Use cases

Five uses for one recorded decision.

Data passes from the Database through the AI globe to the 4loop logo. Hover or focus the diagram to reveal labels for the Database, AI, 4loop, Data, Outcome, Prompt, and all five passages. Data, Outcome, and Prompt sit above their flow lines. The object labels use elbow leaders. A separate Data branch enters 4loop from above. A Prompt branch connects the LLM to 4loop from below. The stationary storage core contains a soft orange orb in its second, middle square while the globe rotates continuously around a larger, softly glowing lavender orb at its center. Two balls travel along the stationary orbital ring. Orange dots flow from the drive into the globe. Lavender dots follow the Outcome path from the globe into 4loop. Two orange dots branch upward along the Data path into 4loop. White dots follow the lower Prompt path into 4loop. The final arrow splits into five passages: Pass Through, Audit, Human Gate, Fine Tuning, and Ai Onboarding. Each passage ends in a white orb. Purple dots continue into Pass Through. Purple, orange, and white dots flow into Audit, Human Gate, Fine Tuning, and Ai Onboarding. Their coronas brighten in a wave from the top orb to the bottom orb and back, while the cores stay fixed.

AI onboarding

Build trust before implementation.

A new agent starts with a person on the decision, so you can trust the system before you implement it. Someone approves, rejects, or sends the work back, and those judgments show what to keep, what to change, and what is not ready to go live. The review criteria can travel with that decision.

Fine tuning

Stored for a later model.

Each approval, rejection, and reason stays with the work it judged. Today 4loop stores that record, so the data can become your future asset. Fine-tuning on it is a later step. Before that training, 4loop will classify the data. A smaller classified set costs less to train, and the work stays your intellectual property.

Human gate

The draft waits.

Live work waits until someone approves it, rejects it, or sends it back. For an important decision, 4loop adds an AI suggestion beside that choice.

Audit

A record you can open.

The record holds who decided, when, and the reason, with the data that went in, the prompt, and the outcome. A stored hash shows which bytes arrived and which were reviewed. You can query that record and check it later.

Pass-through

Verified work moves on.

The record of what passed stays. After the decision, one POST goes to an HTTPS address you registered. Remote calls are on Growth and Enterprise.

AI onboarding

A new agent starts with a person on the decision.

The first stretch of an agent is when a mistake is still new to the team. Use it to vet the system before you implement it. Send that work through 4loop. A person approves it, rejects it, or sends it back with a reason. The review criteria can travel with the decision.

Customer replies, intake notes, and internal drafts can all start on this path.
A review can hold more than one decision.
The record of those early judgments stays with the work.
Those judgments vet the system before implementation: what to keep, what to change, and what is not ready to go live.

Fine tuning

The decision is stored, so a later model has something real to learn from.

Each approval, edit, rejection, and reason is kept with the work it judged: who decided, when, and why. A team can later use that record to tighten a prompt, build an evaluation set, or fine-tune a smaller model on its own reviewed work. Today 4loop stores the human decision. Fine-tuning on that record is a later step. Before that training, 4loop will classify your data. A smaller classified set costs less to train, and the work stays your intellectual property.

Before training, 4loop will classify your data.
That classified set is smaller, so the training costs less.
Your intellectual property stays yours.
The payload is hashed on the way in. The decision is stored as a receipt.
Provenance stays on the record: the source, the agent, the reviewer, the change, and the reason.
A stored decision is the material for that later step. It is not a claim that the draft was right.

Human gate

The draft waits until someone decides. The record stays either way.

Meaningful work can be held until a person approves it, rejects it, or sends it back. For an important decision, 4loop adds an AI suggestion beside that choice. Approval can let it move forward. A return sends it back to be revised. The decision is written down in both cases. The decision record is the audit. Work that is allowed to continue, and the record of that passage, is in pass-through record keeping.

Audit

The decision is a record you can open later.

4loop keeps who decided, when, and the reason with the output. The record also holds the data that went in before the AI outcome, the prompt, and the outcome. The submitted payload is hashed on the way in. A second hash seals that digest to the review, the task, and the template. On submit, 4loop hashes the bytes again and checks the seal.

The data that went in, the prompt, and the outcome stay together on the record.
If the bytes no longer match, the reservation is refused and the first record stays.
The decision is stored as a receipt, with its own SHA-256 digest. Archive does not erase that receipt.
Provenance stays on the record: the source, the agent, the reviewer, the change, and the reason.
A stored hash shows which bytes arrived and which bytes were reviewed. It does not mean the draft was right.

Pass-through record keeping

The verified work can move on. The record of what passed stays.

4loop can receive data before an agent processes it. Those bytes are hashed on arrival, SHA-256, before that work is submitted. After a person records a decision, 4loop can send the verified payload on: one POST to an HTTPS address you registered. The call waits for that decision. If you have a trusted system, it can go on autopilot, and 4loop can collect the data that went into the workflow.

A trusted system can go on autopilot after the decision.
4loop can collect the data that went into that workflow.
The unchecked draft is not what gets sent.
A failed call does not undo the decision. Remote calls are on Growth and Enterprise.
The arrival hash shows which bytes came in.
Provenance stays with what passed: the source, the agent, the reviewer, the change, and the reason.