question.consulting · the services side of asking well · No engagement is open today

Asking a school something worth answering.

Most feedback programmes fail before the first response arrives, in the wording. This is the practice of getting that part right.

Write about a question you are stuck on See how it goes wrong →

question.consulting is a stated practice, not an open engagement. No school or district is being advised through this address today, no fee exists, and nothing on this page books, schedules, or signs anything.

What this is, and what it is not

Every other address in the Question family is about a tool. question.technology explains the response engine. question.software is one teacher’s front door. question.management is how a district stands the tool up across buildings. question.company is the venture behind them.

This address is about the part that happens before any of that: the wording, the audience, the window, the denominator, and the decision the answers are supposed to feed. A school that buys a survey tool and writes bad items now has confident numbers about nothing, faster.

It is a practice, described. Five stages, six named failure modes, and the schema constraints an anonymous instrument has to be designed inside — all set out below in enough detail that you can use them without contacting anyone. That is deliberate. A consulting page that withholds its method so you have to buy it is doing the thing this practice is against.

What it is not: not a software product, not a certification, not a legal or compliance opinion about your district’s obligations, and it does not administer instruments on your behalf. It is also not an open engagement today — see the honest status below, stated in the present tense rather than as a promise about a future date.

The five stages of a feedback programme

These are in order, and the order is the point: the most common failure is starting at stage two.

Stage 01

Decide what the answer will change

Before a single item is written, name the decision the results feed and the person who owns it. If no decision changes on any possible result, the instrument should not be fielded.

In practice This stage kills the most surveys, and it should. A climate survey with no owner produces a slide deck; a climate survey owned by the person who sets the duty roster produces a changed duty roster.

What exists at the end One page: the decision, the owner, the date the decision is actually made, and the result that would change it.

Stage 02

Write the items, then try to break them

Draft each item, then read it as an adversary: double-barrelled, leading, assuming a fact the respondent may not hold, or requiring a scale point that does not exist for them.

In practice “How satisfied are you with communication from the school and the district?” is two questions wearing one scale. A parent happy with one and furious with the other cannot answer it truthfully, so the number that comes back is not a number.

What exists at the end An item bank with the rejected drafts kept beside the survivors, so the next programme does not re-make the same item.

Stage 03

Design for who is answering

A staff instrument, a family instrument and a student instrument are three different designs. The student one is the case where anonymity has to be structural rather than promised.

In practice Reading level, item count and the honest limit of a free-text box all move by audience. A twelve-item staff form and a four-item student form can serve the same decision; a twelve-item student form mostly measures stamina.

What exists at the end One instrument per audience, each with its own item count and its own consent posture written down.

Stage 04

Field it where the guarantees are real

Anonymity that lives in a promise on the cover page is a different thing from anonymity the storage layer cannot violate. The constraints section below names the difference precisely.

In practice If a response row can be joined back to the link it came from, or ordered precisely by arrival, a small population can be re-identified regardless of what the cover page said.

What exists at the end A fielding plan naming the surface, the window, the population, and what is structurally impossible to recover afterwards.

Stage 05

Read it, in public, with the losses named

Report the response rate, the population it is a rate of, the items dropped and why, and the finding you did not like. A reading that carries only good news trains the next cohort not to answer.

In practice A 19% staff response rate reported as 19% is a usable, honest signal. The same rate reported as “staff say” is a claim with the scope stripped off, and staff notice.

What exists at the end A written reading with the denominator on its face, published to the people who answered it.

Six ways a school survey goes wrong

Each is a specific defect with a specific repair. None of them requires software to fix, and none of them is fixed by a better tool.

Wording

The double-barrelled item

What you see. One item asks about two things and the responses cluster oddly in the middle.

Why. A respondent who feels differently about the two halves has no truthful answer available, so they pick the midpoint.

The fix. Split it. Two items, two scales, and accept the longer instrument.

Scale

The missing scale point

What you see. A large share of respondents skip an otherwise simple item, or the free-text box fills with corrections.

Why. The scale has no point for the respondent's actual position, most often no honest “does not apply to me”.

The fix. Add the escape point and count it separately. A non-answer that is recorded is data; a non-answer forced into a neighbouring point is contamination.

Anonymity

Anonymity by promise

What you see. Response rates from the smallest buildings, or the smallest roles, are near zero.

