Waterfall Casting

Waterfall Casting

Tiered casting that heals itself. You have N roles to fill and a ranked, tiered shortlist — premium/mid/value or macro/mid/micro. Instead of reaching everyone at once and overspending, this skill reaches tier-1 only first, then watches the connection tools: when a tier-1 outreach comes back replied_declined or expires at Day-7 with no reply, it releases that slot's budget to tier-2 and reaches the next-best not-yet-contacted creator — cascading down the tiers until each role is filled or budget is exhausted. For brand-side managers and agency casting directors filling a fixed number of slots who want first-choice talent first and an automatic, honest fallback rather than a single all-in wave. It is iterative: it lays out the waterfall plan + budget up front, reaches the first wave, and on a later invocation (or when the user reports back) advances the waterfall a wave at a time.

Read first: ${CLAUDE_PLUGIN_ROOT}/shared/conventions.md (tool schemas, credit prices, the nine conventions) and ${CLAUDE_PLUGIN_ROOT}/shared/connection-flow.md (THE connection contract — the 3 tool schemas, the mandatory pre-flight, graceful degradation, privacy invariants). This skill honors thrifty/thorough credit modes (${CLAUDE_PLUGIN_ROOT}/shared/credit-modes.md), Refusal Recovery (${CLAUDE_PLUGIN_ROOT}/shared/refusal-recovery.md), and the Freshness Gate (${CLAUDE_PLUGIN_ROOT}/shared/freshness-gate.md).

This skill composes siblings by reference and never restates their internals:

• It sources the tiered shortlist from brief-to-shortlist (run that skill's casting flow to build the ranked, enriched, freshness-gated list, then tier it premium/mid/value). If the user already has an in-session tiered shortlist, use it as-is.
• It reuses the reach mechanics + mandatory pre-flight exactly as cast-and-connect applies them (entitlement probe, suppression pre-screen, budget_band auto-fill, 10×N estimate + confirm, one request_creator_connection per creator). This skill differs only in which creators it reaches when: tier-1 first, then cascading on decline/no-reply.
• It reads outreach state via the same connection-read tools connection-pipeline-tracker reads (list_connections / get_connection_status) to decide when to fall through. Where that skill renders the board, this skill acts on it.

If the user wants a single all-in wave (reach everyone now, no cascade), use cast-and-connect. If they want a read-only board of what's running, use connection-pipeline-tracker. If they only want the shortlist, use brief-to-shortlist. This skill is the one that backfills automatically.

Inputs to collect

• The brief OR an in-session tiered shortlist (required) — either a written brief (any format; passed straight into the brief-to-shortlist flow, then tiered) or a tiered shortlist already produced this session (use as-is; don't re-cast). Don't make the user restructure either.
• Roles to fill (headcount) (required) — how many slots the campaign needs filled, e.g. "3 creators". This is the target, not a reach count; the waterfall reaches one creator per open slot per wave, not the whole list.
• Budget (required) — total outreach/creator budget. Drives how far the cascade can run: each open slot draws from the pooled budget, and a released (declined/expired) slot's budget rolls down to the next tier. Express the plan against this ceiling.
• Tiering scheme (optional) — premium/mid/value (default) or macro/mid/micro. If the shortlist already carries tier labels, keep them.
• Opportunity fields for outreach — campaign_type is required (e.g. "paid IG Reel + Story", "UGC licensing", "event appearance"); ask if the brief doesn't state it. Optional: deliverables, comp_tags, timeline, brief_context, brand-authored personal_message. budget_band is auto-filled from a corpus rate band (see Pre-flight) — don't ask unless the user overrides. For a promo/gifting/event/collab (not a plain casting), also collect the structured brief — archetype (casting|product_gift|partner_promo|event_invite|paid_collab), an offer ({what, value, expiry, redemption}), and the on-behalf-of brand ({name, represented_by}) — per the schema and required-field rules in connection-flow.md (the new archetypes require brand + offer.what; don't restate the schema here, reference it).
• Credit mode — default thorough; switch to thrifty on the trigger phrases in the credit-modes module.

Never ask twice for anything the brief or an existing shortlist already states — quote it back instead. The waterfall rule itself is fixed and stated up front; don't ask the user to re-confirm it each wave.

Flow

This skill runs in waves. Wave 1 reaches tier-1. Each later invocation (or report-back) reads current outreach state and advances the cascade. The flow below covers both the first wave and an advancement wave — Step 2 detects which.

