aurora-agent

# AURORA Agent (bioprism)

**Query-compiled inference for executable biology.**

Context engineering, with receipts.

An MCP server and CLI built on the FIBER decision-context compiler: a typed decision query is
compiled into the smallest decision-sufficient evidence region, delivered with a Context
Certificate stating exactly what was omitted.

[![CI](https://github.com/AURORA-NEURO/aurora-agent/actions/workflows/ci.yml/badge.svg)](https://github.com/AURORA-NEURO/aurora-agent/actions/workflows/ci.yml)
[![Release](https://img.shields.io/github/v/release/AURORA-NEURO/aurora-agent)](https://github.com/AURORA-NEURO/aurora-agent/releases)
[![License](https://img.shields.io/badge/license-Apache--2.0-blue)](LICENSE)
[![MCP registry](https://img.shields.io/badge/MCP%20registry-io.github.MurariAmbati%2Faurora--agent-blue)](https://registry.modelcontextprotocol.io/?search=aurora-agent)

Implementation of the AURORA BioPRISM / OncoWorld / FIBER blueprint (v0.6, 935 registered spec
modules). A Rust workspace whose central idea is that **context assembly is a compiler pass**:
instead of retrieval plus summarisation plus vibes-based compaction, a typed decision query is
compiled into the smallest decision-sufficient evidence region, delivered as a **Decision
Section**, and accompanied by a **Context Certificate** that states exactly what was omitted and
whether the omission could have changed the decision.

> Compile the smallest decision-sufficient evidence region. Never traverse the whole knowledge
> structure by default.

## What the measurements actually say

The reference world ships 761 facts, 750 of them exploratory distractors that all consume the same
protected `cohort_id` hub. FIBER compiles the query down to **11 facts (1.45% of the world)** and
the deterministic oracle still returns the correct verdict with all four leakage witnesses.

**It is not alone in doing so.** Under equal tuning, a 5-hop incidence walk and a BM25 retriever at
k=11 select *exactly the same eleven facts*. The distribution's own `compare_baselines.py` measures
the graph baseline only at depth 7 and unbounded — the two settings where it returns everything —
and reports a 69× advantage that vanishes under equal tuning. That is a strawman comparison, and
correcting it is what 43.38 and 43.41 require.

So the reference world cannot tell these methods apart. [`crates/worldgen`](crates/worldgen) makes
the structure a parameter and builds one that can — distractors attached near the target instead of
at a hub leaf, decisive facts behind a relay chain, and distractor tags camouflaged to tokenise into
the protected vocabulary:

| Strategy | Facts | Sound? | Closure | Admissible |
|---|---:|:-:|---:|:-:|
| full-context | 762 | yes | 100% | yes |
| graph-5-hop | 750 | **no** | 0% | **no** |
| graph-7-hop | 750 | **no** | 0% | **no** |
| graph-11-hop | 761 | yes | 100% | yes |
| lexical-top-11 (BM25) | 11 | yes | **91%** | **no** |
| **fiber** | **11** | **yes** | **100%** | **yes** |

Three distinct failure modes appear. The graph walk has **no usable depth**: 5–10 pull in 98% of the
world *and still miss every decisive witness*; 11 is the first sound setting and by then it has
taken everything. BM25 reaches the *right verdict* from a **91% protected closure** — right by luck,
having dropped a protected fact that happened not to matter, and raising k to 50 never recovers it.
FIBER is the only admissible strategy: right verdict **and** full closure, at 11 facts.

That last failure is why the harness ranks on admissibility rather than verdict alone — ranking on
verdict would have crowned the strategy that violated the mandatory closure and got away with it.

This does not show FIBER wins generally: the discriminating world was built to expose these modes,
just as the reference world was built to expose hub expansion. Both are single points, the full
sweep is not done, and an embedding retriever and a *directed* dependency walk are still missing
from the panel. Full analysis: [docs/FINDINGS.md](docs/FINDINGS.md). How much of the blueprint the
workspace actually covers, and which sections have nothing standing in for them:
[docs/COVERAGE.md](docs/COVERAGE.md). The crate layout and the blueprint path:
[docs/ARCHITECTURE.md](docs/ARCHITECTURE.md).

## Autonomous agent process boundary

The Python SDK includes a secret-safe operator entry point for the autonomous brain:

```bash
cd python
python -m prism_sdk catalogue
python -m prism_sdk evidence-plan --domain science
python -m prism_sdk provider-status --provider openai
```

For keyless local development, the same boundary supports an explicit credentialless fixture:
`python -m prism_sdk provider-status --provider local` and `run --provider local --model local-model`
use the runtime's bounded in-memory transport; no key or network provider is contacted.

Use `run` with a caller-owned MCP server when you are ready to invoke a provider. Keys are accepted
only through a hidden prompt or an explicitly named environment variable; they are never command
line arguments, MCP arguments, plans, or persisted state. See [the autonomous brain guide](docs/AUTONOMOUS_BRAIN.md#operator-process-boundary)
for model discovery, durable inventory refresh, model-selection, approval, and credential-lifecycle details.

