rtr-role-type-retrieval 0.4.0__tar.gz

rtr_role_type_retrieval-0.4.0/.github/DISCUSSION_TEMPLATE/falsify-a-claim.yml

@@ -0,0 +1,84 @@
+title: "[Falsify] "
+labels: ["falsify-a-claim"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        This is the room for attacking a claim we have made — a theorem, a
+        measurement, a number in a README, a sentence in a commit message.
+
+        You do not need to be right. You need to name the claim precisely and
+        say what would settle it. A refutation that fails is still worth having:
+        it turns an assumption into something that survived an attack.
+
+        If you already have the counter-example in hand and it is unambiguous,
+        open a **false claim** issue instead — that path is for defects with
+        evidence attached. This one is for the experiment before the verdict.
+
+  - type: textarea
+    id: claim
+    attributes:
+      label: The claim, quoted verbatim
+      description: >
+        Copy it exactly as written. Paraphrase is where most disagreements
+        actually live — half of them dissolve once the original wording is on
+        the table.
+    validations:
+      required: true
+
+  - type: input
+    id: location
+    attributes:
+      label: Where it is stated
+      description: >
+        A permalink, or `path/to/file.ext:LINE`. If it is a claim we made in
+        conversation rather than in the tree, say so — that is a finding in
+        itself, because it means the claim was never written down.
+    validations:
+      required: true
+
+  - type: textarea
+    id: refutation
+    attributes:
+      label: What would refute it
+      description: >
+        The observation that, if it occurred, would force us to withdraw the
+        claim. Be concrete: an input, a command, an exit code, a counter-example.
+        A claim with no stated refutation is not a claim we should have made.
+    validations:
+      required: true
+
+  - type: textarea
+    id: attempted
+    attributes:
+      label: What you ran, and what happened
+      description: >
+        Paste the command and its real output, including the exit code read
+        directly. Negative results belong here too — "I tried to break it this
+        way and it held" is evidence, and we would rather have it than not.
+      render: shell
+    validations:
+      required: false
+
+  - type: dropdown
+    id: standing
+    attributes:
+      label: Where does this stand right now?
+      options:
+        - I think the claim is wrong and I can show it
+        - I suspect it is wrong but I cannot yet demonstrate it
+        - The claim may be true but is stated more strongly than the evidence supports
+        - The claim is untestable as written and needs restating before it can be judged
+        - I tried to refute it and failed — recording the attempt
+    validations:
+      required: true
+
+  - type: checkboxes
+    id: scope
+    attributes:
+      label: Before you post
+      options:
+        - label: I have quoted the claim rather than summarising it
+          required: true
+        - label: I have said what would change my mind, not only what I believe
+          required: false

rtr_role_type_retrieval-0.4.0/.github/DISCUSSION_TEMPLATE/general.yml

@@ -0,0 +1,19 @@
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Anything that does not fit the other categories. If you are unsure where
+        something belongs, this is the right place — it will be moved for you and
+        nobody will mind.
+
+        Two things that do **not** belong here:
+        - **Security findings.** Never post those publicly — use the affected
+          repository's private advisory form.
+        - **Conduct concerns.** Those go privately to the organisation owner.
+
+  - type: textarea
+    id: body
+    attributes:
+      label: What's on your mind
+    validations:
+      required: true

rtr_role_type_retrieval-0.4.0/.github/DISCUSSION_TEMPLATE/ideas.yml

@@ -0,0 +1,38 @@
+title: "[Idea] "
+labels: ["idea"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Early and half-formed is welcome here. This is the room for thinking out
+        loud — a worked proposal with costs and migrations belongs in an issue,
+        but it does not have to start there.
+
+  - type: textarea
+    id: idea
+    attributes:
+      label: The idea
+    validations:
+      required: true
+
+  - type: textarea
+    id: problem
+    attributes:
+      label: What problem it would solve
+      description: >
+        Stated without reference to the solution, if you can. An idea that
+        cannot name its problem is worth discussing anyway — say so, and we will
+        look for the problem together.
+    validations:
+      required: false
+
+  - type: dropdown
+    id: stage
+    attributes:
+      label: How settled is this?
+      options:
+        - A hunch — I want reactions before I think harder about it
+        - A real design — I have considered the cost and I am arguing for it
+        - I am willing to implement it if there is appetite
+    validations:
+      required: true