Step 0 — Build (or accept) the tiered shortlist.

• If given a brief: run the brief-to-shortlist casting flow exactly as that skill defines it (normalize → search → profile fan-out → Freshness Gate → optional conflict check → one slate-level rate call wrapped in Refusal Recovery → rank). Honor the same credit mode. Then tier the ranked output into premium/mid/value (or macro/mid/micro) — by follower band, rate band, or fit-rank as the brief implies. Do not restate or alter brief-to-shortlist's internals; it owns the casting contract.
• If given an in-session tiered shortlist: use it as-is.

The output of Step 0 is a tiered, ranked, freshness-gated shortlist plus the slate's inferred vertical and rate band (reused for budget_band below, so no extra rate call is needed).

Step 1 — Lay out the waterfall plan + budget (no reach yet, 0 outreach credits). Before reaching anyone, show the plan so the user understands the cascade and the spend ceiling:

• The tiers and who's in each (ranked within tier).
• The fixed fallback rule, stated honestly: "I reach tier-1 first. If a tier-1 creator declines or hits Day-7 with no reply, I release that slot and reach the next-best not-yet-contacted creator — tier-2, then tier-3 — until each role is filled or budget runs out. A no-reply or a decline triggers the fall-through; impatience does not."
• The budget math: headcount × budget ceiling, and how a released slot's budget rolls down to the next tier. Reaching is 10 credits/creator regardless of tier; the campaign budget (the dollars paid to talent) is what cascades, framed by the corpus rate band per tier.

Step 2 — Read current outreach state (free) and decide the wave. Call the connection-read tools exactly as connection-pipeline-tracker does — this is how the skill knows whether it's Wave 1 or an advancement, and which slots have fallen through:

list_connections { "limit": 100 }

For any outreach this campaign already started, detail it:

get_connection_status { "conn_ref": "<ref>" }

→ lifecycle status, touches sent (of 3), latest reply classification, next scheduled touch. Classify each prior outreach (per connection-pipeline-tracker's reading):

• Filled / advancing — replied_interested or accepted_inapp: the slot is working; do not backfill it. (The double-opt-in intro is the matchmaker's on a confirmed yes — never an address here.)
• Released → backfill — replied_declined, or expired, or delivered with no reply and the sequence already past its Day-7 end: this slot is open again; its budget rolls to the next tier and the next-best not-yet-contacted creator is the new reach target.
• Still live, leave alone — queued | sending | delivered mid-sequence (before Day-7), or replied_question (a reply is owed, not a decline): do not fall through yet. A question is not a no. Point the user to reply-triage for the question; hold the slot.

If list_connections returns the non-entitled refused envelope, jump to the graceful-degradation branch (below) — never error.

Step 3 — Compute this wave's reach set. For each open role (Wave 1: all roles; advancement: only the released ones), pick the next-best not-yet-contacted creator: stay within the current tier until it's exhausted, then drop to the next tier down. Skip anyone already reached for this brand (handled in the pre-screen too). Cap the wave at min(open roles, budget-allowed reaches) — if the remaining budget can't fund another tier-appropriate slot, stop and say so rather than reaching into talent the budget can't cover.

Step 4 — Mandatory pre-flight (before any request_creator_connection). Apply cast-and-connect's pre-flight verbatim, per connection-flow.md, in order:

1. Entitlement probe + graceful degradation. Already covered by Step 2's list_connections; if it returned a refused / not-entitled envelope, do NOT error — deliver the full tiered plan + budget + advice (the read-only half: the waterfall plan, the per-tier rate bands, the fallback rule) plus one line: "Connections aren't on your plan — here's the waterfall casting plan and budget, and you can reach these creators by upgrading to Creatorland Pro (connections)." Stop; the run still succeeded.
2. Suppression / active-sequence pre-screen. Drop from this wave's reach set (a) anyone already in an ACTIVE sequence for this brand (queued | sending | delivered | replied_* — a repeat is refused + a harassment risk) and (b) anyone who opted out. Report who was skipped and why. (The Step 2 read already has this data; match by the identifier/campaign each was reached under.)
3. Auto-fill budget_band from the corpus rate band. Reuse the rate band from Step 0. If Step 0 produced none (e.g. thrifty dropped it), make one query_market_intelligence rate call now, wrapped in Refusal Recovery:

   query_market_intelligence { "mode": "rate", "vertical": "<slate vertical>", "deal_type": "<only if the campaign specifies one>" }
   

