RFQ to Quote

Chain 1. customer-inquiry-intake → data-consolidate → design step → quote-build.

This chain does not run in one sitting, and building it as though it does is the main way it goes wrong. Steps 1 to 5 happen when the RFQ lands. Then it pauses while the customer answers, often for days, which is usually the longest single stretch in quote turnaround. Treat the inquiry as resumable state.

Every member step runs through its member skill — load its SKILL.md before the step, and produce what it produces. Steps 1 to 5 are customer-inquiry-intake; step 6 is data-consolidate, backflush check included; steps 9 to 11 are quote-build with its ../quote-build/reference/hours-ladder.md, ending in its two delimited documents. A chain that prices from memory of its members drops exactly the rules that make the price defensible: the ladder, the default variance threshold, the outbound check.

Quick start

→ Steps 1-5 on arrival. Then STOP and wait for the customer
→ Step 3 can short-circuit: a hard disqualifier ends the sequence before we ask anything
→ Step 5 never waits. Provisional score on what is known
→ Answers re-enter at step 2, NOT step 5. The gaps register shrinks and step 5 re-scores
→ No answer? Stale. Never cancelled, never declined
→ Step 6 BEFORE step 7: a current-revision routing makes the design a cross-check, not the basis
→ Step 7 builds an internal model per path: C2 drawing, B prior model saved as new, A text (ROM only), C1 STEP (unverified)
→ Integrity check before and after every CAD step. The prior model is never overwritten
→ Fusion missing? Hand-entered features or no design step. Labelled, never a stop
→ Four gates: send questions, confirm bid/no-bid, confirm the design, release the quote

Inputs

First, find the plant setup: from the task's own folder (never cd into the plugin), run python3 -I <this skill's folder>/../../shared/scripts/plant_setup.py find, do its setup_page_first step, then read the profile at profile_path and any rate at rates_path, per ../../shared/replies.md.

The twelve steps

1. The RFQ arrives, in whatever format the customer sent it. Hand it to customer-inquiry-intake. A multi-attachment package is one inquiry.

2. Extract into the intake form, with missing and ambiguous flagged separately, every tolerance carrying how it was read, and each line's geometry_path classified — C2 drawing, C1 STEP, B prior part, A text only, none. This is also where a customer's later answer re-enters.

3. Check against the plant's constraints. A hard disqualifier stops the sequence here with a no-bid recommendation and its specific reason, before the plant engages the customer at all. There is no point asking five questions about a part the plant cannot make.

Two limits, both load-bearing. It stops the sequence, not the inquiry: the estimator can override and proceed, and the inquiry stays open. And only violations determinable from stated facts qualify. Anything resting on a missing field is a scoring input for step 5, not a stop.

4. Draft the open questions back to the customer, as sendable email text — including a drawing with no revision and any change the customer says is pending. Gate 1 — the estimator sends them. Nothing goes to a customer from here. When they say they have, record questions_sent_at and _by: the stale clock starts there.

5. Return the provisional bid/no-bid, with its reasoning and the specific unanswered questions that would change it, per line as well as overall: an RFQ whose lines deserve different answers is partial, with each line's line_recommendation. Gate 2 — the estimator confirms or declines, per line where the lines differ, and bid_decision_basis records what they ruled on.

Step 5 never waits. Most RFQs arrive with at least one gap, so a recommendation that blocked on answers would stall the chain on nearly all of them. It sits after the questions rather than at the front because a sound call often needs those questions already in flight, but it is still ahead of every costing step.

Then the chain stops. While any drafted question is unanswered, the first pass ends here, and says so in these terms: "Paused after step 5, at Gate 1 and Gate 2 — the questions are drafted for you to send, and the bid/no-bid is provisional until you confirm it. Steps 6 to 11 run when the customer answers, or when you tell me to quote with those questions as exclusions." No customer-facing quote — no price, no lead time, no terms — is drafted in the same pass as the questions it depends on, even when the request was "take this all the way to a quote." A quote built on an open question prices a guess, and once it exists it gets sent. The estimator may still choose to quote on exclusions; that is their call at Gate 2, made after seeing the questions, not a default.

A pause is saved, so a new session can resume it. Save the intake form on the Open work page as the contract names it, with open_questions word for word as sent, set paused_at_step and waiting_on, and write the inquiry's line to the page's ## Open inquiries block, per ../../shared/contracts/intake-form.md and ../../shared/replies.md. Name the page and the file on it. On resume, bring in the page, read the line, open the form from it, and continue at paused_at_step; never start the inquiry over.

6. Pull cost history, comparables, routings and operation actuals via data-consolidate and the ERP, with the backflush check — load ../data-consolidate/SKILL.md; actuals agreeing within 1% are probable_standard, all of them, not the one that happens to equal the routing. This runs before the design step on purpose: where a current-revision routing or trustworthy actuals exist, the model is a cross-check, not the basis.

7. The design step. Per line, by path, build the internal estimating model in a new Fusion document, every feature citing its source, with the document-integrity check before and after: C2 from the drawing; B by opening a different prior part's model, applying the drawing's changes and saving as a new document, the prior model never overwritten, falling back to C2 when it is unavailable; A from text specs, every dimension from_text, a ROM only, never a firm quote; C1 by importing the STEP and reading it, nothing generated, the import unverified. Emit the geometry estimate, approval.status: draft, per ../../shared/contracts/geometry-estimate.md. Skipped for a line with a current-revision routing unless the estimator asks for a cross-check. Read reference/design-step.md before the first CAD call.

8. Gate 3 — the estimator confirms the design. The model, the tolerance reading, the datum assumptions (every inferred datum, by name), the operation list, and the open questions the model raised. A line priced from a current routing still passes Gate 3 on the routing's operation list and the tolerance reading. No geometry-derived hour is computed for the quote until this yes, or an override the estimator gives and the record keeps.

