Standing service staff-leave-assistant-retested-and-chats-read-every-month · revised 12 October 2026
Standing service
Keep your staff leave assistant tested and its chats read every month
Each month the leave assistant's frozen test and leak test are re-run on your latest export and a sample of chats is read for severe errors, then reported to your HR person.
This starts a conversation by email. Nothing is charged, and nothing is run, until we have agreed scope, access and terms with you in writing.
The responsibility you hand over
Leave rules turn over every year: allowances reset, carry-over caps apply and part-year starters change. An assistant tuned on last year's rules gives staff the wrong days, and one wrong answer about a colleague can reveal who is off sick. The monthly job re-checks the sums against your rules table, re-tests the colleague-leak questions and reads real staff chats for health or pay details.
Who it’s for: Office manager or HR lead whose staff use a leave assistant and who needs to know each month that it still answers correctly and shows nobody else's leave.
Usually starts when: The assistant is live, the leave year turned over or the export changed, and nobody re-tested it.
The result: Each month you receive the re-run of the frozen 100-question test and the leak test on your latest export, a read of up to 30 randomly drawn chats and a list of cases that changed since last month. A severe error goes first in the report.
What stays true, and what we do about it
No response-time guarantee is published for this new service. A target for delivering each month's report is agreed in writing before it starts, set to what a service at this stage can keep.
Hours are agreed in writing before the service starts. At launch the service is not staffed round the clock and runs once each month. The assistant runs in your tool and you hold its off switch.
What must remain true
- Every checked figure matches the expected answer from the dated export and rules version used for that chat, and the export date is stated
- The assistant gives one person's row only, and the leak test finds none
- Every handover trigger still hands over to your HR person with the chat attached
What we watch
- A test copy of the current leave export in the live layout
- The frozen regression pack, expected answers and fresh buyer-held cases for changes
- Up to 30 randomly drawn chats a month, redacted by you
When something happens
Scroll the table sideways to read it all.
| When | What we do |
|---|---|
| The leave year turns, or the export layout or rules table changes. | We re-run the whole pack on the new layout before you rely on it. Linked job: Staff leave-balance assistant |
| A severe error appears in a sampled chat. | It goes first in the report with the transcript, and the report points to the switch-off note. |
| The chat tool or its model version changes. | We re-run the full pack and report every case that now fails. |
We do on our own
- Re-run the frozen test on a test copy
- Read the sampled chats and write the report
- Offer the one-off job when the leave year or layout changes
We ask you first
- Any change to what the assistant may say, its handover triggers or pass marks
- Any access to your live HR system or chat tool
- Switching the assistant on or off
We escalate to you when
- Any checked figure differs from the expected answer for the dated export and rules version used for that chat
- Any chat shows another person's data
- The same failure appears two months running
How you know it held. Each month you can re-run the frozen test with the steps we hand over. The report states the cases run, passed and failed, chats read, severe errors found and cases that moved since last month.
How we keep it true
This service is never finished. Each month's report shows whether the thing it watches stayed true, and it continues until you end it.
Let staff ask about their own leave balance and get an answer from your HR export Job Optional
An automated assistant tells each signed-in member of staff their own leave balance from your HR export, and hands pay, health, discipline and legal questions to your named HR person.
Bought separately, each time it is needed; not part of the monthly fee
What is included, and what is not
- Monthly test report
- Chat-sample findings
- List of cases that changed
Included
- One assistant in one chat tool
- The frozen regression pack re-run on a test copy of the latest export, plus 20 fresh buyer-held labelled cases before a changed workflow is accepted
- The leak test repeated on 20 test accounts
- A read of up to 30 randomly drawn chats, redacted by you
Not included
- Switching the assistant on or off: your HR person does that
- Changing what it may say without your written approval
- Reading every chat: up to 30 a month are sampled
- Any response-time guarantee or out-of-hours cover
- HR, legal or pay advice
How we know it’s done
Agreed with you before work starts. Each check produces evidence you keep.