## Status

**79 crates, 518,324 lines, clippy -D warnings enforced in CI.** Byte-level parity with the
CPython reference runtime is enforced by test and holds across *three* implementations: CPython, the
Rust eager path, and the Rust indexed store.

The table below is generated. It used to be hand-maintained and drifted to claiming twenty-three
crates and 820 tests — the same hand-copy drift [`crates/devx`](crates/devx)'s exit-code audit
exists to catch, sitting in the README of the repository that wrote the audit. Regenerate it, and
the test count, with:

```bash
tools/status.sh --tests
```

The **Blueprint** column is derived rather than declared: it lists the sections whose module ids a
crate actually cites in its own source, using the token rule [`tools/coverage.sh`](tools/coverage.sh)
runs. A crate that stops citing a section drops it here without anyone remembering to edit a row.

How much of the blueprint is covered, and what the remainder is:
[docs/COVERAGE.md](docs/COVERAGE.md) and [docs/BACKLOG.md](docs/BACKLOG.md). Every uncovered module
carries a typed verdict in [`crates/residue`](crates/residue) explaining why nothing implements it.

<!-- generated by tools/status.sh at 655e219 -->

| Crate | Blueprint | What it does |
|---|---|---|
| [`bioprism-adapter`](crates/adapter) | 04,28,40,43 | Data adapter contract with mandatory semantic-loss reporting |
| [`bioprism-adaptive`](crates/adaptive) | 08,43 | Adaptive evaluation: capability posterior, information-gain suite selection, parent-aware uncertainty |
| [`bioprism-api`](crates/api) | 11 | Bounded HTTP API, event stream, and signed webhook outbox for the Prism MCP kernel |
| [`bioprism-atlas`](crates/atlas) | 03,33,43 | BioCapability atlas and metrics: capability ontology, coverage, failure atlas |
| [`bioprism-atlashub`](crates/atlashub) | 09,27,34 | BioAtlas surfaces: world cards, connector registry, value-of-experiment, federated evaluation, research CI |
| [`bioprism-atlasx`](crates/atlasx) | 34 | Capability atlas and public-hub remainder: coverage debt as a derived claim, and the failure-atlas browsing surface |
| [`bioprism-backends`](crates/backends) | 32,43 | Physical backend portfolio: variable elimination, worst-case-optimal joins, structural estimation and the honest fallback |
| [`bioprism-baseline`](crates/baseline) | 43 | Equal-engineering context baselines: full-context, k-hop incidence, connected component, lexical top-k and query-graph |
| [`bioprism-benchcompiler`](crates/benchcompiler) | 06,35 | Benchmark compiler: trajectory to decision cell, first causal divergence, minimization, oracle synthesis |
| [`bioprism-bioethics`](crates/bioethics) | 13,30,36 | Section 36 remainder: biology security, privacy, ethics and governance beyond policy and safety |
| [`bioprism-bioeval`](crates/bioeval) | 26,31,43 | Biological evaluation engine: scoring planes, partial credit, biological error classes |
| [`bioprism-bioevalx`](crates/bioevalx) | 07,26 | Bio evaluation engine remainder: scoring planes, reader models, adjudication and the evaluation contract |
| [`bioprism-bioir`](crates/bioir) | 25,39 | Biological IR: BioWorld, specimen lineage, AssayLens, cohort and split, uncertainty and reference standards |
| [`bioprism-biolang`](crates/biolang) | 25,28,39,43 | The biological IR family and BioQL: typed world, state, intervention, worldline, oracle, mutation and bundle representations |
| [`bioprism-bioworlds`](crates/bioworlds) | 30,38,43 | Reference bioworlds and vertical slices: worlds built to make blocked platform claims exercisable |
| [`bioprism-brain`](crates/brain) | 09,11 | Provider-neutral autonomous brain kernel: model routing, prompt assembly, bounded plans, and online bandit state |
| [`bioprism-bundle`](crates/bundle) | 10,12,13,34,43 | Signed result bundles and reproduction: attestation, replay, and what symmetric authentication cannot promise |
| [`bioprism-choreography`](crates/choreography) | 23 | Multiparty choreography: session types with projection, bounded protocol model checking, adjudication, quorum with checked independence, and sagas with honest compensation |
| [`bioprism-cli`](crates/cli) | 40,43 | The bioprism command-line interface |
| [`bioprism-conformance`](crates/conformance) | 14,40,43 | Conformance suites, the test pyramid and release quality gates |
| [`bioprism-cookbook`](crates/cookbook) | 03,11,13,14,19,21,38,39,40,41,43 | Reference examples: worked recipes with the claim each one demonstrates and the property a reader can check |
| [`bioprism-dataops`](crates/dataops) | 12 | Section 12 remainder: storage topology, relational catalog, SLOs, compute placement and federated deployment, each answer carrying