Skip to main content
Turn your demand report into launch copy you can post. launch drafts one asset per channel — Product Hunt, Show HN, an X thread — each in that channel’s voice, with every claim backed by a real quote. channel_plan gives you a step-by-step checklist to run by hand. Drafting only. metalworks never posts. launch(...) returns drafts; channel_plan(...) returns a checklist you execute yourself. Nothing here touches a network. 
from metalworks import Metalworks
mw = Metalworks()
research = mw.research("an affordable, jitter-free focus supplement for developers")
positioning = mw.positioning(research)

assets = mw.launch(research, positioning)   # list[LaunchAsset] — [] on a no-go report
for asset in assets:
    print(asset.surface, "—", asset.title)
    print(asset.body)
    for v in asset.variants:                # alternate hooks you can pick from
        print("  variant:", v)

plan = mw.channel_plan(research)            # a checklist you run by hand
for step in plan.steps:
    print(step.scheduled_offset, step.surface, "—", step.action)
From the CLI (report-id comes from metalworks research list): 
metalworks research launch <report-id>

​
What you give it / what you get back

You give it: a finished Research bundle (the report lives on .demand). Positioning is optional — pass it to keep the copy consistent with your positioning statement. You get back: a list[LaunchAsset], one per channel. Each asset carries a title (the hook), a channel-native body, alternate variants, and claim_citations — the claims it could back. Launch copy is the easiest place to over-claim, so every factual, quantified, or attitudinal claim is backed by a real quote. The model writes the body and lists each claim with the verbatim Reddit quote it says supports it; metalworks keeps a claim only when both the supporting quote resolves to a real quote in the report and the claim text appears verbatim in the body. Anything it can’t back is dropped — never softened, never guessed. 
asset = assets[0]
for c in asset.claim_citations:
    quote = asset.body[c.span_start:c.span_end]   # == c.claim_text
    print(quote, "→ backed by", c.evidence_ref.evidence_id)
The char offsets are Python code-point offsets. A non-Python consumer (JS counts UTF-16 code units) should treat claim_text as authoritative and re-find it rather than slicing by offset when the body contains emoji.

​
The channel plan

mw.channel_plan(research) returns a ChannelPlan: a fully deterministic sequence, no model involved. One step per channel, each marked requires_human and posting_gated — the library plans, you post. Show HN is never automated. Its step says to post it manually and answer replies yourself — the HN audience expects a human, and the plan encodes that. Holding a LaunchAsset or a ChannelPlan never posts anything.

​
When the result is thin

launch(...) refuses — returns [] — when the report says don’t launch: a negative demand verdict (thin signal / no demand), or no demand cluster with at least 2 distinct people behind it. It won’t manufacture a launch for an idea the report couldn’t validate. A strong-demand report with a thin price signal is still launch-worthy — the refusal reads only the demand-strength part of the verdict, so a “not enough price signal” caveat is never mistaken for a no-go. If a single channel’s model call fails, that channel is skipped; the rest still draft. This is honest, disclosed work only. metalworks won’t write fake reviews, invent personas, or post on your behalf — see the usage policy.
Next: a content & SEO plan from the same report, or the authentic Reddit engagement loop. Or read why you can trust the output — the rule that drops any claim it can’t back.