The Three-Tier Content Review Pattern: Why One Approver Is Never Enough on Air

By the KAVANA engineering team — June 2026

A few years ago we watched a broadcast compliance incident unfold that had a straightforward cause: a single editor had approved a piece of content under deadline pressure, the editor was also the person who had written the content, and neither the writing nor the approval was done carefully. The content aired. The regulatory complaint arrived six days later. The station had a complete record of who had written the content and what time it was written. They had no record of the review — because there had been no real review, just a checkbox ticked by someone who wanted to go home.

This post explains the content review architecture we built to prevent that pattern from recurring at the stations we serve, why three tiers are structurally necessary rather than administratively convenient, how the audit trail is structured, and how the same configuration maps onto the different regulatory frameworks our international customers operate under.

The Failure Mode of Single-Approval Systems

Single-approval content review fails in a specific and predictable way under production pressure. The failure is not that reviewers are careless people. It is that the single-reviewer architecture creates conditions where carelessness is the rational response to the incentive structure.

When one person is responsible for both production and approval — or even when one person is responsible for final approval with no structured handoff from production — several things tend to happen. First, the reviewer develops familiarity blindness: having worked with or written the content, they read what they expect to see rather than what is actually there. This is a well-documented phenomenon in editorial quality control and it applies equally to broadcast content. Second, when there is one approval gate and it is approaching deadline, the cost of a rejection falls entirely on the reviewer: saying no means the content does not go to air and the reviewer must explain why. The rational choice under time pressure is to approve and hope. Third, there is no structural separation between the person who knows the editorial context and the person who knows the compliance requirements — they are the same person, trying to apply both lenses simultaneously, and one of them usually wins.

The rubber-stamping problem — where the approval record exists but the review did not happen — is a variant of the same issue. It produces an audit trail that looks clean but is meaningless as a compliance artifact. When something goes wrong, the record shows that someone approved the content. It does not show that anyone read it carefully.

Adding a second approver helps with the familiarity blindness problem. It does not necessarily help with the deadline pressure problem or with the structural mixing of editorial and compliance perspectives, because the second approver is usually drawn from the same pool and working under the same time pressure as the first.

Three tiers, with distinct role assignments and distinct review focuses at each tier, are the minimum structure that addresses all three failure modes simultaneously.

The Three Tiers: Distinct Focus, Distinct Personnel

The three tiers in KAVANA's review system are not three people doing the same review. They have explicitly different charters.

First review — editorial correctness. The first reviewer's job is to verify that the content is factually accurate, editorially coherent, and appropriate for the format. For AI-generated content this means checking the output against the source material: did the synthesis stay faithful to the facts, did it hallucinate claims, does the phrasing match the station's editorial style? For human-produced content it means the standard editorial review: facts checked, claims sourced, phrasing clear. The first reviewer should ideally be someone close to the content domain — a journalist reviewing news, a traffic editor reviewing road conditions updates. They should not be the content producer, because the familiarity blindness problem applies.

Second review — compliance and standards. The second reviewer's job is regulatory and standards compliance. Does the content conform to the broadcast code applicable in this jurisdiction? Does the sponsor identification language meet the required format? If AI synthesis was used, is the disclosure language present where required? Is the content appropriate for the daypart and the likely audience profile? This review requires different knowledge than the first — familiarity with the specific regulatory framework, not just editorial judgment — and it should be performed by someone with that specific training. In a small station this may be the station manager; in a larger organization it may be a dedicated compliance officer.

Third review — final authority and air clearance. The third reviewer is not doing a complete re-read of the content. They are making the final authorization decision, verifying that the previous two reviews have been completed and signed, checking for any issues flagged as requiring senior judgment, and taking ownership of the decision to air. In regulatory terms, this is the person whose name goes on the authorization. In operational terms, this is the person who has to be accountable when something goes wrong. The third reviewer should have broadcast authority — the legal and organizational standing to authorize transmission — not just editorial standing.

These role definitions mean that the three reviewers need different things from the review interface. The first reviewer needs to see the content alongside its source material. The second reviewer needs to see the compliance checklist for the applicable regulatory framework. The third reviewer needs to see that the previous reviews have been completed, any flags or reservations that were noted, and the key parameters (air time, daypart, content type) without necessarily re-reading the full content. The KAVANA review workflow implements these three different views — the same underlying content record, presented differently at each tier.