rtr_role_type_retrieval-0.4.0/.github/DISCUSSION_TEMPLATE/q-a.yml

@@ -0,0 +1,43 @@
+title: "[Q&A] "
+labels: ["question"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        **Questions are never noise here.** If our documentation had answered
+        this, you would not be asking — so an unanswerable question is usually
+        our defect, and often becomes a documentation fix.
+
+        You do not need to be polished. You do need to be reproducible.
+
+  - type: textarea
+    id: question
+    attributes:
+      label: What you are trying to do
+      description: The goal, not only the error. Knowing where you were headed often changes the answer.
+    validations:
+      required: true
+
+  - type: textarea
+    id: tried
+    attributes:
+      label: What you tried, and what happened
+      description: Commands and their real output, including exit codes where relevant.
+      render: shell
+    validations:
+      required: false
+
+  - type: input
+    id: env
+    attributes:
+      label: Repository and platform
+      placeholder: "rtr-role-type-retrieval on Windows 11 / PowerShell 7"
+    validations:
+      required: false
+
+  - type: markdown
+    attributes:
+      value: |
+        ---
+        If it turns out to be a defect rather than a question, we will say so and
+        move it to an issue. You do not need to guess correctly to get help.

rtr_role_type_retrieval-0.4.0/.github/DISCUSSION_TEMPLATE/show-and-tell.yml

@@ -0,0 +1,34 @@
+title: "[Show] "
+labels: ["show-and-tell"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Something you built with one of these tools, a workflow you worked out,
+        a proof you wrote, a way you broke it and what you learned.
+
+        **Reports of breaking our tools are welcome here as much as successes.**
+        "I pushed this until it fell over, here is where" is a genuinely useful
+        post.
+
+  - type: textarea
+    id: what
+    attributes:
+      label: What you made or found
+    validations:
+      required: true
+
+  - type: input
+    id: link
+    attributes:
+      label: Link, if there is one
+    validations:
+      required: false
+
+  - type: textarea
+    id: surprise
+    attributes:
+      label: What surprised you
+      description: The part that did not go as expected is usually the most useful part for everyone else.
+    validations:
+      required: false

rtr_role_type_retrieval-0.4.0/.github/ISSUE_TEMPLATE/adaptation-request.yml

@@ -0,0 +1,79 @@
+name: Adaptation request — RE:Symbiose for your CLI or agent
+description: Ask for RE:Symbiose wiring on a CLI/agent harness we do not ship yet. The README promise applies - adaptations requested here will be delivered.
+title: "adaptation: "
+labels: ["adaptation"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Any CLI. Any agent. Ask here and the wiring will be delivered:
+        Keeper plugin (or its hook equivalent), memory MCP config, and the
+        measured bounds for your model's window. To deliver fast, we need
+        the harness facts below — guesses cost both of us a round-trip.
+
+  - type: input
+    id: harness
+    attributes:
+      label: CLI / agent harness and version
+      description: e.g. "aider 0.86", "Claude Code 2.x", "our in-house runner (link if public)".
+    validations:
+      required: true
+
+  - type: dropdown
+    id: mcp
+    attributes:
+      label: Does it speak MCP over stdio?
+      options:
+        - "Yes"
+        - "No"
+        - "I do not know"
+    validations:
+      required: true
+
+  - type: dropdown
+    id: hooks
+    attributes:
+      label: Does it expose a message-transform or plugin hook (the Keeper surface)?
+      description: Anything that can rewrite the outgoing message list before a model call. Name it below if you know it.
+      options:
+        - "Yes"
+        - "No"
+        - "I do not know"
+    validations:
+      required: true
+
+  - type: textarea
+    id: hook-detail
+    attributes:
+      label: Hook / plugin details (optional)
+      description: Docs link, hook name, or the closest extension point you found.
+    validations:
+      required: false
+
+  - type: input
+    id: model
+    attributes:
+      label: Model + context window + input cap (if known)
+      description: The Keeper trigger must sit below the model's input cap; 320k/128k defaults assume a big window.
+    validations:
+      required: false
+
+  - type: dropdown
+    id: os
+    attributes:
+      label: Operating system
+      options:
+        - "Windows"
+        - "Linux"
+        - "macOS"
+        - "Other"
+    validations:
+      required: true
+
+  - type: textarea
+    id: corpus
+    attributes:
+      label: What will it remember?
+      description: Rough corpus shape (docs/code/chat logs), size, and where the files live. This decides defaults, not eligibility.
+    validations:
+      required: false

