Acquired

This graph starts from an approved company topic, gathers source-backed research from the web, produces a three-act podcast script, and generates separate Audio Director manifests for the host and guest.

The contract path uses fixed material. The live path requires working Codex credentials and network access to the domains allowed by the research Rules.

Complete graph definition

The runnable configuration contains six parallel research branches, an evidence gate, three parallel writing branches, and two audio-manifest outputs. You can also download the raw YAML.

version: 1
project:
  name: acquired-podcast
  default_targets: [audio_manifest_host, audio_manifest_guest]
slots:
  topic_prompt: { kind: file, path: fixtures/topic.md }
  gate_request: { kind: file, path: fixtures/gate-request.json }
  topic: { kind: file, path: topic.md }
  origins: { kind: file, path: research/origins.json }
  business_model: { kind: file, path: research/business-model.json }
  product: { kind: file, path: research/product.json }
  competition: { kind: file, path: research/competition.json }
  leadership: { kind: file, path: research/leadership.json }
  inflections: { kind: file, path: research/inflections.json }
  research_gate: { kind: file, path: research-gate.json }
  supplement: { kind: file, path: research/supplement.json }
  meeting: { kind: file, path: production-meeting.md }
  act_one: { kind: file, path: script/act-one.md }
  act_two: { kind: file, path: script/act-two.md }
  act_three: { kind: file, path: script/act-three.md }
  draft_script: { kind: file, path: episode-draft.md }
  script: { kind: file, path: episode.md }
  qa: { kind: file, path: fixtures/qa.bootstrap.json }
  qa_result: { kind: file, path: qa.json }
  audio_manifest_host: { kind: file, path: audio-director-host.json }
  audio_manifest_guest: { kind: file, path: audio-director-guest.json }
rules:
  - id: choose_topic
    in: [topic_prompt]
    out: [topic]
    run: { using: human, command: "Approve the exact company topic file" }
  - id: research_origins
    in: [topic]
    out: [origins]
    run:
      using: codex
      model: gpt-5.6-luna
      reasoning_effort: none
      command: >-
        Research the origins and founding history of the company named in
        in/topic using live web sources. Write only out/origins as JSON with
        dimension, summary, claims, and sources. sources must contain at least
        two objects with url, title, retrieved_at, and evidence fields. Every
        material claim must be supported by those sources; do not use prior
        knowledge as unrecorded evidence.
  - id: research_business_model
    in: [topic]
    out: [business_model]
    run:
      using: codex
      model: gpt-5.6-luna
      reasoning_effort: none
      command: >-
        Research how the company in in/topic makes money, allocates capital,
        and reports its operating segments using live web sources. Write only
        out/business_model as JSON with dimension, summary, claims, and sources.
        sources must contain at least two objects with url, title, retrieved_at,
        and evidence fields. Bind every claim to recorded evidence.
  - id: research_product
    in: [topic]
    out: [product]
    run:
      using: codex
      model: gpt-5.6-luna
      reasoning_effort: none
      command: >-
        Research the company's principal products, services, and customer value
        from live web sources. Write only out/product as JSON with dimension,
        summary, claims, and sources. sources must contain at least two objects
        with url, title, retrieved_at, and evidence fields. Do not emit an
        unsupported product list.
  - id: research_competition
    in: [topic]
    out: [competition]
    run:
      using: codex
      model: gpt-5.6-luna
      reasoning_effort: none
      command: >-
        Research the company's competitive landscape and defensibility using
        live web sources. Write only out/competition as JSON with dimension,
        summary, claims, and sources. sources must contain at least two objects
        with url, title, retrieved_at, and evidence fields. Separate sourced
        facts from analysis.
  - id: research_leadership
    in: [topic]
    out: [leadership]
    run:
      using: codex
      model: gpt-5.6-luna
      reasoning_effort: none
      command: >-
        Research the leadership history, governance, and succession of the
        company in in/topic using live web sources. Write only out/leadership as
        JSON with dimension, summary, claims, and sources. sources must contain
        at least two objects with url, title, retrieved_at, and evidence fields.
  - id: research_inflections
    in: [topic]
    out: [inflections]
    run:
      using: codex
      model: gpt-5.6-luna
      reasoning_effort: none
      command: >-
        Research the decisive strategic inflection points in the company's