The Audit Trail: What Gets Stored and Why the Structure Matters

An audit trail that cannot be queried is not an audit trail. It is a log. The distinction matters when a regulatory body asks you to demonstrate that a specific piece of content was reviewed in accordance with your stated procedures.

Each content item in KAVANA's system carries a review record with the following structure. The record is append-only — existing entries cannot be modified, only new entries can be added, and the system records any attempt to add a retroactive entry.

Each review action is recorded with: the reviewer's authenticated identity (not a display name — the account credential that was used, tied to the specific user record in the system); a cryptographic timestamp from the server (not the client, to prevent clock manipulation); the version hash of the content that was reviewed (so the record shows precisely what text was approved, not just that some version of this content was approved); the tier and role of the reviewer; any flags or reservations entered by the reviewer; and the disposition (approved, approved with conditions, referred to next tier, rejected).

When content is edited between review tiers — which happens — the edit creates a new version record. The second reviewer's approval record references the version hash they reviewed, not the version the first reviewer reviewed. If the versions differ, the system flags this and the third reviewer sees that the content was modified between first and second review, with a diff.

The diff capability is not a nice-to-have. It is structurally necessary for compliance purposes. The audit question "was this content approved as aired?" cannot be answered without a record that shows both what was approved at each tier and whether what aired matched the final approved version. KAVANA-DOG generates a playout record that includes the audio file hash for each transmitted segment. The compliance query can therefore trace from the transmitted audio back to the approved content record, confirming the chain.

Retention periods are configurable by jurisdiction. The system flags records approaching the configured retention window and requires an explicit action to either extend retention or archive. The default configuration retains records for seven years, which covers the most demanding regulatory frameworks we currently support.

What an Auditor Can Actually Find

Regulatory inquiries about broadcast content typically arrive weeks or months after the content aired. The auditor's question is not "what do your procedures say?" — any station can produce a procedures document. The question is "can you demonstrate that your stated procedures were actually followed for this specific piece of content on this specific date?"

The scenario: a complaint has been lodged about a news segment that aired on a Tuesday morning. The segment contained a factual claim that the complainant argues was false and damaging. The regulator wants to see the review records for that segment.

In KAVANA's system, the query path looks like this. The administrator queries the content library by air date and time. The system returns the content record including the scheduled air time, the actual transmission time from the DOG playout log, and the audio file hash. The review record attached to that content shows: first review completed by [reviewer identity] at [timestamp], version hash [X], flagged issues: none; second review completed by [reviewer identity] at [timestamp], version hash [X, same]; third review authorized by [reviewer identity] at [timestamp], version hash [X, same]. No version changes between review tiers. The authorized version matches the transmitted audio by hash.

This demonstration takes about three minutes if the content record exists and the chain is intact. It takes considerably longer — and may not be completable — if the review was performed informally with records scattered across email threads and personal notes.

What the auditor cannot find in this query is evidence that the first reviewer actually read the content carefully rather than ticking the box. The audit trail proves the process was followed; it cannot prove the quality of attention that was applied. This is an honest limitation. The three-tier structure makes careless approval more difficult by distributing the review responsibility and requiring three separate people to make affirmative decisions. It does not make it impossible.

Regulatory Mapping: How One Configuration Serves Four Frameworks

The stations we serve operate under four major regulatory frameworks: Ofcom in the UK, FCC in the United States, ACMA in Australia, and NRTA in China. Each has distinct requirements for broadcast content review and record-keeping. The good news is that the underlying structure of three-tier review with complete audit trails satisfies the record-keeping requirements of all four. The differences are at the configuration layer, not the architecture layer.

Ofcom (UK) requires that licensees maintain systems for ensuring compliance with broadcast codes, that those systems include editorial oversight, and that records be retained for a period sufficient to respond to potential complaints. Ofcom's Broadcasting Code does not specify a minimum number of reviewers. It requires that stations demonstrate a functioning compliance system. The three-tier review record, with the second tier explicitly designated as compliance review, maps cleanly onto this requirement.

FCC (US) requirements for review and record-keeping are tied to the station's license class and format. The political broadcasting requirements (equal opportunity, sponsorship identification) require specific documentation that the applicable rules were followed. The KAVANA compliance checklist for second-tier review includes FCC-specific items for stations that configure the US regulatory profile: political content identification, sponsorship disclosure language, obscenity/indecency classification for the daypart.

