Cautilus Promise Ledger

This ledger names the current Cautilus promises and cross-cutting rules. The names are the reader-facing addresses. Compact keys mirror those names for packets, tables, and checks.

Model index: Promise Ledger. View guide: How The Views Relate. Evidence state: Evidence State.

Workflow Promises

Readiness: Cautilus shows whether the selected repo is ready for a bounded workflow and what setup remains.
Claim Discovery: Cautilus turns selected source docs into source-referenced candidate claims and proof-planning work.
Behavior Evaluation: Cautilus evaluates intentful behavior across supported dev and app surfaces.
Bounded Improvement: Cautilus improves a selected behavior target under explicit budget, protected checks, and held-out evidence.

Workflow Audit Matrix

promise	key	commitment	user workflow	maintainer evidence routes	cross-cutting rules
Readiness	`promise.readiness`	A user can tell whether Cautilus can operate in the selected repo and what setup is missing before spending workflow budget.	Readiness	Readiness And Runtime Status, Adapter And Host Ownership	Evidence Gaps, Vocabulary Consistency, Agent-Human Resumability, Host-Owned Execution
Claim Discovery	`promise.claim-discovery`	A user can scan selected source docs into broad source-referenced candidates and leave curation, likely false negatives, and proof planning visible.	Claim Discovery	Claim Discovery Workflow, Evidence State And Review Artifacts, Binary And Skill Boundary	Reviewable Artifacts, Evidence Gaps, Agent-Human Resumability, Host-Owned Execution
Behavior Evaluation	`promise.evaluation`	A user can evaluate intentful behavior across supported `dev` and `app` surfaces when deterministic tests alone do not explain the behavior.	Behavior Evaluation	Evaluation Surfaces And Runners, Reporting And Review Variants, Live Invocation Runtime	Reviewable Artifacts, Packet Freshness, Cost And Proof Freshness, Host-Owned Execution
Bounded Improvement	`promise.improvement`	A user can improve a selected behavior target while preserving intent, explicit budget, protected checks, held-out evidence, and reviewable revision artifacts.	Bounded Improvement	Improvement Loop, Scenario History And Proposal Normalization, Active Run And Workspace Lifecycle	Reviewable Artifacts, Evidence Gaps, Cost And Proof Freshness, Host-Owned Execution

Cross-Cutting Rules

cross-cutting rule	key	commitment	user attachments	maintainer evidence routes
Reviewable Artifacts	`rule.reviewable-artifacts`	Workflow output remains reopenable as machine-readable packets and readable reports.	Reviewable Artifacts	Evidence State And Review Artifacts, Reporting And Review Variants, Active Run And Workspace Lifecycle
Evidence Gaps	`rule.evidence-gaps`	Missing or weak evidence stays visible until the claim is proven, narrowed, deferred, or removed.	Evidence Gaps	Evidence State And Review Artifacts, Improvement Loop, Readiness And Runtime Status
Host-Owned Execution	`rule.host-owned-execution`	Host repos own prompts, models, credentials, runtime wiring, fixtures, and acceptance policy while Cautilus owns generic workflow packets and boundaries.	Host Ownership	Adapter And Host Ownership, Live Invocation Runtime, Binary And Skill Boundary
Vocabulary Consistency	`rule.vocabulary-consistency`	One concept keeps one name across user prose, maintainer specs, packets, tests, and Cautilus Agent guidance.	Readiness, Claim Discovery, Host Ownership	Binary And Skill Boundary, Adapter And Host Ownership, Evidence State And Review Artifacts
Packet Freshness	`rule.packet-freshness`	Readable views show current packets and source artifacts rather than stale copied state.	Reviewable Artifacts, Evidence Gaps, Behavior Evaluation	Evidence State And Review Artifacts, Reporting And Review Variants, Active Run And Workspace Lifecycle
Cost And Proof Freshness	`rule.cost-and-proof-freshness`	Expensive eval and improve evidence is shown honestly as selected, prepared, stale, or newly executed proof.	Behavior Evaluation, Bounded Improvement	Evaluation Surfaces And Runners, Improvement Loop, Scenario History And Proposal Normalization
Agent-Human Resumability	`rule.agent-human-resumability`	A human or agent can resume from durable packets, next actions, source refs, and source-bound review feedback that can become reusable learning evidence instead of chat memory.	Readiness, Claim Discovery, Host Ownership	Evidence State And Review Artifacts, Binary And Skill Boundary, Active Run And Workspace Lifecycle, Reporting And Review Variants

Ledger Checks

Verify all primary reading views exist.

node -e 'const fs = require("node:fs"); for (const path of ["docs/specs/user/index.spec.md", "docs/specs/contracts/index.spec.md", "docs/specs/rules/index.spec.md", "docs/specs/evidence/index.spec.md"]) { if (!fs.existsSync(path)) throw new Error("missing " + path); }'

run:shell