mechanical.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
1.8 KiB
1.8 KiB
jouleclaw-stack — starter cascades
This directory exists because the five-lens wave-7/8 cross-talk audit (Phillip / Rafie / Bloom / Hyper-Agent / Priscilla) independently flagged the same on-ramp gap: a multi-crate workspace, one end-to-end example. Small runnable cascades give a new operator a working starting point per use case.
Each example is self-contained and runs with:
cargo run -p jouleclaw-stack --example <name>
| Example | What it shows | Audit lens |
|---|---|---|
minimal_cache_only |
The smallest cascade walk; L0 cache vs L3 echo. | Phillip — text-is-state |
rag_with_retrieval |
Retrieve → enrich → rerank → read over a seeded index. | Bloom — verifiable zone |
comprehend_demo |
Stack::comprehend — L0–L2 only, with intent tagging. |
Priscilla — understand before you generate |
agent_with_skill_induction |
Register a skill, watch a query class resolve at L0.5. | Phillip — agent-ready APIs |
save_memory_and_promote |
Manual save_memory + auto-learned NotifyPromotedFact events. |
Hyper-Agent — three modes of memory |
ledger_breakdown |
A workload, a ledger, a CostSummary rendered with intent shares. |
Rafie — breakdown at the top |
full_cascade_with_triangle |
Cascade walk + triangle (elicitor → cone → conservation → oracle) + JCR-1 receipt — in one run. | The consolidated end-to-end stake. |
openpay_per_joule_billing |
Cross-standard composition: JouleClaw → JCR-1 → OpenPay. Pay-per-joule pitch made real, three artifacts emitted. | Open-standards quartet — composing at the wire layer, not the crate-graph layer. |
Doctrine in one line
"Inference is the last resort, not the entry point." — SPEC §1.1
Read the comprehend_demo example first if you want to feel why that
sentence is the project's centre of gravity.