   Express budget_band per tier as a market range (e.g. tier-1 "$5k–$12k (market p25–p75, Beauty macro tier)"), never a single number or a per-creator quote. Disclose any Refusal-Recovery broadening at the clearance level.
4. Credit estimate + explicit confirm. This wave reaches W creators = 10×W credits. State it and get an explicit yes before firing any request_creator_connection:

   This wave reaches creators (tier-) to fill open slot(s) — costs <10×W> credits ($<0.25×W>). The plan and budget are already yours; this starts a 3-touch sequence per creator that auto-stops on reply, and I'll backfill from the next tier if any decline or hit Day-7 with no reply. Proceed? Never fire a wave without confirmation. (Reuse credit-budget-guardian framing if the user wants a ledger.)

Step 5 — Reach this wave. One request_creator_connection per confirmed creator, all sharing the same opportunity (tier-specific budget_band):

request_creator_connection {
  "creator": { "type": "creatorland_user_id", "creatorland_user_id": "<id from shortlist>", "display_name": "<name>" },
  "opportunity": {
    "campaign_type": "<required, from inputs>",
    "deliverables": "<if stated>",
    "comp_tags": ["<if stated>"],
    "budget_band": "<auto-filled market band for THIS creator's tier, from step 4.3>",
    "timeline": "<if stated>",
    "brief_context": "<one-line slate context, no contact info>",
    "personal_message": "<brand-authored only — never invented; omit if the brand gave none>"
  },
  "send_connection_request": true
}

Use whichever creator identifier type the shortlist row carries (creatorland_user_id preferred; else social_handle with platform; email only if the brand supplied the address, with a required display_name). Capture each returned conn_ref + status. If a single creator comes back refused (already-active, not charged), note it and continue with the rest — never abort the wave.

Step 6 — Write the deliverable. The tiered plan + budget, this wave's Reached table (conn_refs), the fallback rule, and what's pending (live slots, released slots backfilled this wave, and what the next advancement wave would do). Point the user to report back (or re-invoke) to advance the cascade.

Advancing the waterfall: on the next invocation (or when the user reports "creator X passed"), re-run from Step 2 — read state, find released slots, compute the next wave's reach set (one tier down), pre-flight + confirm, reach, re-write the deliverable. The skill never auto-fires a new wave without a fresh 10×W estimate + confirm.

Deliverable

Client-ready markdown — the waterfall plan, the current wave's outreach record, and the standing fallback rule:

# Waterfall Casting — <campaign/brand name>
_Wave <k> · <date> · <R> roles to fill · budget ceiling <budget> · Creatorland Data_

## The waterfall plan
Reach tier-1 first; on a **decline** or **Day-7 no-reply**, release that slot's
budget and reach the next-best not-yet-contacted creator one tier down — until
each role is filled or budget runs out. A no-reply or decline triggers the
fall-through; impatience does not. We never promise a yes.

| Tier | Creators (ranked) | Tier rate band (market p25–p75) |
|---|---|---|
| Tier 1 (premium/macro) | <Name @handle>, <Name @handle> | <$x–$y, vertical/tier> |
| Tier 2 (mid) | <Name @handle>, … | <$x–$y> |
| Tier 3 (value/micro) | <Name @handle>, … | <$x–$y> |
> <rate provenance line(s) exactly as the tool returned them>
Bands are the **market band for the vertical/tier — not any individual creator's
rate** — used to frame budget, not to quote talent. <If broadened: clearance-level note.>

## Slot ledger
| Slot | Status | Reached (conn_ref) | Result so far | Next if it falls through |
|---|---|---|---|---|
| 1 | live (tier-1) | <conn_ref> | queued | → tier-2: <next-best Name> |
| 2 | live (tier-1) | <conn_ref> | delivered, 1/3 | → tier-2: <next-best Name> |
| 3 | backfilled (tier-2) | <conn_ref> | queued | → tier-3: <next-best Name> |
<one row per role; "result so far" from get_connection_status>

## Reached this wave
Outreach started via Creatorland's matchmaker — a 3-touch sequence (day 0 / 3 / 7)
that auto-stops on reply, with a double-opt-in intro only if the creator says yes.
No creator contact details are shared with you through this; the matchmaker holds them.

