Compliant human-in-the-loop, gamified

When a human in the loop is the law, make it work.

For teams where review is a regulatory requirement, not a nicety. LoopQuest serves your automation's output through review games that keep people genuinely vigilant, then slips in decoys to prove they still are. You get oversight you can evidence, on work people don't dread.

Start reviewing free Why it's required

Drop-in API. Works with Zapier, n8n, UiPath, LangChain and custom scripts.

Swiper

ZendeskAI: Auto-approve 96%

Refund: duplicate annual charge

“I was charged twice for my renewal on 3 June. Please refund the duplicate.”

Plan price: £240 / yr
Amount charged: £2,400

← FlagApprove →

Approve or flag

Connects to the tools you already automate with, now including MCP so Claude, Cursor and your own agents can request human review directly.

View all integrations →

Team vs team

Turn review work into a contest your team actually wants to win

Split a call centre or ops floor into teams and run a competition for a day or a week. Every review scores points; the leaderboard updates live; the winning team takes the prize. It is the simplest way we have found to keep quality high when the work is repetitive.

Pooled XP per team, ranked live — show it on the office screen
Scored fairly by average per member, so uneven teams still compete
Pick one game or all of them for each contest
Everyone’s notified by email the moment it goes live

Included on the Team and Enterprise plans.

updates live as
reviews come in

June Sprint

Live · ends in 2d

All games · avg XP / member · 🏆 £50 voucher

🥇

Day Shift92

🥈

Night Owls78

🥉

Weekend Crew61

Live feed

Priya cleared 8 tasks+80 Day Shift
Marcus caught a decoy+25 Night Owls
Sam fixed an extraction+10 Weekend Crew

A few ways teams run them

Friday Fixer Frenzy

Fixer1 day

🏆 Coffee on the house

Accuracy Cup

All games1 week

🏆 £50 team voucher

Decoy Derby

Swiper1 day

🏆 Leave-early Friday

Sorter Showdown

Sorter1 week

🏆 Trophy + bragging rights

The mandate

Where human review isn't optional

Across regulated domains, the law increasingly requires a person, not just a model, to be answerable for consequential decisions over sensitive data. LoopQuest is the layer that makes that oversight real and auditable.

References are provided for context and are not legal advice. Your obligations depend on your jurisdiction, sector and use case. Confirm them with qualified counsel.

GDPR · Article 22

Individuals have the right not to be subject to decisions based solely on automated processing. Lawful automated decisions on personal data need meaningful human intervention.

EU / UK GDPR

EU AI Act · Article 14

High-risk AI systems must be designed for effective human oversight, and the text itself calls for countering automation bias.

Regulation (EU) 2024/1689

Healthcare

Clinical decision support and PHI workflows keep a qualified human accountable for the outcome.

HIPAA · FDA guidance

Financial services

Sanctions, fraud and AML alerts require human disposition, and model outputs require validation.

FATF · SR 11-7 · ECOA/FCRA

Online platforms

Content moderation decisions need human review paths, statements of reasons and appeals.

EU Digital Services Act

Governance frameworks

Human oversight is a core control in modern AI risk management standards.

NIST AI RMF · ISO/IEC 42001

Two ways to work

Block the automation, or watch it. One API, your call per task.

Some decisions legally need a person to sign off before anything happens. Others just need checking for quality. LoopQuest does both — set mode on each task.

⏸

Gate

Blocks until a human approves

The automation pauses and the consequential action only fires on an approve. For decisions where a person has to be answerable: payments, clinical sign-off, sanctions, content takedowns.

⚙️Automationyour flow

⏸LoopQuestwaits

👤Humandecides

✓Approve → proceeds

✕Reject → blocked

Wait via webhook callback or poll — works in n8n, MCP agents, code, RPA
Fail-closed timeout: no verdict in time → escalates (never silently approves)
Require N reviewers to agree before it resolves (consensus)

📊

Monitor

Reviews quality in the background

The flow proceeds immediately; LoopQuest reviews a copy out-of-band. You get an auditable measure of quality, drift and accuracy — without slowing anything down.

⚙️Automationproceeds now

→Doneno wait

👤Copy → human review

📈→ quality metrics & alerts

Nothing waits — verdicts are logged, scored and trended
Sample a fraction for QA, or review every item
Decoys prove reviewers are genuinely engaged, not rubber-stamping

How it works, across every integration →Recipes for n8n, MCP, Zapier, Make, code and more.

A game for every kind of review

Each task routes to the game that fits the decision — one ingest endpoint, the right interface for the work.

Swiper

Approve or flag outputs, one card at a time

Binary · yes / no

How it works →

Versus

Pick the better of two AI answers

Compare · A vs B

How it works →

Sorter

Drop each item into the right category

Classify · 1 of N

How it works →

Detective

Hunt errors in long generated text

Verify · find the mistake

How it works →

Fixer

Confirm or correct extracted fields

Correct · check the data

How it works →

Redact

Mask the sensitive data an AI left in the open

Redact · tap what's sensitive

How it works →

Grounding

Check an AI's claim against its source

Verify · supported or not

How it works →

The failure mode the law names

Mandating a reviewer isn't the same as having one.

When a model is right almost every time, people stop truly looking and start rubber-stamping. That is automation bias, well documented in human-factors research alongside the vigilance decrement, the way attention fades when you monitor reliable systems.

It matters enough that EU AI Act Article 14 explicitly requires oversight measures that keep people aware of the tendency to over-rely on AI output. A bored box-ticker satisfies the org chart, not the regulation.

The Decoy Matrix answers it directly

LoopQuest seeds a mathematically generated decoy error every 50 to 100 tasks. Catch it and you are rewarded. Miss it and the system breaks the trance with an alert, then logs it. You get a measurable, auditable signal that your reviewers are genuinely engaged, not just present.

Per-reviewer catch-rate and automation-bias score
Immutable audit log of every decoy
Accuracy-weighted scoring that punishes blind approvals

The reasons that outlast the regulation

Accountability

Regulators and customers want a named human who can stand behind the decision, not just a model that produced it.

Edge cases

Models fail on the unusual: distribution shift, novel inputs, adversarial content. That is precisely where a human call earns its keep.

Better signal

Every verdict is labelled evaluation data. Your review queue quietly becomes the dataset that improves the model.

People stay sharp

Review work is monotonous, and monotony breeds error. Making it a game keeps skilled reviewers engaged and accurate over the long haul.

One integration, and your pipeline keeps flowing.

Ingest

POST output to /api/v1/tasks with a workspace key. The Decoy Matrix may quietly alter a copy, while the real data flows on untouched.

Review

A human judges it through the fitting module, earning XP, ranks and league spots while staying vigilant.

Return

The verdict and metadata webhook straight back to your system, with every decision logged and auditable.

curl -X POST https://loopquest.tomphillips.uk/api/v1/tasks \
  -H "authorization: Bearer $LOOPQUEST_KEY" \
  -d '{ "module": "swiper",
        "payload": { "ticket": 4821, "predicted": "escalate" },
        "callback_url": "https://your-app.com/verdict" }'