        history using live web sources. Write only out/inflections as JSON with
        dimension, summary, claims, and sources. sources must contain at least
        two objects with url, title, retrieved_at, and evidence fields, and each
        inflection must include a date and sourced consequence.
  - id: supplement_competition
    in: [topic, competition, gate_request]
    out: [supplement]
    run:
      using: codex
      model: gpt-5.6-luna
      reasoning_effort: none
      command: >-
        Read in/gate_request and in/competition, perform the requested targeted
        follow-up research about in/topic, and write only out/supplement as JSON
        with status equal to complete, dimension, request, findings, and sources.
        sources must contain at least two source objects with URLs and evidence.
  - id: review_research
    in: [origins, business_model, product, competition, leadership, inflections, supplement, gate_request]
    out: [research_gate]
    run: >-
      for artifact in in/origins in/business_model in/product in/competition in/leadership in/inflections;
      do jq -e '(.sources | length) >= 2 and (.summary | length) > 0' "$artifact" >/dev/null || exit 1; done;
      jq -e '.status == "complete" and (.sources | length) >= 2' in/supplement >/dev/null;
      jq -n --arg request "$(jq -r '.request' in/gate_request)"
      '{decision:"accept",round:2,request:$request}' > out/research_gate
  - id: production_meeting
    in: [topic, origins, business_model, product, competition, leadership, inflections, supplement, research_gate]
    out: [meeting]
    when: "test \"$(jq -r '.decision' research_gate)\" = accept"
    permissions: { network: none, secrets: [] }
    run:
      using: codex
      model: gpt-5.6-luna
      reasoning_effort: none
      command: >-
        Synthesize the accepted source-backed research in the declared inputs.
        Write only out/meeting as a three-act production brief with the central
        thesis, narrative tension, evidence assigned to each act, and source
        URLs beside the claims they support.
  - id: write_act_one
    in: [meeting, origins, business_model]
    out: [act_one]
    permissions: { network: none, secrets: [] }
    run:
      using: codex
      model: gpt-5.6-luna
      reasoning_effort: none
      command: >-
        Write only out/act_one from the supplied production brief, origins, and
        business-model evidence. Preserve source URLs inline and distinguish
        narration from host dialogue.
  - id: write_act_two
    in: [meeting, product, competition]
    out: [act_two]
    permissions: { network: none, secrets: [] }
    run:
      using: codex
      model: gpt-5.6-luna
      reasoning_effort: none
      command: >-
        Write only out/act_two from the supplied production brief, product, and
        competition evidence. Preserve source URLs inline and build the central
        strategic tension without inventing facts.
  - id: write_act_three
    in: [meeting, leadership, inflections]
    out: [act_three]
    permissions: { network: none, secrets: [] }
    run:
      using: codex
      model: gpt-5.6-luna
      reasoning_effort: none
      command: >-
        Write only out/act_three from the supplied production brief, leadership,
        and inflection evidence. Preserve source URLs inline and close the thesis
        with sourced consequences.
  - id: assemble_episode
    in: [act_one, act_two, act_three]
    out: [draft_script]
    run: "cat in/act_one in/act_two in/act_three > out/draft_script"
  - id: revise_episode
    in: [draft_script, qa]
    out: [script]
    when: "decision=$(jq -r '.decision' qa); test \"$decision\" = pending || test \"$decision\" = revise"
    run: "{ cat in/draft_script; printf '\nRevision applied: source transitions verified.\n'; } > out/script"
  - id: qa_episode
    in: [script, qa]
    out: [qa_result]
    run: >-
      if grep -Eq 'https?://' in/script && grep -q 'Revision applied' in/script;
      then printf '{"decision":"accept","round":2}\n';
      else printf '{"decision":"revise","round":1}\n'; fi > out/qa_result
  - id: direct_host_audio
    in: [script, qa_result]
    out: [audio_manifest_host]
    when: "test \"$(jq -r '.decision' qa_result)\" = accept"
    run: >-
      jq -n --arg script "$(cksum in/script | cut -d' ' -f1)"
      '{schema:"audio-director/v1",voice:"host",script_checksum:$script,segments:["act_one","act_two","act_three"]}'
      > out/audio_manifest_host
  - id: direct_guest_audio
    in: [script, qa_result]
    out: [audio_manifest_guest]
    when: "test \"$(jq -r '.decision' qa_result)\" = accept"
    run: >-
      jq -n --arg script "$(cksum in/script | cut -d' ' -f1)"
      '{schema:"audio-director/v1",voice:"guest",script_checksum:$script,segments:["act_one","act_two","act_three"]}'
      > out/audio_manifest_guest