ACMA (Australia) operates a content standard framework under the Broadcasting Services Act. The ACMA's compliance requirements include record-keeping obligations for sponsored content and restrictions on certain content categories by time of day. The second-tier compliance checklist for the Australian profile includes ACMA's classification categories.

NRTA (China) has the most detailed content review requirements of the four frameworks, reflecting a regulatory philosophy that broadcast content should be reviewed before air rather than subject to post-hoc complaints. The three-tier structure is a direct analog to the review requirements in NRTA guidelines for broadcast production organizations. The record retention requirements align with the seven-year default configuration.

The system supports multiple regulatory profiles simultaneously — relevant for international broadcasting groups that operate stations in more than one jurisdiction from a shared content operation. A piece of content intended for both Australian and UK distribution carries compliance checklists for both profiles, and the second-tier reviewer must clear both.

Why AI-Generated Content Must Clear the Same Three Tiers

The argument we sometimes hear is that AI-generated content does not need the same review structure as human-produced content because it can be checked computationally — content filters, factual verification APIs, hallucination detection. Run the automated checks, and if they pass, the content is cleared.

We disagree with this reasoning, and not primarily because AI systems are unreliable — though current systems do hallucinate, do produce confident-sounding false claims, and do generate content that passes automated filters while containing errors that a knowledgeable human reader would immediately catch.

The deeper reason is that three-tier review exists to create accountability, not just to catch errors. When a broadcast incident occurs involving AI-generated content, the regulatory question is not just "was this content checked automatically?" It is "who authorized this content for transmission, and what was the basis of that authorization?" An automated filter is not an authorizing party. A human reviewer who applies their professional judgment and signs their name to an authorization record is.

The volume argument cuts the other way from what it might seem. AI-generated content enables stations to produce significantly more content per day than human production workflows allow. Higher volume means more opportunities for something to go wrong, not fewer. The AI Three Gods system is specifically designed to handle high-volume AI-generated content production with the three-tier review workflow as a non-negotiable step — not an optional add-on for stations that want extra assurance, but a structural requirement of the content pipeline.

The practical implementation for AI content does have one adjustment: the first-tier review is configured to surface the specific risk areas for AI-generated content (factual claims that should be verified against the source data, phrasing that sounds generated rather than broadcast-natural, duration anomalies that suggest the synthesis model truncated or padded the output). The first reviewer's checklist for AI content is different from the checklist for human-produced content. The tier structure and the audit trail are identical.

The Overhead Question

Three-tier review takes more time than single-tier review. This is true and there is no point pretending otherwise. For a station that has been running on informal single-approval workflows, implementing structured three-tier review will slow down the content production process, at least initially.

The friction is real. The overhead is also real. The question is whether the overhead is proportionate to the risk it mitigates.

For a county-level station producing straightforward format programming with low controversy potential, the risk of a significant regulatory compliance incident is low, and the three-tier overhead may be disproportionate. The system supports configuring lighter review structures for content categories that carry lower risk.

For a station producing news and current affairs, political programming, sponsored content, or AI-generated content at volume, the risk profile is different. A single compliance incident of meaningful severity — a broadcast standards violation, a false claim in a news segment, an undisclosed sponsorship — has costs that include regulatory fines, license review, and reputational damage. The three-tier review overhead is small relative to those costs.

The audit trail itself has value beyond compliance. Stations that have been operating with informal review processes consistently report that moving to structured review with a complete record reduces the internal time spent on post-incident analysis, because the record makes it straightforward to determine what happened and why. That reduction in investigation overhead partially offsets the upfront review overhead.

For technical documentation on configuring the review workflow, the KAVANA AI Three Gods documentation covers the tier structure and role assignment. The KAVANA AI Utilities page covers the integration with the broader content production pipeline. Stations evaluating the system are welcome to contact us at international@kavanafm.com with questions about regulatory mapping for specific jurisdictions.

KAVANA is developed by Hunan ShengGuang Technology Co., Ltd. (湖南声广科技有限公司), incorporated 2012, team active since 2005. We hold a broadcast production and distribution license (湘字第00565号) and operate under Chinese cybersecurity Level 3 certification. Technical documentation and open specifications: github.com/kavanafm.