rtr_role_type_retrieval-0.4.0/.github/ISSUE_TEMPLATE/bug_report.yml

@@ -0,0 +1,85 @@
+name: Something is broken
+description: A reproducible defect — wrong output, a crash, a gate that passes when it should not.
+title: "bug: "
+labels: ["bug"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Reproducibility is the whole game. A report we can run is worth ten we
+        can only read.
+
+        If you are reporting that a **check passes while the thing it checks is
+        broken**, that is the highest-value defect class we have — say so
+        explicitly in the summary.
+
+  - type: input
+    id: repo
+    attributes:
+      label: Which repository, and which version
+      placeholder: "rtr-role-type-retrieval @ 0.3.0, or commit 673266b"
+    validations:
+      required: true
+
+  - type: textarea
+    id: repro
+    attributes:
+      label: The exact commands, and their exit codes
+      description: >
+        Read exit codes DIRECTLY — `cmd; echo "exit=$?"`. Never through a pipe:
+        `cmd | grep …` reports grep's status, not cmd's, and that has produced a
+        false green in this organisation's own history.
+      render: shell
+    validations:
+      required: true
+
+  - type: textarea
+    id: expected
+    attributes:
+      label: What you expected, and what happened instead
+    validations:
+      required: true
+
+  - type: dropdown
+    id: class
+    attributes:
+      label: What kind of defect is this?
+      options:
+        - Wrong output or wrong behaviour
+        - A crash, hang, or a command that never returns
+        - A check passes while the thing it checks is broken (vacuous gate)
+        - A check fails while the thing it checks is fine (false alarm)
+        - An installer or release artefact does not match its sources
+        - Something else
+    validations:
+      required: true
+
+  - type: input
+    id: env
+    attributes:
+      label: Platform
+      placeholder: "Windows 11 / Git Bash 2.55, or Ubuntu 24.04 / bash 5.2"
+    validations:
+      required: true
+
+  - type: textarea
+    id: unverified
+    attributes:
+      label: What you could NOT check
+      description: >
+        An honest gap is a result. "I only have Windows" is a fine answer and a
+        far better one than silence.
+    validations:
+      required: false
+
+  - type: checkboxes
+    id: hygiene
+    attributes:
+      label: Before submitting
+      options:
+        - label: I reproduced this on a clean checkout, not only in my working tree.
+          required: false
+        - label: I read the exit code directly rather than through a pipe.
+          required: false
+        - label: This is not a security issue (those go through the private advisory form).
+          required: true

rtr_role_type_retrieval-0.4.0/.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1,14 @@
+blank_issues_enabled: true
+contact_links:
+  - name: Security vulnerability (private)
+    url: https://github.com/Nova-Violet-Role/.github/blob/main/SECURITY.md
+    about: Gate evasion, record forgery, fence escape, credential exposure. Never open a public issue for these -- use the affected repository's private advisory form.
+  - name: Ask a question
+    url: https://github.com/Nova-Violet-Role/.github/blob/main/SUPPORT.md
+    about: Questions are never noise here. If our documentation had answered it, you would not be asking -- so an unanswerable question is usually our defect.
+  - name: Code of Conduct
+    url: https://github.com/Nova-Violet-Role/.github/blob/main/CODE_OF_CONDUCT.md
+    about: Attacking the claims is the work. Attacking people is not.
+  - name: How to contribute
+    url: https://github.com/Nova-Violet-Role/.github/blob/main/CONTRIBUTING.md
+    about: Read the exit code directly. Watch your test fail. Say what you could not verify.

