Job plain-english-questions-on-a-read-only-warehouse-copy · revised 12 October 2026
Let staff ask the warehouse questions in plain English, with each answer's query shown
A question tool over approved views of a read-only warehouse copy. Every answer shows its query and row count, out-of-scope questions are refused, and a question set you hold back tests it.
You might be seeing
- Simple questions wait days for an analyst
- A chat tool gave a number that looked right and was not
No passwords, keys, card details or admin invites needed to start.
What usually happened
In one vendor test, a tool that writes its own queries against raw tables gave more wrong answers on messy real schemas than one answering through defined metrics, and wrong answers look convincing. One practitioner who built such a tool reported these controls as useful: a read-only role, approved views, query limits and logging. Even then every answer needs its query and row count shown.
Who it’s for: Head of operations or analytics who wants staff to ask the data questions without waiting for an analyst.
Usually starts when: Staff queue for the one analyst, and a pilot that let a chat tool loose on the warehouse gave answers nobody trusted.
The result: A question tool answers only from approved views on a read-only copy of your warehouse. Every answer shows its query and row count. On a 50-question set you wrote and 20 held back, the share answered correctly is reported, wrong numbers fail the whole test, and out-of-scope questions are refused.
Check whether this job fits
Answer these without sending files or logins. Nothing is submitted unless you choose to contact us.
What you get
- The question tool and its settings
- The approved-views list
- The question set with expected answers
- The results table, including refusals
- The kill-switch steps and the log format
Included
- One read-only copy of one schema with personal columns masked, and up to 10 approved views
- A question tool that writes queries only against those views
- A read-only role, a row limit, a time limit and a log of every question and query
- A 50-question set with expected answers: 30 for tuning and 20 you keep from us until the test run
- A kill switch you hold, and a handover with the log format
Not included
- Access to your live production warehouse
- Write access of any kind
- Questions outside the approved views
- Answers presented without their query
- Personal data columns
How we know it’s done
Agreed with you before work starts. Each check produces evidence you keep.
Of the 20 held-back questions, at least 18 are answered correctly against your expected values and the other 2 are refused or marked unsure; any wrong figure fails the whole test. 18 of 20 correct shows the true rate is at least 71% with 95% confidence, and 0 wrong figures in 20 bounds the wrong-figure rate under 14%.
Evidence: The results table beside your expected answers.
10 questions outside the approved views, which you add and we do not see beforehand, are all refused, not guessed; one figure returned for any of them fails the whole test.
Evidence: The ten questions and the replies.
Every answer in the held-back set shows its query and row count, and the log records each question.
Evidence: A sample of 10 answers and the log.
A write or delete attempt that we make as a test is blocked by the platform's role, not by the model.
Evidence: The attempt and the platform's refusal. This is our own test, so it is a smoke test.
An instruction inside a text column, which you plant and we do not see beforehand, changes no answer, and the row and time limits stop a heavy query that you plant.
Evidence: The planted cell, the heavy query and the stop records.
Sign-off. Your tool owner inspects the held-back results, the refusals and the log, and signs off in writing. Payment follows sign-off. You run the tool in your own account.
If it fails. If the held-back test is not passed, you do not pay for this fixed scope. You keep the results, the failures and the test set, and we agree whether to stop or re-quote. No surprise work.
When it fits, and when we stop
It fits when
- A read-only copy with personal columns masked, reached by a secure route after agreement
- Up to 10 approved views whose meaning your team has agreed
- You write 50 questions with expected answers, and a named person owns the tool
- You can switch the tool off yourself
We stop and tell you if
- Only live production access is available
- Definitions of the views are disputed
- You want 'ask anything' over raw tables
- Personal columns cannot be masked
What could go wrong
The job reads copies you send and writes files. It changes nothing in your systems, so there is nothing to revert. Our copies of your files are deleted when the job ends.
Scroll the table sideways to read it all.
| Risk | How we handle it |
|---|---|
| A plausible wrong number is trusted. | Every answer shows its query and row count. A wrong number on the held-back set fails the whole test. |
| The tool is tricked into a destructive or off-limits query. | The role is read-only, limits are enforced outside the model, and a write attempt made as a test must be blocked by the platform. |
| Text in a table acts as an instruction. | Cell text is data. An instruction you plant in a text column must change nothing. |
| Your files go to a model provider outside your control. | Before any file is shared we name the provider, its country and its retention terms in writing. Entry jobs run on redacted, synthetic or public copies, or on your own provider account with a key you hold. |
A separate automated reviewer checks the limits, the log and the write-attempt test before hand-over and does not mark the held-back test. The tool owner on your side writes the questions and expected answers, keeps 20 back and signs off. No person on our side is included unless your agreement names one, and at launch the work is largely automated.
How we deliver
We arrange the work and independent review, then show you the result against the agreed checks. You keep authority over your systems.
- Agree the views, the limits, the question set and the severe errors
- Receive the masked read-only copy and your views' definitions
- Build the tool with a read-only role, limits and logging
- Run the 30 tuning questions, then the 20 held-back ones and the refusals
- Independent review of the log, the limits and the write-attempt test
- Hand over the tool and kill-switch steps; delete our copies
This is a one-off job, not emergency cover or a subscription. We confirm eligibility, the price, a start window and a delivery date before you accept, and work starts only after inputs, a secure handoff route and permissions are in place. Model usage up to a cap agreed in writing is included, and an enquiry creates no charge or booking.
Need to keep it working?
Want the question set re-run each month to catch drift? Ask about a quoted monthly re-run.
Ongoing work is separately scoped and quoted: no monitoring, response-time guarantee or automatic subscription is included in this job.
Explore an ongoing managed lane, or mention the responsibility you need in your enquiry.
What you can check
This is a new service. We have not delivered this job for a client yet.
Other ways to get this done
- The ten-questions job turns ten recurring questions into saved, checked queries and writes no new query at run time. It is the safer start. syntheticindustry.ai
- The defined-metrics job puts the numbers behind a tested semantic layer first, so a question tool has something solid to answer from.
Questions
Can staff ask anything?
No. Only questions the approved views can answer. Everything else is refused.
Can the tool change my data?
No. Its role cannot write, and the platform enforces that.
Do you need my production warehouse?
No. A masked read-only copy is enough.
Send an enquiry
Send us
- The warehouse and the tables behind the views you have in mind
- Five example questions staff ask today
- No real records, personal data or credentials in the first enquiry
Later, once you agree
- A read-only masked copy by an agreed secure route
- The 50-question set with expected answers
- A named owner for the tool
- A written cap on model usage
You keep your warehouse and every decision. We work on a read-only masked copy reached by an agreed secure route, return a tool you run in your own account, and delete our copies when the job ends. After handover the tool runs on your own provider account with your key; each question and its query result go to that provider. The tool's role cannot write, and you hold the kill switch.
Email fallback: open your mail app
If website submission is unavailable, review and send the fallback email yourself. An email fallback is not a website receipt. Or write to hello@syntheticindustry.ai with “plain-english-questions-on-a-read-only-warehouse-copy” as the subject.