corpus-qa

1# SPDX-License-Identifier: Apache-2.02# yaml-language-server: $schema=https://nika.sh/spec/v1/workflow.schema.json3#4# TEMPLATE · corpus-qa · canonicalize corpus → retrieve → answer or abstain.5#6# Identical duplicates are idempotent, while one id carrying conflicting text7# is refused before indexing. An empty retrieval is a valid answer with empty8# citations, never a fabricated model response.9#10# Needs · nothing. Runs offline in an empty directory with mock/echo.11# Run · nika run <file> --model mock/echo12nika: corpus-qa13​14model: mock/echo                     # SLOT: replace with a local or hosted seat15​16const:17  query: "What is the holiday carry-over policy?"   # SLOT: the question18  documents:                         # SLOT: replace with retrieved source chunks19    - { id: handbook, text: "Support hours are 09:00 to 17:00 UTC." }20    - { id: security, text: "Security incidents are reported within one hour." }21    - { id: handbook, text: "Support hours are 09:00 to 17:00 UTC." }22​23⛨permits:24  ⛨tools: ["nika:jq", "nika:assert"]25​26tasks:27  conflicting_ids:28    ◆invoke:29      tool: nika:jq30      args:31        input: ${{ const.documents }}32        expression: 'sort_by(.id) | group_by(.id) | map(select((map(.text) | unique | length) > 1) | .[0].id)'33​34  duplicates_consistent:35    with:36      conflicting_ids: ${{ tasks.conflicting_ids.output }}37    ◆invoke:38      tool: nika:assert39      args:40        condition: ${{ size(with.conflicting_ids) == 0 }}41        message: "One corpus id names conflicting source text"42​43  index_once:44    after:45      duplicates_consistent: success46    ◆invoke:47      tool: nika:jq48      args:49        input: ${{ const.documents }}50        expression: "sort_by(.id) | unique_by(.id)"51​52  index_twice:53    with:54      index: ${{ tasks.index_once.output }}55    ◆invoke:56      tool: nika:jq57      args:58        input: ${{ with.index }}59        expression: "sort_by(.id) | unique_by(.id)"60​61  index_is_idempotent:62    with:63      first: ${{ tasks.index_once.output }}64      second: ${{ tasks.index_twice.output }}65    ◆invoke:66      tool: nika:jq67      args:68        input: { first: "${{ with.first }}", second: "${{ with.second }}" }69        expression: ".first == .second"70​71  compiled:72    with:73      idempotent: ${{ tasks.index_is_idempotent.output }}74    ◆invoke:75      tool: nika:assert76      args:77        condition: ${{ with.idempotent }}78        message: "Corpus index changed when canonicalized twice"79​80  retrieve:81    with:82      index: ${{ tasks.index_once.output }}83    ◆invoke:84      tool: nika:jq85      args:86        input: { documents: "${{ with.index }}", query: "${{ const.query }}" }87        expression: '.documents as $docs | .query as $query | ($docs | map(select((.text | ascii_downcase) | contains($query | ascii_downcase))))'88​89  answer:90    with:91      hits: ${{ tasks.retrieve.output }}92    when: ${{ size(with.hits) > 0 }}93    ◇infer:94      max_tokens: 500                # SLOT: answer spend ceiling95      prompt: |96        Answer from these passages only and cite their ids.97        Question: ${{ const.query }}98        Passages: ${{ with.hits }}99      schema:100        type: object101        additionalProperties: false102        required: [answer, citations]103        properties:104          answer: { type: string, minLength: 1 }105          citations: { type: array, minItems: 1, items: { type: string } }106​107  finalize:108    with:109      hits: ${{ tasks.retrieve.output }}110      answer: ${{ tasks.answer.output }}111    ◆invoke:112      tool: nika:jq113      args:114        input: { hits: "${{ with.hits }}", answer: "${{ with.answer }}" }115        expression: 'if (.hits | length) == 0 then {known: false, answer: "I do not know from this corpus.", citations: []} else {known: true, answer: .answer.answer, citations: .answer.citations} end'116​117outputs:118  result: ${{ tasks.finalize.output }}119​