Opportunity sent: **<campaign_type>**<, deliverables, timeline if set>.

| # | Creator | Tier | conn_ref | Status |
|---|---|---|---|---|
| 1 | <Name> @<handle> | T1 | <conn_ref> | queued |
<one row per creator reached this wave; status as returned>

**Skipped in pre-screen:** <Name — already in an active sequence for this brand>;
<Name — opted out> · or "None — all picks were clear to reach (in our records)."

## Pending — what the next wave would do
- **Live, holding:** <slots still in-sequence before Day-7, or replied_question>
  — no backfill yet. <replied_question → "creator asked a question; reply owed,
  not a decline — draft it with **reply-triage**; the slot stays held.">
- **Released this wave → backfilled:** <slots that declined/expired and were
  reached one tier down>.
- **Next fall-through targets:** <per open slot, the next-best not-yet-contacted
  creator one tier down>, fundable within remaining budget <or "budget exhausted
  below tier-N — no further backfill without raising the ceiling">.
- Report back (or re-invoke) when a slot declines or lapses and I'll advance the
  waterfall — with a fresh 10×W estimate + confirm before reaching.

---
Data freshness: <N>/<M> shortlisted creators synced within the last sync window; <K> flagged for re-verification.
Provenance: Creatorland Data MCP · cast via brief-to-shortlist (search + <M> profiles + 1 rate benchmark, <clearance level>) · waterfall state via list_connections + get_connection_status (free) · <W> connection requests this wave · <date>.
Credits used this run: ~<T> (cast: ~<C> · state reads: 0 · reach this wave: <W> × 10 = <10W>).

If the entitlement probe returned not-entitled, the deliverable is the waterfall plan + tier rate bands + fallback rule only (the advice half), with the Reached / Slot-ledger outreach columns replaced by the single upgrade line — and a credit tally reflecting the cast only (no reach charges). On an advancement wave with no released slots, the deliverable is the slot ledger + "nothing to backfill — all slots live or filled" and a 0-reach-credit tally.

Honesty rules

• Never promise a reply or a yes. Frame every reach as "we'll reach them and run a 3-touch sequence that stops on reply" — outcomes are the creator's. A queued/delivered status is contact made, not a booking. "Backfill" replaces a released slot, not a rejected creator's worth.
• Fall-through triggers on decline or Day-7 no-reply — never impatience. A replied_question is a reply owed, not a no; a mid-sequence delivered (before Day-7) is still live. Hold those slots; do not cascade early. State this rule up front and apply it literally.
• Privacy floor (conventions 7 + connection-flow). No creator contact info appears anywhere — not in the plan, the slot ledger, or the Reached table. The reveal is the matchmaker's double-opt-in only. conn_ref and status are brand-safe; a creator's address is not, and is never returned or inferred.
• budget_band is a per-tier market band, never a per-creator quote — auto-filled from the corpus rate band, expressed p25–p75 for the vertical/tier. Disclose any Refusal-Recovery broadening at the clearance level.
• personal_message is brand-authored only. Never invent a quote, tone, or promise on the brand's behalf — omit it if the brand gave none.
• Suppression pre-screen is "in our records," never an absolute guarantee the creator isn't being contacted elsewhere.
• Always show the credit cost of a wave before reaching, and the actual tally after. Each wave gets its own 10×W estimate + explicit confirm — the skill never auto-fires the next wave. The plan + budget are delivered regardless of whether any outreach runs.
• Budget is honest about its ceiling. When the remaining budget can't fund a tier-appropriate backfill, say "budget exhausted below tier-N" and stop — never reach into talent the budget can't cover just to fill a slot.
• All brief-to-shortlist honesty rules (vertical-level rate framing, provenance never stripped, stale data never silently mixed) carry through the composed shortlist unchanged.

Credit footprint

Per wave: cast happens once (Wave 1 only); reaching is 10 × (creators reached this wave); all state reads (list_connections / get_connection_status) are free.

thorough: 31 cast credits on Wave 1 (brief-to-shortlist: 1 search ×2 + ~24 profiles + 1 rate call ×5; +5 per refusal-ladder rung) + 10 × (creators reached this wave). Filling 3 tier-1 slots in Wave 1 ≈ 31 + 30 = ~61 credits ($1.53); a later backfill wave reaching 2 ≈ 0 cast + 20 = ~20 credits. · thrifty: ~17 cast credits on Wave 1 (1 search ×2 + ~10 profiles + 1 rate call ×5, max 2 ladder rungs) + 10 × (creators reached this wave); reach only the open slots this wave, never the whole list. Not-entitled plans incur cast credits only (no reach charges); advancement waves with no released slots incur 0.