You design the policy and the machine that enforces it for a single, dated, physically located event. The deliverable is two things, never one: a published document, and a pipeline of named people who can act on a report before the event ends.

A code of conduct with no pipeline behind it is the failure this skill exists to prevent. It signals the stated values are decorative, and it spends a reporter's trust on a channel that answers nothing.

The output is a policy, a staffing plan, and a runbook. Sanctions remove people from a paid event and reporting channels collect sensitive personal data about named individuals, so parts of this touch real legal exposure.

Say this before the interview. An organizer who mistakes a published policy for legal cover is worse off than one with no policy.

Ask one question at a time, multiple-choice where possible. Questions 5-7 exist because the menus below diverge sharply on time-to-effect, durability and effort - their default orderings cannot be picked for the user.

Harassment is the same conduct and needs the same enumerated list whoever bought the ticket. What genuinely changes every mechanic below is who controls the people in the room.

State which model a recommendation assumes whenever they differ. Most of the policy anatomy transfers to both; the response-team staffing menu, the appeals route and the public-statement authority do not.

Sponsor and booth staff sit on this seam either way. They are in scope of your policy - MLH, PyCon US and all name sponsors in their own scope clauses - but they are bound by a contract their employer signed, not by a ticket they bought. Say plainly in the policy that they are covered; the clause wording lives in .

If your harness has persistent memory, store the scope decisions, the response team roster with their conflict declarations, the published response clock, and every sanction with its rung and reason. The next edition starts from a ladder that already exists and a record that makes a repeat offence visible - which is the only thing that lets rung 2 mean anything.

Every ordering below is a default, not a law; it shifts with context and with who executes it. Before presenting any of the three menus, re-rank it against the interview and say which answer moved which option:

Where the user's constraints rule an option out, delete it from their plan rather than parking it at the bottom. A ruled-out option left on the list reappears as scope at T-2 weeks.

Efficiency (recommended order): adapt an established text > adopt one verbatim > adapt plus a culture preamble.

Adopting verbatim is near-free and still loses, because the template names roles and contacts that do not exist at your event - which makes its reporting sentence unusable - and disclaims its own enforceability. Adapting costs an afternoon and fixes exactly that. The on compliance cost is real rather than a dodge: both adapting postures keep the same Creative Commons attribution obligation and trigger the same single read by whoever holds legal authority; a culture preamble adds words, not legal surface.

Default: adapt an established text. Promote to the culture preamble whenever a stated identity or values document already exists.

A narrower variant of adopting verbatim: for a small volunteer-run event, linking directly to a peer event's own live, maintained code-of-conduct and values pages - rather than a generic template - inherits that peer's local legitimacy and enforcement record along with its text, at the same near-zero drafting cost. It is sound only when a suitable peer document exists in the same region or community and its scope (in-person versus online, jurisdiction) genuinely matches this event, and it still needs the same fix the generic template needs: confirm the linked policy is current and its host organization still exists before pointing to it, and replace any contact or role it names with this event's own.

Starved: adapt plus a culture preamble - highest value and highest effort, so efficiency starves it every round.

Deleted, not demoted: writing your own from scratch. Delete it unless the event has in-house counsel and a specific reason no template fits.

Efficiency: published email alias > physically identifiable on-site responders > named conflict-of-interest alternates > anonymous intake path > staffed phone or SMS hotline.

The alias leads on efficiency because it costs minutes, works before and after the event, and is the address the printed policy has to carry. Visible responders lead on value because a distressed person reaches for someone they can see: PyCon US puts its trained responders in "brightly colored shirts", a channel with no online equivalent. The on compliance cost holds because both visible responders and named alternates trigger the same single review - publishing an individual's name and contact, which is a consent conversation with that person, not a data-processing regime.

Compliance cost is a real axis here, unlike in most menus. Each channel carries a different exposure:

Default: published alias plus named conflict-of-interest alternates plus physically identifiable on-site responders, for any event with a physical day. Drop to alias plus alternates only for an online-only or pre-event-only scope.

Starved: the anonymous intake path - highest value against exactly the reports the other channels never receive, and high effort because it forces a decision convention that works with no reporter to go back to. Promote it whenever there is a power gradient inside the reporting population: paid staff, sponsor booth staff who answer to a sponsor, students at a university-hosted hackathon, or a response team that includes the event's most senior organizer. PyCon US publishes no anonymous channel while accepting anonymous reports at intake; CNCF runs a documented relay where any committee member forwards a report without the reporter's name (project practice, not an event's).

Efficiency: a named published team of 3-6 > routing into an umbrella body's incidents team > an external trust-and-safety contractor.

Sizing 3-6 is deliberate: it has to still function with a third of it unavailable, which at a three-day event is a certainty. Fix the decision method in advance including tie and absence rules, and let it err toward taking action - the dominant observed failure is inaction and delay on valid reports, not haste.

The umbrella route is near-zero effort where it exists and unavailable everywhere else, so treat it as conditional rather than as a rung anyone can pick. Its compliance cost is that the affiliation agreement makes someone else's procedure binding on you and removes your discretion to reverse the outcome.

Default: a named published team of 3-6, with every member's name in the reporting instructions and a recusal rule that starts the moment a member realizes it applies.

Starved: the external trust-and-safety contractor - highest value in the one case nothing else answers, highest effort (procurement, briefing, a retainer, and handing an outside party attendee personal data). Promote it when the event's most powerful person sits on the organizing team and no independent body exists above them. A code of conduct that will not apply to the most powerful people in the community should not be adopted at all; a contractor is the only rung that answers that precondition when there is no board to escalate to.

Deleted, not demoted: a single named organizer as sole responder. Delete it for any event where that organizer is also the most senior person on the team, because the channel then cannot receive the one report most likely to need it, and one person deciding alone at 2am is not a process. Left at the bottom of a list it reappears as "we'll just have someone handle it" two weeks out.

Measure the pipeline, not the policy - a policy has no observable output until a report arrives.

Pass threshold, and iterate until it is met: every report acknowledged inside the published window, and every reporter told the outcome before the published follow-up deadline. That threshold is self-set, and the two published bands above are what make it realistic rather than invented.

Expected output: