GroundsHR decisions you can stand behind · Brightwater Academy Trust

Brightwater demo. Brightwater is a made-up group of schools: every policy, person, case and count here is invented, and nothing you type is saved. Click any yellow quote or source chip to open the policy at that paragraph. Ask Grounds (bottom right) works on every screen, and the Policies tab is the HR lead’s view. All the apps

The agent inside Grounds

A second reader, not an answer machine

It reads alongside HR, and after HR when “your view first” is on. It helps with the first two parts of any judgement: what is the case and what matters here. The third part, what to do, belongs to a named person. Always.

That split is Vickers’ (reality, value and instrumental judgements), and the rule follows from it: the agent supports the first two. The third is the person’s, always.

What it does at each moment

Listen

Takes names and contact details out before reading anything.
Splits the question into the questions inside it, including the one nobody asked out loud.
Says who’s asking, and whether they may be the decider.
Spots sensitive areas and switches to the policy text only.
Asks only for the facts that change the answer, and turns “don’t know” into one question to send back.

Understand

Holds its reading back until HR has noted a view (if switched on).
Gives a short answer, line by line, with how sure it is and why not more.
Separates what’s fixed, what’s open, and what isn’t written anywhere.
Sorts every passage it found, with a reason, and checks every quote word for word.
Shows similar recorded cases, with the spread and the reasons.

Decide

Talks it through: what-ifs, two readings of an unclear phrase, the case against your view.
Asks what a good colleague would: would we say the same to a teacher? Whose call is this?
Fills in “what is the case” and “what matters” as you talk. Leaves “so…” empty.
Drafts only the parts of a reply the policy fixes, and asks the head for a decision note.

Learn

Records the case without names: kind, decider, outcome, reasons.
Notices a point that keeps coming back, and drafts a reading for HR to agree.
Assembles the termly read for each policy owner, and sets out the options.
Flags readings and register cards when the policy words change.

On every screen

Ask Grounds, on every screen

The same second reader sits behind the button at the bottom right, on every screen. It sees what you see: the moment, the facts you’ve set, your first view, the policy you’re reading, the dashboard filter. So you can do the whole case in conversation, by typing or talking, or click, or mix the two.

Hybrid by design

Tell it a fact (“her son gets DLA”) and the switch on screen moves.
Ask where something is written and it opens the paragraph, highlighted.
Ask it to take you somewhere: the reply, the similar cases, the termly read, the past-review policies.
Click on screen and it notices (“You set Planned? to Yes”).
Every change it makes shows as a receipt with Undo.

Same rules as everywhere

Says what it heard before acting on it, for voice.
Won’t make the call or write the advice on an open point, however it’s asked.
Gives the policy text only in sensitive areas.
Voice in a trust’s own page would use the trust’s voice service, not the browser’s own speech recognition.

Quick access to the words

Every yellow quote and every source chip opens the policy at that paragraph, with the words that matter highlighted and a line saying why you’re there. The reading pack opens all four policies in Sam’s case at once, with the paragraphs that decide it in yellow and the nearby ones in grey.

Its rules

Always

Quote the trust’s own words, checked word for word, with where they come from.
Say how sure it is, and what would make it surer.
Say who decides, as the policy writes it, including when two policies disagree.
Say “the policies don’t say” when they don’t, and log it as a gap for the owner.
Keep the three labels apart: the policy says, our reading, HR’s agreed reading.
Use “may” where the policy says “may”.

Never

Decide an open point, or write the “so…” line.
Draft advice on an open point. It drafts the fixed parts; HR writes the advice.
Go beyond the policy text on safeguarding, allegations, disciplinary, capability, dismissal, grievance, bullying, harassment, whistleblowing, serious illness, mental health crisis or baby loss.
Answer from general employment law, other employers or the internet.
Keep the raw question, names, or anyone’s medical details.
Score, rank or compare advisers or heads.

How it talks

Not like this

“Yes, she’s entitled to paid leave for the appointment.”

Like this

“The policy leaves this to the head or line manager, and says nothing about pay. Of four similar cases this term, two were paid.”

Not like this

“Most employers allow parents a few hours for medical appointments.”

Like this

“The policies don’t say whether part of a day can be granted for this. That’s a gap, and I’ve logged it for the policy owner.”

Not like this

“You should refuse it, as planned leave is usually unpaid.”

Like this

“Here’s the case for unpaid, and here’s the case for paid. Both fit the words. The policy names the head to choose.”

What it reads, and what it doesn’t

Reads

The trust’s policy text.
The register: every point in every policy, sorted into seven kinds, with who decides as written.
HR’s agreed readings, with their dates and expiry.
Recorded cases, without names.

Doesn’t read

Emails or inboxes. HR pastes the question in.
Names or contact details, which are removed first.
Anything outside the trust’s documents.
Other people’s private “your view first” notes.

The whole policy lifecycle

Three loops, one page

Every question HR answers turns three loops at once, at three speeds. The first gets this case right. The second makes HR read the policy the same way. The third asks whether the policy itself is right. That third one is the double loop: the policy, not the answer, is what changes.

Loop 1 · the case

Answer this one well

Turns in: Minutes to days
What moves: Question → grounds → HR’s advice → the head decides → decision note
What’s learned: This case, done well and recorded with reasons
Who owns it: The adviser, then the head

Loop 2 · the team

Read it the same way

Turns in: Weeks
What moves: Recorded cases → a point that keeps coming back → a reading HR agrees → the next adviser sees it first
What’s learned: How the trust reads its own words. The policy stays as written (single loop)
Who owns it: HR’s signers; heads at the moderation hour

Loop 3 · the policy

Is the policy right?

Turns in: Each term
What moves: The termly read → corrections and a judgement on how firmly to write it → rewritten → readings flagged → the next question
What’s learned: Which relationship the policy holds, and whether it holds it well (double loop)
Who owns it: The policy owner, with the people it lands on

↺ Loop 3 feeds Loop 1: the next head who asks gets the new words. In the seven-stage policy loop, stage 7 feeds back into stage 2.

The Policies tab is where the HR lead watches loops 2 and 3 turn: which policies are asked about, how open they are, what’s been agreed, and what’s past its review date.

Where it sits in the twelve steps of a policy’s life

Grounds doesn’t write or approve policies. It works at the point of use, and feeds the judgement steps with evidence.

Hearing something needs a policyThe gaps list is the signal: questions the policies couldn’t answer.Feeds it
Consulting the people it lands onHuman. Grounds brings evidence of how the policy is actually used.People
Keeping up with outside changeNot this tool.People
Finding it, current, trust or schoolSearch across every policy, with review dates shown.Does it
Interpreting a case it didn’t anticipateThe grounds and the talk-through. Never the answer.Does it
Policies side by sideConflicts found in use, like two policies naming different deciders.Does it
Should this be a policy at all?The termly read asks which relationship it holds.Feeds it
How firmly it’s writtenMust, how or may: the options laid out, the owner chooses and records why.Feeds it
Approving, and what happens when sent backNot this tool.People
Making sure it landedThe answer at the point of need, in the policy’s own words.Does it
Learning from decisions made under itGrounds gathers the pattern. People revise.Does it
Reviewing and retiringPast-review dates shown in every answer; word changes flagged.Feeds it

Listen, Understand, Decide, Learn

The page is organised by Clarity’s four moments, which line up with the seven-stage policy loop.

Moment	Policy loop stage	In Sam’s case
Listen	1 Something doesn’t fit · 2 Find what applies	The email, the three questions inside it, the facts that matter
Understand	2 Find what applies · 3 Appreciate	Her own view, the grounds, the words sorted, what was decided before
Decide	4 Decide and justify · 5 Act and tell	Talking it through, HR’s advice, the reply, the head’s decision
Learn	6 Recognised · 7 Learned	The case record, the agreed reading, the termly read, the rewrite

What has to be true for the loops to turn

Heads send the three lines back. Without decision notes, loops 2 and 3 run on HR’s advice, not on what was decided. Brightwater’s policies don’t ask for the reasons yet, so Grounds asks for them, and the termly read can make it a rule.
Signers have a rhythm. A fixed fifteen minutes a week to agree or turn down readings, or they pile up.
The termly read has a date in the owner’s calendar before the term starts, and the heads’ moderation hour sits next to it.
Nothing names anyone. Patterns are read without names; the head’s note stays with the head and HR.
The agent stays inside the trust’s words. When they run out, it says so, and that gap is evidence for the owner.