Why. The people in a group of four know that a role field plus a building field identifies them, whatever the cover page says.

The fix. Cut the demographic fields to what the decision actually needs, and field it where re-identification is structurally prevented rather than promised.

Timing

Fielded at the wrong moment

What you see. Results swing violently between two administrations with no intervening change.

Why. The window sat beside an event -- a schedule change, a closure, a report-card date -- that dominates the responses.

The fix. Choose the window against the calendar first, and record which events sat inside it, so the next reading is comparable.

Follow-through

The survey with no reply

What you see. Each administration draws fewer responses than the last, from the same population.

Why. Nothing visible changed after the previous one, and nobody was told what was read.

The fix. Publish the reading and the decision it fed, including the part you did not act on and why. This is the cheapest single intervention on response rate, and it costs no software.

Scope

The stripped denominator

What you see. A headline says “families say”, a board member asks how many families, and the room goes quiet.

Why. A true statement about respondents was published as a statement about the population.

The fix. Put the denominator on the face of every figure. “112 of 640 families” survives the question; “families say” does not.

What the engine forces a design to do

If a programme is fielded on the Question engine, its schema constrains what can be designed — and the constraints are mostly the ones you would want. Each row names the rule in our own migration, so you can check it rather than take it on trust.

The rule, and where it lives	What it means for a design
Three answer shapes, enforced by the database `ck_qt_question_type CHECK (type IN ('text','scale','choice'))`	An instrument composes from free text, a scale, and a choice. That is a narrower palette than a general form builder, and a useful one: matrix grids and ranking items -- the two shapes that most often produce uninterpretable data -- are not available to design badly.
A response row carries no user identity `qt_submission has no user_id; the migration states it is designed so a leak is impossible, not merely absent`	Anonymity is a property of the storage, not a sentence on the cover page. This is what makes a small-building instrument worth fielding at all.
No arrival timestamps on a response or an answer `no created_at / updated_at on qt_submission or qt_answer, on purpose`	Responses cannot be put back in the order they arrived, so a respondent cannot be identified by having answered right after a meeting ended.
A share link is never joined back to a response `no share_link_id on qt_submission, on purpose`	A per-recipient link can still control distribution without the link becoming a name attached to the answer -- the usual way a confidential survey quietly stops being one.
Moderation runs before storage on the anonymous path `moderation is contractual on the anonymous path (qt_submission.moderation_state)`	An open free-text box on a student instrument is designable, because a flagged response is labelled for a human rather than published or silently dropped.

Scope bound, stated because it matters. These are properties of the question engine’s database schema. They are not a legal opinion, not a certification, and not a statement about your own district’s retention, logging or access controls, which sit outside this schema and outside this practice.

Who this is for, and what it costs to go without

The third column is the honest one: the cost of skipping this work is not a missing feature. It is a decision made on bad information, and a population that stops answering.

Reader	What they are asking	The cost of going without
A principal running a climate survey	How do I ask staff something real without the results being unusable or the smallest team being identifiable?	A survey nobody trusts, a response rate that falls every year, and a staff room that has learned the exercise is decorative.
A district assessment or operations lead	Four buildings are each running their own form. How do I get one instrument that compares across them?	Four incomparable datasets, a board presentation built on the largest one, and a decision made on the building that answered most.
A programme director with a grant report due	The grant requires participant feedback. What do I field, and what can I honestly report from it?	A report written from a 12% response rate presented as a finding -- the kind of claim that does not survive a funder's second look.
A PTO or booster board	We want to ask families one thing well, once a year, and act on it.	A thread of replies in three places, no denominator, and a decision the loudest ten families made.
A school already running the Question tool	We have the software. The wording is where we keep getting stuck.	A good engine fielding weak items, which produces confident numbers about nothing.

The objections, answered

These are the six things a sceptical administrator actually says. They are answered here rather than left for an email.

Why would we pay someone to write questions? Anyone can write a survey.

Anyone can, and the failure modes above are what usually comes back. Also: nothing here is for sale today, so this page is not asking you to pay anyone. It states a practice and names the things that go wrong, which is useful to you whether or not you ever write to us.

You sell the tool. Is the consulting just a way to sell more of it?