What each Slot does

Slot	Stage	Role in the graph
topic_prompt	input	The company topic presented for approval.
gate_request	input	A bounded follow-up question required by the evidence gate.
topic	scope	The approved topic Revision read by every research Rule.
origins	research	Source-backed founding and early-history research.
business_model	research	Revenue, capital-allocation, and operating-segment research.
product	research	Product, service, and customer-value research.
competition	research	Competitive landscape and defensibility research.
leadership	research	Leadership, governance, and succession research.
inflections	research	Strategic inflections with dates and sourced consequences.
supplement	research	Targeted competition follow-up requested by the gate.
research_gate	acceptance	The decision written after validating the six reports and supplement.
meeting	production	A three-act production brief built from accepted evidence.
act_one	writing	Origins and business-model act.
act_two	writing	Product and competition act.
act_three	writing	Leadership and inflections act.
draft_script	join	Direct concatenation of the three acts.
script	revision	Episode script after source-transition revision.
qa	input	Bootstrap QA decision that allows pending or revise to trigger revision.
qa_result	acceptance	Machine check for source URLs and the revision marker.
audio_manifest_host	output	Audio Director manifest for the host voice.
audio_manifest_guest	output	Audio Director manifest for the guest voice.

What each Rule does

Rule	Reads	Writes	Work
choose_topic	topic_prompt	topic	Asks a person to approve the exact topic file.
research_origins	topic	origins	Researches founding history and records at least two sources.
research_business_model	topic	business_model	Researches revenue, capital allocation, and segments.
research_product	topic	product	Researches products, services, and customer value.
research_competition	topic	competition	Researches competition and separates sourced facts from analysis.
research_leadership	topic	leadership	Researches leadership history, governance, and succession.
research_inflections	topic	inflections	Researches strategic turning points with dates and consequences.
supplement_competition	topic, competition, gate_request	supplement	Performs one bounded follow-up investigation.
review_research	six reports, supplement, gate_request	research_gate	Uses jq to validate source counts, summaries, and supplement status.
production_meeting	topic, all research, research_gate	meeting	Creates a three-act brief after the gate accepts.
write_act_one	meeting, origins, business_model	act_one	Writes act one and preserves source URLs.
write_act_two	meeting, product, competition	act_two	Writes act two around the central strategic tension.
write_act_three	meeting, leadership, inflections	act_three	Writes act three with sourced consequences.
assemble_episode	act_one, act_two, act_three	draft_script	Concatenates the three acts in order.
revise_episode	draft_script, qa	script	Adds source-transition revision when QA is pending or revise.
qa_episode	script, qa	qa_result	Checks for source URLs and the revision marker.
direct_host_audio	script, qa_result	audio_manifest_host	Produces the host manifest after QA accepts.
direct_guest_audio	script, qa_result	audio_manifest_guest	Produces the guest manifest after QA accepts.

The six research Rules share only topic, so they can run in parallel. The three writing Rules share meeting and can also run in parallel. The two manifests read the same accepted script without depending on each other.

Run and inspect it

Start with the contract profile.

scripts/test-real-world-cases.sh acquired-podcast contract

It preserves the same Slots, Rules, gate, and JSON schemas while using fixed inputs to test scheduling and recovery.

Run live research after preparing network access and Codex credentials.

scripts/test-real-world-cases.sh acquired-podcast live

Inspect the six files under research/. Each should contain at least two sources with URLs, titles, retrieval times, and evidence. Then inspect research-gate.json, the three acts, and both Audio Director manifests.