Invariants¶

Invariants are the promises that should survive ordinary implementation change.

This page names the truths the package is trying hardest not to lose. If an invariant changes, that should feel more like a design event than a routine code edit.

Treat the quality pages for bijux-canon-ingest as the proof frame around the package. They should show how trust is earned and where skepticism still belongs.

Visual Summary¶

flowchart RL
    page["Invariants<br/>clarifies: see proof | see limitations | judge done-ness"]
    classDef page fill:#dbeafe,stroke:#1d4ed8,color:#1e3a8a,stroke-width:2px;
    classDef positive fill:#dcfce7,stroke:#16a34a,color:#14532d;
    classDef caution fill:#fee2e2,stroke:#dc2626,color:#7f1d1d;
    classDef anchor fill:#ede9fe,stroke:#7c3aed,color:#4c1d95;
    classDef action fill:#fef3c7,stroke:#d97706,color:#7c2d12;
    proof1["tests/invariants for long-lived repository promises"]
    proof1 --> page
    proof2["tests/unit for module-level behavior across processing, retrieval, and interfaces"]
    proof2 --> page
    proof3["tests/e2e for package boundary coverage"]
    proof3 --> page
    risk1["CHANGELOG.md"]
    risk1 -.keeps trust honest.-> page
    risk2["pyproject.toml"]
    risk2 -.keeps trust honest.-> page
    risk3["README.md"]
    risk3 -.keeps trust honest.-> page
    bar1["proof before confidence"]
    page --> bar1
    bar2["done means defended behavior"]
    page --> bar2
    bar3["package trust after change"]
    page --> bar3
    class page page;
    class proof1,proof2,proof3 positive;
    class risk1,risk2,risk3 caution;
    class bar1,bar2,bar3 action;

Invariant Anchors¶

package boundary stays explicit
interface and artifact contracts remain reviewable
tests continue to prove the long-lived promises

Supporting Tests¶

tests/unit for module-level behavior across processing, retrieval, and interfaces
tests/e2e for package boundary coverage
tests/invariants for long-lived repository promises
tests/eval for corpus-backed behavior checks

Concrete Anchors¶

tests/unit for module-level behavior across processing, retrieval, and interfaces
tests/e2e for package boundary coverage
README.md

Use This Page When¶

you are reviewing tests, invariants, limitations, or ongoing risks
you need evidence that the documented contract is actually defended
you are deciding whether a change is truly done rather than merely implemented

Decision Rule¶

Use Invariants to decide whether bijux-canon-ingest has actually earned trust after a change. If one narrow green check hides a wider contract, risk, or validation gap, the work is not done yet.

What This Page Answers¶

what currently proves the bijux-canon-ingest contract instead of merely describing it
which risks, limits, and assumptions still need explicit skepticism
what a reviewer should be able to say before accepting a change as done

Reviewer Lens¶

compare the documented proof story with the actual test layout and release posture
look for limitations or risks that should have moved with recent behavior changes
verify that the claimed done-ness standard still reflects real validation practice

Honesty Boundary¶

This page explains how bijux-canon-ingest is supposed to earn trust, but it does not claim that prose alone is enough. If the listed tests, checks, and review practice stop backing the story, the story has to change.

Next Checks¶

move to foundation when the risk appears to be boundary confusion rather than missing tests
move to architecture when the proof gap points to structural drift
move to interfaces or operations when the proof question is really about a contract or workflow

Purpose¶

This page records the kinds of promises that should not drift casually.

Stability¶

Keep it aligned with invariant-focused tests and documented package guarantees.