It would be, if this page pretended otherwise. So the arrangement is stated plainly: the Question family builds a response engine, and this address is about the work that happens before the engine renders anything. The five stages, the failure modes and the denominator discipline apply identically on a paper form or on somebody else's tool.

Our district already has a survey platform.

Then keep it. The constraints section explains what a design has to be true of. If your platform gives you those guarantees, design inside them there. If it does not -- if a response can be joined back to a recipient token -- that is worth knowing before you promise staff anonymity.

Can you analyse the results for us with AI?

No. Nothing in the Question family is AI, and no reading produced by this practice is machine-generated. A reading is written by a person who states the denominator, the drops, and the finding they did not like.

Can we survey students?

That is the case needing the most care and the least software. Student responses are consent-gated and never public, demographic fields are cut to what the decision needs, and the instrument is fielded where anonymity is a property of the storage rather than a promise on the cover page.

What does an engagement cost?

There is no engagement and no fee. No rate, retainer or estimate exists to quote, and this page would be lying if it named one.

How this fits the Question family

Five addresses, five different jobs. This one owns the judgement; the others own the machinery.

Address	What it is	Why it is not this page
question.technology	The shared response engine: architecture, storage guarantees, the measured gaps.	That page explains how the machine works. This one is about what you point it at.
question.software	The classroom front door for one teacher.	One teacher, one class, one set of items. No programme design.
question.management	The district-operations face: standing the tool up across buildings, who can see what, what is still manual.	That is the operations of running the tool. This is the design of the instrument.
question.company	The venture behind the family.	Who we are, not what the work is.
question.email	The mailbox product in the family.	A different product entirely, named here so the family is complete rather than curated.

The practice on this page is deliberately portable. Nothing in the five stages or the six failure modes requires the Question engine, or any software at all. Only the constraints section is engine-specific, and it says so on its face.

Honest status, in the present tense

What exists today is on this page: the practice, written down, with its stages, its failure modes and its constraints. That is real, and it is usable by you without contacting anyone.

What does not exist: an offer, a scope of work, a rate, an availability calendar, a signed engagement, or any surface on this address that takes a booking. This paragraph is written the way it is because the alternative — a page describing a service in the future tense — ages into a promise, and we would be the ones who broke it.

That email reaches a person and starts a conversation. It does not open an engagement, quote a fee, reserve time, or create any obligation on either side, because none of those exist here today.

Nothing here is AI. No item is drafted, no response is summarised, and no reading is produced by a model, in this practice or anywhere in the Question family.

Minors. Student responses are consent-gated and never public. A student-facing instrument is designed with the demographic fields cut to what the decision needs, and fielded where anonymity is structural, for the reason set out in the failure modes above.

Money

There is no price on this page because there is nothing to price. No fee, rate, retainer, day rate, package or estimate exists, and inventing a number to look like a real business would be the first dishonest thing on the site.

There is no checkout, no card field, no invoice and no payment of any kind on this address. If that changes, the change will be a working checkout with a real number on it, not a sentence added here.

The next concrete step

Step one, and you can do it today without us: take the instrument you are about to field and run it against stage 01. Write down the decision it feeds, the person who owns that decision, and the date that decision gets made. If any of the three is blank, the instrument is not ready, and no amount of software fixes that.

Step two: read your draft items against the six failure modes. The double-barrelled item and the missing scale point account for most of what comes back unusable.

Step three: That email reaches a person and starts a conversation. It does not open an engagement, quote a fee, reserve time, or create any obligation on either side, because none of those exist here today.

Write about a question you are stuck on

Questions

What is question.consulting, in one sentence?

The services side of the Question family: the judgement that goes into a survey or feedback programme before any tool renders it.

Is anything on this page purchasable?

No. No engagement is open, no fee exists, and there is no form, checkout or booking anywhere on this address.

What happens if I write to the address?

It reaches a person and starts a conversation. It does not create an engagement, reserve time, or oblige either side to anything.

Do you administer the survey for us?

No. The practice described here is design and reading. Fielding runs on your own surface, or on the Question engine, operated by you.

Is the engine's anonymity claim something we can check?

Yes, and you should. Every constraint below names the schema rule it comes from, in our own migration, so it is checkable rather than asserted.

Does any of this require the Question software?

No. The stages and the failure modes apply on paper. Only the constraints section is engine-specific, and it says so on its face.