Budget Allocator

Budget Allocator

You have a total campaign budget and a goal ("$50k for a Q3 fitness launch") and need a defensible plan for how to spend it: how many creators, at which tiers, at what per-deal cost, and who specifically could fill each slot. This skill benchmarks tier-level deal values from the Creatorland corpus, proposes a mix that fits the budget, and names candidate creators per slot. It is a planning skill for the whole budget — not a price check on a single deal (that's fair-price-brief / one-number-rate).

Read first: ${CLAUDE_PLUGIN_ROOT}/shared/conventions.md (tool schemas, credit prices, the nine conventions). This skill honors thrifty/thorough credit modes (${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).

Precision & filters for outreach (convention 14). When this list will feed an outreach step, default to precision over raw recall: search precision: "tight" with the hard-gate filters the brief supports (platform, niche, data_freshness_days, content_format, audience_country), and reserve broad plus heavy lookalike unioning for market sizing. Do not fan out dozens of lookalike calls to hit a volume target (each hop drifts from the seed), and trim the weak tail by each result row relative_fit (within-set fit, 1.0 = strongest) rather than padding to a round number. Canon: ${CLAUDE_PLUGIN_ROOT}/shared/conventions.md convention 14.

Inputs to collect

• Total budget (required) — in USD; convert and note if quoted otherwise.
• Vertical (required) — infer from the campaign goal if stated ("fitness launch" → Fitness); ask only if genuinely ambiguous.
• Campaign goal / shape (required) — reach-led, conversion-led, or awareness/credibility-led. This drives the default mix bias: reach → more micro/mid breadth; credibility → fewer, bigger names.
• Deal type (optional) — map "sponsored posts" / "affiliate program" to the closest deal_type. Omit if unspecified; tier bands come from the vertical level.
• Constraints (optional) — geo, platform, must-include tiers, max creators, reserved spend (e.g. "hold $10k for paid amplification" — allocate only the remainder and say so).
• Mode — default thorough; thrifty on request or stated budget concern.

Never ask twice; never ask for what the conversation already contains.

Flow

1. Benchmark tier economics — query_market_intelligence calls in rate mode to establish per-deal value bands for the vertical, each wrapped in Refusal Recovery (rate-mode floor: 10 brands / 50 deals):

   • { mode: "rate", vertical: <vertical>, deal_type: <if specified>, creator_tier: <the tier slot being priced — macro/mid/micro — one call per slot> }
   • In thorough mode, optionally one { mode: "market", vertical: <vertical> } call (floor: 5 brands / 25 deals) for deal-volume/deal-type-mix context that informs which deal types are actually active in this vertical.

   Pass creator_tier (emerging <1k / nano 1k-10k / micro 10k-100k / mid 100k-500k / macro 500k-1M / mega 1M+) for the tier slot you are pricing so each slot benchmarks against same-size creators; a slot whose tier is unknown or too thin for the privacy floor broadens (disclosed) to the vertical band, and you position within it per the constraint below.

   IMPORTANT honesty constraint: where a tier is unknown or too thin to clear the floor, the corpus returns vertical/filter-level bands, not literal per-follower-tier price lists. Those tier slots (macro/mid/ micro) are constructed by positioning within the returned band — macro slots planned near p75+, mid near median, micro near p25 — and the plan must label this method explicitly. If a refusal ladder broadened the slice, disclose the clearance level. 5 credits per call incl. each retry. Thrifty: 1 rate call, max 2 ladder rungs, skip the market-mode call.

2. Propose the mix (no tool calls) — solve the allocation: number of slots per tier × per-slot planning rate (from band positions) ≤ budget, biased by the campaign shape. Present 1 primary mix + 1 alternate (e.g. "breadth mix" vs "anchor mix"). Show the arithmetic.

3. Name candidates per slot — one search_creators per tier slot type (convention 9: tiers are distinct targets — never one merged search): { mode: "brief", brief: <campaign goal text>, filters: { niche: <niche>, min_followers: <tier floor>, max_followers: <tier ceiling>, platform: <if constrained>, country: <if constrained> }, limit: 2×slots-for-tier (cap 15), precision: "balanced" } 2 credits each. If a tier's exact niche comes back thin, apply the thin-niche honesty ladder (convention 8): disclose, broaden one labeled step (exact niche → adjacent → vertical), and tag picks exact/adjacent. Thrifty: search only the two largest-spend tiers, limit = slots+2.

4. Freshness Gate on named candidates — in thorough mode, get_creator_profile { identifier: { type: "creatorland_user_id", creatorland_user_id: <from search result> } } (or social_handle form) for each named candidate; classify fresh/aging/ stale; stale go to "re-verify before pitch", never the main slot list. 1 credit each. Thrifty: skip profiling; label candidates "unverified — search-result freshness only" and say so in the data note.

5. Pre-run estimate — fire the credit-modes estimate before step 3/4 if the planned total exceeds ~30 credits (typical thorough runs do).

Deliverable

A budget allocation plan in markdown:

# Budget allocation — $<total> · <vertical> · <campaign goal>

## Recommended mix
| Tier | Slots | Planning rate/deal | Subtotal |
|---|---|---|---|
| Macro (<follower range>) | N | $<~p75 of band> | $ |
| Mid (<range>) | N | $<~median> | $ |
| Micro (<range>) | N | $<~p25> | $ |
| **Total** | | | **$<total≤budget>** (<headroom> headroom) |

Method note: planning rates are positions within the corpus rate band for
<vertical><, deal type> (p25/median/p75 = $x/$y/$z) — the corpus benchmarks
the vertical, not follower tiers directly; tier rates are band positions,
labeled as such.

## Alternate mix — <name>
<same table, one-line rationale for when to prefer it>

## Candidate creators per slot
### Macro slots
- @<handle> — <one-line fit rationale> · <fresh/aging> · [exact fit/adjacent fit]
...per tier...

### Re-verify before pitch
<stale-profile candidates, or "none">

---
**Benchmark basis:** <clearance level + floor-disclosure note if broadened>
**Provenance:** <provenance line(s) from market-intel responses, verbatim>
· recency window: <window>
**Data freshness:** N/M candidates synced within the last sync window; K
flagged for re-verification. <thrifty: "candidates unprofiled — verify
before client use">
**Honesty note:** rates are corpus-level market bands for this vertical/
tier-position — not any named creator's personal rate. Confirm actual quotes
per creator before committing budget lines.
Credits used this run: ~N (breakdown: <rate calls>×5 + <searches>×2 + <profiles>×1)

Honesty rules

• Band positions, not tier price lists. The corpus doesn't return per-tier rates; the plan constructs tier planning rates as positions within the vertical band and must say so. Never present a slot rate as "what micro creators in fitness charge."
• Never a named creator's rate. Candidates are listed with the slot's planning rate, framed as the budget line — not their quote. The plan tells the user to get actual quotes.
• Disclose every broaden (Refusal Recovery clearance level) and every thin-niche widening step (convention 8 labels).
• The arithmetic must close. Subtotals ≤ budget, headroom shown; if the goal can't be met inside the budget at band rates, say so rather than shading rates down.
• No contact info in the plan, regardless of source (convention 7).

Credit footprint

thorough: ~25–40 credits (1–2 market-intel ×5 + 3 tier searches ×2 + ~15–25 profiles ×1; refusal rungs add 5 each — pre-run estimate fires) · thrifty: ~9–14 credits (1 rate call + 2 searches, no profiling)