offer

offer: design it, price it, close it

Pricing is where founders bleed silently: too low out of fear, too clever out of theory, or changed weekly out of doubt. This skill runs the offer conversation like an operator: the founder's numbers first, tested mechanics second, receipts on everything.

  the offer question
        |
  [ ground ]     recall: their numbers, past pricing lessons, the vision
        |
  [ probe ]      wishful pricing dies here, before the market kills it
        |
  [ design ]     ladder -> price -> close -> launch, heuristics applied
        |
  [ receipts ]   their fact + external comparable, or PARKED

Before any step, read references/CONVENTIONS-core.md. This resolves from the installed skill directory to the plugin's shared contract, independent of the founder's working folder. Its receipts, filing, decision, and state rules govern this workflow.

Step 0: Ground in their reality

Run the shared contract's Universal preflight. A Fresh, Partial, or Wrong-folder result stops and runs the co-founder-setup skill before anything else (re-sync for Partial), without designing or filing an offer. On Ready, recall first: stage and margins from the charter, every pricing-related learning and decision in the vault, what customers have actually paid before, the vision's direction. Cite by file. An offer designed against an imagined business fails in the real one.

Before sizing or creating anything, search index.md and every file in wiki/initiatives/ for the same destination and price move; titles alone are not a dedupe key. Record the matching path and current state. An exact move gets one initiative for its whole lifecycle.

This is a money surface: the always-ask law applies to everything here.

Done when: their real numbers and history are on the table, or their absence is stated.

Step 0.5: Size the move, and stop if it is BIG

Public price changes, repricing existing customers, and anything that materially moves revenue are BIG. A truly small reversible test may size SMALL only when its bounded audience, cost, rollback, and non-impact on existing customers are recorded. A BIG move found here follows the dedupe result before any stub is created:

• Exact idea or survived match: reuse that file. An idea returns to the gauntlet; a survived file proceeds only when its latest verdict is SURVIVES with no parked recommendation.
• Exact planned or executing match: amend that initiative and preserve its state. If the approved Key decisions already contain this exact price move, offer may design the remaining terms against it. If the ask changes the approved move, route the change through plan's same-destination amendment checkpoint first. Never create or re-run the gauntlet on a parallel initiative.
• Exact reviewed or banked match: the arc is closed; a renewed move creates one new idea with reopens: pointing to the closed file.
• No exact match: create one new idea stub, append ## [date] offer | <initiative title> to root log.md, add the same event to the initiative's local Log, and hand it to the gauntlet. Durable state creation always has a root event even when offer design stops at the gate.

No designing or recommending happens while a required gauntlet or plan amendment is pending. The stub-creation event above is the invocation's one offer row. A genuinely small, reversible experiment proceeds only when sizing records it as small; a BIG label has no experiment-shaped bypass. Discovering mid-design that the move needed interrogation is discovering it too late.

Done when: the move is sized, and BIG moves are with the gauntlet instead of here.

Step 1: Probe the wishful thinking

Inventory every parameter the design would need: delivery input, capacity, cap, spot count, guarantee window, buyer behavior, and launch audience. Run each through the shared parameter gate before applying a heuristic. A missing value gets one question or a parked/unverified label and is excluded from current terms and customer-facing close.

Ask before you assert: never present a quantity, duration, sample size, date, or causal motive in prose before asking the founder for it or citing a receipt for it.

Before designing anything, the pricing-specific gap probes, each only where a real gap shows:

• What has anyone actually paid? The strongest pricing fact is a payment that already happened. Interest, compliments, and survey answers are not payments.
• What is the price of the current alternative? What the customer spends today (money or hours) bounds what the offer can charge.
• Does the math survive their real volume? Revenue fantasies use fantasy volume. Run the unit economics at the volume the vault supports.
• Can they operationally honor what the price implies? A cap, a guarantee, a service-heavy tier: each is a promise with an operating cost.