rtr_role_type_retrieval-0.4.0/.github/ISSUE_TEMPLATE/false_claim.yml

@@ -0,0 +1,63 @@
+name: A claim in our docs is false
+description: Something a README, CHANGELOG or release note asserts does not hold. The most welcome report there is.
+title: "false claim: "
+labels: ["false-claim"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        This organisation's documentation makes **measurable** claims on purpose —
+        theorem counts, timings, exit codes, "this gate has a negative control".
+        Any of them can be wrong.
+
+        A report that falsifies one is treated as a defect in the documentation,
+        not as an inconvenience and not as an attack. It is the fastest way to
+        improve any of these projects, and it is credited in the changelog.
+
+  - type: input
+    id: where
+    attributes:
+      label: Where the claim is
+      description: Repository, file, and line if you have it.
+      placeholder: "rtr-role-type-retrieval — eval/TAXONOMY.md, rtr-full leads combined"
+    validations:
+      required: true
+
+  - type: textarea
+    id: claim
+    attributes:
+      label: The claim, quoted exactly
+    validations:
+      required: true
+
+  - type: textarea
+    id: contradiction
+    attributes:
+      label: The measurement that contradicts it
+      description: >
+        A command and its real output. If the claim is a number, show how you
+        counted. Read exit codes directly — never through a pipe.
+      render: shell
+    validations:
+      required: true
+
+  - type: dropdown
+    id: kind
+    attributes:
+      label: What kind of wrong is it?
+      options:
+        - The claim is simply false
+        - The claim is true but weaker than it sounds (overclaim)
+        - The claim was true and has expired (a snapshot presented as a standing property)
+        - The claim cannot be checked at all as written
+        - The claim is true but the thing it measures is vacuous
+    validations:
+      required: true
+
+  - type: textarea
+    id: environment
+    attributes:
+      label: Where you measured it
+      description: OS, tool version or commit hash. A claim can be true on one platform and false on another — that is still a defect, and knowing which one saves a round trip.
+    validations:
+      required: false

rtr_role_type_retrieval-0.4.0/.github/ISSUE_TEMPLATE/proposal.yml

@@ -0,0 +1,58 @@
+name: Proposal or design argument
+description: A feature, a change of shape, or an argument that something should be done differently.
+title: "proposal: "
+labels: ["proposal"]
+body:
+  - type: markdown
+    attributes:
+      value: |
+        Proposals are welcome and are argued on merit. The bar is not
+        enthusiasm — it is a clear statement of the problem and an honest
+        account of the cost.
+
+        If this is an early idea rather than a worked argument, an **Ideas
+        discussion** is the better home and nobody will mind if you start there.
+
+  - type: textarea
+    id: problem
+    attributes:
+      label: The problem, stated without reference to your solution
+      description: If the problem cannot be stated on its own, the proposal is a solution looking for one.
+    validations:
+      required: true
+
+  - type: textarea
+    id: proposal
+    attributes:
+      label: What you propose
+    validations:
+      required: true
+
+  - type: textarea
+    id: cost
+    attributes:
+      label: What it costs
+      description: >
+        Complexity, new dependencies, maintenance burden, breaking changes,
+        things that become harder. A proposal with no stated cost reads as
+        unexamined.
+    validations:
+      required: true
+
+  - type: textarea
+    id: alternatives
+    attributes:
+      label: What you considered and rejected, and why
+    validations:
+      required: false
+
+  - type: dropdown
+    id: breaking
+    attributes:
+      label: Would this break existing users?
+      options:
+        - "No"
+        - "Yes, and I describe the migration below"
+        - "I do not know"
+    validations:
+      required: true