9. The hours step, through quote-build — load ../quote-build/SKILL.md and ../quote-build/reference/hours-ladder.md first: the CAM path where Fusion Manufacturing and plant cutting data exist, the feature path always — setup hours and run minutes per operation, low, expected and high, each labelled with its basis, each toolpath verified by geometry.

10. The ladder per operation, through quote-build: routing, then recorded actuals, then geometry-derived, then first principles; variance where two exist; setup and run apart; capability flags; the load check by work center.

11. The customer quote and the internal sheet, as quote-build emits them: === CUSTOMER QUOTE — SENDABLE === and === INTERNAL COST SHEET — NOT FOR CUSTOMER ===, each closed with === END ===, the outbound check run on the first. Gate 4 — the estimator reviews and releases.

12. When something comes back. See Resuming, below.

Approval gates

Four, and the count matters — a reader who thinks there are three has lost the one that stops an unconfirmed model pricing a job, and a reader who thinks there are two has lost the one that stops mail reaching a customer.

Nothing reaches a customer without human release. The generated model never reaches a customer at all.

Asked where sign-off is needed, answer with this table — each gate, its step, and what the estimator does there. "All four gates are still with the estimator" is true and tells them nothing. And whenever the chain stops, it names the gate and the step it stopped at. What a yes at a gate does and does not cover — including that a Gate 3 yes does not carry to a later drawing revision — is ../../shared/approval-gates.md.

Resuming

When the answers come back

An answer is new source material. It re-enters at step 2, not step 5 — it is extracted like anything else with its own source reference, the gaps register shrinks, and step 5 re-scores on the fuller picture.

It revises the existing inquiry. It never opens a new one. When the inquiry came from the Open work page, the reply's first line says so: "Picking up EH-RFQ-0912 from your Open work page, paused at step 5." Match each answer to its question by the numbers in the saved open_questions.

Where an answer flips the recommendation, say so and say why. A recommendation that silently reverses between two readings is worse than either version of it. And an earlier Gate 2 yes lapses: it covered the recommendation as shown in bid_decision_basis, not the new one. Return the decision (and the changed line's) to draft, set bid_decision_reset to what changed, and ask Gate 2 again naming the change. Never quote a line on a yes given before it flipped.

When a revised drawing arrives

A new drawing revision re-runs step 7 for the lines it touches: a new model, a new geometry estimate, a new Gate 3. The previous Gate 3 yes covered the previous revision and nothing else. Hours and price follow from the new confirmation.

When the customer counters on price

A price counter does not re-run step 7 and never reopens CAD. Nothing about the part changed. quote-build re-derives from the stored operation record, and the revision goes through Gate 4 like any other.

When they never come back

After the interval (MFG_RFQ_STALE_DAYS, the profile, or 5 business days by default, per customer-inquiry-intake) the inquiry surfaces as stale, carrying its provisional recommendation and its still-open questions, so the estimator quotes it with exclusions or drops it deliberately.

A declined inquiry is never surfaced as stale: it has been decided.

Stale is a visibility change, never a status change. Nothing here cancels, declines, closes or no-bids an inquiry because a customer went quiet. Letting it go quiet is how an RFQ joins the unbid book, which is one of this chain's own metrics.

Output

Every document this skill produces is printed in the reply, in full. Saving a copy to a file as well is fine, as Word or Excel per ../../shared/replies.md, never .txt; a file instead of the reply is not. The person reads the reply, and every delimited block and outbound check lives in it — a document only on disk has been neither seen nor checked. A record handed to another skill is saved and summarised, never pasted as raw YAML; memory blocks are written where the plant's memory lives; and the name on an approval is asked, never taken from the account. All three per ../../shared/replies.md.

Everything the member skills produce, in order: the intake form, the open questions as sendable text, the provisional recommendation with what would change it, the geometry estimate per line (or the reason there is none), the customer-facing quote, the internal cost-basis sheet with its hours-basis table — rung, path, cutting-data basis, setup and run, range, variance and drivers per operation — and the assumptions and exclusions list.

The Sales command-center page — commercial performance for the period, carrying the intake record and the quote by reference. An artifact you place; nothing is written to a system. Read ../../shared/command-center.md.

Edge cases and failure modes

• A hard disqualifier fires at step 3 — the chain stops before step 4. The inquiry stays open and the estimator may override.
• The estimator declines at Gate 2 — quote-build blocks. Reopening is a new decision, not a quote.
• The estimator declines the model at Gate 3 — no geometry-derived hours. Correct the model and re-confirm, or price from the rungs that remain.
• A revised RFQ arrives mid-chain — it supersedes, and the chain resumes against the revision rather than running twice.
• The customer answers only some questions — re-score on what arrived, and keep the rest open. Partial is normal.
• The only routing is for an earlier revision — superseded. Step 7 runs as though there were no routing, and the old routing is shown, never priced.
• Foreign geometry appears in the Fusion document — the step fails. Ask the user to restart Fusion. Never build on top of it.
• Asked to update the prior part's model in place (path B) — decline in one line. The adapted model is always a new document.
• Asked to send a path A line as a firm quote — decline. A firm quote needs a drawing, and the drawing re-runs step 7.
• The model suggests a cheaper way to make the part — it goes in cost_down_proposals, on the internal sheet. Never to the customer.

Degradation

Chain position

This is Chain 1 itself. Members: customer-inquiry-intake, data-consolidate, quote-build, with the design step run here.

Metric. Quote turnaround, owned jointly by customer-inquiry-intake and quote-build. Hit rate and the unbid book, owned by quote-build.

Reference files