Each month the 100-question test is rerun on a test copy: figures match your HR person's expected answers from the rows and current rules, including calculated balances, and no leak occurs. One wrong figure or leak fails the month; report each failure and the rerun evidence.
Evidence: The monthly test report.
Each month the frozen test pack is re-run and every previously passing case still meets its agreed check. A regression fails this check even if reported, and the report lists it first with its cause.
Evidence: The monthly test report.
The chats you share (up to 30, drawn at random by you) are read and every severe error is listed with its transcript.
Evidence: The findings list and the transcripts.
Nothing was written to any live system by us, and originals are unchanged (checksums match).
Evidence: Checksums before and after, and your own audit trail.
Before any changed workflow is accepted, you supply 20 new labelled cases, held back from us until the test run, including severe-error cases from the matching one-off test. All 20 must meet their labels and no severe error may occur; we return the case file, results and rerun steps. This is a smoke test on those cases, not a future-rate claim.
Evidence: The fresh buyer-held labels, transcripts and rerun steps.
Sign-off. You read the report and test results and accept or send them back in writing. A month counts as delivered only when its agreed acceptance checks pass; receiving a report alone is not acceptance.
If it fails. If we cannot run a month because an input is missing or has changed shape, we say so, explain what we need and do not guess. If the service is not working for you, you can end it at the end of any month.
When it fits, and when we stop
It fits when
- The staff leave assistant is set up and your rules table is current, with the leave-year dates written down
- Each month's annual-leave export arrives with no reason columns, in the agreed layout
- A named HR person who reads each report and holds the off switch
We stop and tell you if
- The export gains reason or sickness columns that cannot be removed
- Nobody on your side reads the monthly report for two months running
- You want it to answer pay, health or legal questions
What could go wrong
The service reads, tests and reports. Ending it leaves your systems and the way you work exactly as they are, and you keep every report we sent.
Scroll the table sideways to read it all.
| Risk | How we handle it |
|---|---|
| The chat sample is read as proof that no error exists. | Thirty chats a month is a tripwire for large failures. The report says so and gives the number read. |
| Staff chats hold health or personal details. | You redact before sending, the assistant is not meant to hold any, and a chat that does is reported as a severe error. |
| Text inside a document, record or message tries to give the automation instructions. | All input text is data. The automation has no tool to send, approve, file or change anything beyond what is listed here, and planted-instruction items must change nothing. |
| Real documents or records pass through a model provider. | First tests use invented, synthetic or redacted data. Real data moves only after a written agreement names the model provider, its country, how long it keeps data and whether it trains on it. |
A separate automated reviewer checks each month's report against the agreed test before it is sent, and your named person decides what to do with it. No human supervisor from us is included unless your agreement names one. At launch the work is largely automated, and we say so.
Stays with a person
- Your named HR person decides what to do with each report
- You switch the assistant on or off
Access we would need
- Test copies and redacted samples sent through a secure route agreed in writing
- No login, key or write access to your live HR system or chat tool
Questions
Do you read all my staff chats?
No. You draw up to 30 at random, remove names and send them.
What if a severe error appears?
It goes first in the report, and your HR person decides whether to switch the assistant off.
Also part of
Send an enquiry
Send us
- Which assistant and tool you use and roughly how many chats a month
- Who receives the report. No staff data in the first enquiry
Later, once you agree
- The current test-export copy and redacted random chat sample, plus the dated export version and settings used for each sampled chat, through the agreed secure route
- The frozen regression set, current expected answers and pass marks; 20 fresh labelled cases held back until the run before any changed workflow is accepted
- A named HR person
You keep the leave system, the chat tool and every login. We work from test copies and redacted chat samples sent through a secure route agreed in writing and hold no access to your live systems. Our copies are deleted, in writing, when the job ends.
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 “staff-leave-assistant-retested-and-chats-read-every-month” as the subject.