rtr_role_type_retrieval-0.4.0/.github/PULL_REQUEST_TEMPLATE.md

@@ -0,0 +1,41 @@
+<!-- Delete any section that genuinely does not apply, and say why it does not.
+     A deleted section with no explanation reads as an unanswered question. -->
+
+## What this changes, and what question it answers
+
+<!-- One paragraph. State the adversarial question your change answers —
+     "can a check pass while the thing it checks is broken?" — not just the
+     files you touched. -->
+
+## The evidence
+
+```
+<!-- The command you ran, its tail, and its exit code. -->
+```
+
+<!-- Read the exit code DIRECTLY, never through a pipe. `cmd | grep …` reports
+     grep's status, not cmd's. When you must pipe, use ${PIPESTATUS[0]}. -->
+
+## Checklist
+
+- [ ] The repository's test or gate suite exits **0**, and I read the exit code
+      directly rather than through a pipe.
+- [ ] Any behavioural change ships with a test **I have watched fail** — I broke
+      the code on purpose, saw the assertion fire, restored it, saw it pass.
+- [ ] If my check has a negative control, it asserts the mutation is genuinely
+      **present** before running, so "did not apply" cannot be misread as
+      "survived".
+- [ ] No carriage returns: `git ls-files -z | xargs -0 grep -lI $'\r'` prints
+      nothing. (When in doubt, count bytes: `tr -dc '\r' < FILE | wc -c`.)
+- [ ] New files carry the repository's `SPDX-License-Identifier` header.
+- [ ] Executable scripts are executable **in the git index**:
+      `git ls-files -s path/ | grep -v '^100755'` prints nothing.
+- [ ] I did not edit a check's expected value merely to make it pass. Any
+      changed expectation is justified from first principles in this PR.
+- [ ] Documentation numbers I touched were re-measured, not carried over.
+
+## What I could not verify
+
+<!-- Say it plainly. "I have no macOS machine; CI covers it" is a fine answer
+     and a far better one than silence. An honest gap is a result; a fabricated
+     green is a conduct breach. -->

rtr_role_type_retrieval-0.4.0/.github/workflows/template-lint.yml

@@ -0,0 +1,41 @@
+name: template lint
+on:
+  push:
+    paths:
+      - '.github/DISCUSSION_TEMPLATE/**'
+      - '.github/ISSUE_TEMPLATE/**'
+      - '.github/workflows/template-lint.yml'
+  pull_request:
+  workflow_dispatch:
+
+jobs:
+  title-not-empty:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      # GitHub rejects a form whose `title:` is present but empty, with
+      # "title must be of type String and cannot be empty". The key is
+      # OPTIONAL -- absent is fine; present-and-empty is not. A YAML parser
+      # does NOT catch this: `title: ""` is valid YAML and invalid schema.
+      - name: title must be absent or a non-empty string
+        run: |
+          rc=0
+          for f in .github/DISCUSSION_TEMPLATE/*.yml .github/ISSUE_TEMPLATE/*.yml; do
+            [ -f "$f" ] || continue
+            line=$(grep -n '^title:' "$f" | head -1) || true
+            if [ -z "$line" ]; then echo "ok   $f (no title: key)"; continue; fi
+            val=$(printf '%s' "$line" | sed 's/^[0-9]*:title:[[:space:]]*//; s/^"//; s/"$//; s/^ *//; s/ *$//')
+            if [ -z "$val" ]; then echo "FAIL $f -- title: present but EMPTY"; rc=1
+            else echo "ok   $f title=\"$val\""; fi
+          done
+          exit $rc
+
+      # NEGATIVE CONTROL: an alarm nobody has tripped on purpose is untested.
+      - name: the check must be able to FAIL
+        run: |
+          mkdir -p /tmp/ctl && printf 'title: ""\nbody: []\n' > /tmp/ctl/bad.yml
+          line=$(grep -n '^title:' /tmp/ctl/bad.yml | head -1)
+          val=$(printf '%s' "$line" | sed 's/^[0-9]*:title:[[:space:]]*//; s/^"//; s/"$//; s/^ *//; s/ *$//')
+          if [ -z "$val" ]; then echo "PASS: control rejected as designed"; else
+            echo "FAIL: the control was NOT rejected -- this lint proves nothing"; exit 1; fi

rtr_role_type_retrieval-0.4.0/.gitignore

@@ -0,0 +1,23 @@
+# SPDX-FileCopyrightText: Nova-Violet Role ORG
+# SPDX-License-Identifier: EUPL-1.2 OR AGPL-3.0-or-later
+.venv/
+__pycache__/
+*.pyc
+dist/
+build/
+*.egg-info/
+.local_vector_db/
+wiki_registry.json
+extra_roots.json
+lexikon_global.json
+.DS_Store
+# CocoIndex Code (ccc)
+/.cocoindex_code/
+# Lab-only noise (via negativa 2026-10-27): caches, logs, session state, vendored references
+.ruff_cache/
+bun-err.log
+.lake/
+state/
+.opencode/
+opencode-acp/
+rolling-context/