Done when: the design brief rests on facts, with the wishes named as wishes.

Step 2: Design with the heuristics

STOP. Read references/offer-heuristics.md now: it holds the distilled operator mechanics for ladders, caps, grandfathering, anchors, close sequence, guarantees, annual-versus- monthly, urgency legality, and launch timing, each with its trap. Advising on an offer without them is recalling generic frameworks, which is the thing this skill exists to replace.

Apply them to THIS business, one decision at a time, recommended answer attached: the ladder (what is free, what is gated, why), the price and its cap logic, the close (anchor, sequence, guarantee window), the launch (timing signal, warm slice). Where a heuristic and the founder's vault evidence disagree, the vault wins and the disagreement gets said.

Done when: each element of the offer has a recommendation with its mechanism and its trap named.

Step 3: Receipts before the recommendation ships

Every element of the final recommendation carries both legs: at least one fact about THEIR business (a vault file: past payment, margin, churn, audience number) and at least one checked external comparable (run research; category pricing, a real benchmark). A leg missing moves that element to Parked recommendations in the shared canonical shape and names recall or research as the next action. "Needs your call" is only for a judgment fork after the evidence is present. No confident-sounding heuristic bypasses this: the heuristics choose the QUESTIONS, the receipts justify the answers.

Done when: the founder can click through to why, on every number.

Run the shared durable claim admission gate on every proposed cap, duration, capacity, price, buyer behavior, and urgency statement. Reopen each cited source and quote the exact supporting passage; recompute capacities from recorded inputs. Anything supplied only by the founder is Founder-stated: unverified. A heuristic supplies a question, never an invented input.

Step 4: Bank it

The canonical home is always wiki/business/offer.md. Update it through the shared dedupe rule with Audience · Promise · Ladder and price · Terms · Receipts · Parked recommendations · History. Its base-five frontmatter carries offer_state: draft while any recommendation is parked, approved after explicit founder confirmation with every receipt present, active only after the founder confirms it is live, and retired when a successor takes over. Preserve superseded terms in History with dates; one file carries current truth.

Every price or term row is a T1 consent surface (shared contract, Immutable consent receipts): each carries [receipt: raw/inbox/<file>#<exact words>] pointing at the founder's actual words, captured into an immutable raw/inbox/session-receipt-<date>-<slug>.md when the yes is given. Offer may not move offer_state to approved or write a priced/term row as current truth without that receipt. Absent the founder's yes, the price stays a parked recommendation and offer_state stays draft; a system-generated date is never a founder-stated one. Keep index.md and hub/business.md pointed at the canonical offer. Before close, run the linked-state transaction across the offer's current constraints, its customer-facing close, index summary, linked initiative decisions, and metrics. A known-wrong or parked cap cannot remain in active close copy.

Then route without duplicating the design:

• Execution work is needed (a launch, a tier build): append a pending queue.md entry from offer to plan whose source_paths contains wiki/business/offer.md. Plan owns the initiative and marks the entry done after the initiative reaches planned.
• An initiative already exists: add the offer path plus the relevant rationale to Key decisions. The initiative points to the canonical terms instead of copying them.
• The call passes the 3-test bar: write the short dated decision page for history and link it from offer.md. It is part of this offer event, not a second event. A wiki/decisions/** page is a T1 consent surface: it may be created only with the founder's [receipt: raw/inbox/<file>#<exact words>] for the exact call. Without that yes, write no decision page and leave the term parked.

Append exactly one row: ## [date] offer | <what was designed>.

Close in chat: the offer in three lines, the strongest receipt, the ONE next move.

Mandatory filing gate

1. Run scripts/graph-audit on every touched durable file, including any initiative stub created while gating the offer.
2. Fix every violation and rerun it. Offer state, initiative creation, root event row, and queue writes remain incomplete until it exits zero.

Done when: offer.md holds the current design, exactly one log row records the event, and any execution queue item has a durable queue ID.