# Subliminal Blocker

> Agent-facing counterpart to the [human project page](/projects/subliminal-blocker/).

## Record metadata

- Record: 029
- Slug: subliminal-blocker
- Domain: Software
- Domain code: SW
- Type: Browser extension
- Status: Prototype
- Period: 2026
- Portfolio role: Exploratory tool
- Publication state: Public idea and limitations
- Case-study readiness: Prototype unreliable
- Compendium edition: 0.4

## Summary

A browser-extension experiment for distinguishing subtle promotional influence from ordinary page content.

## Overview

Conventional ad blockers are strongest when advertising lives in a known technical container. Subliminal Blocker investigates the harder category: promotion embedded in ordinary-looking prose, imagery, product placement, or recommendation.

A Chrome Manifest V3 extension with No Block, Subliminal, and Full Block modes, local deterministic rules, and optional browser or API classification.

Purpose: Give users more control over advertising and persuasion that does not look like a conventional ad slot.

## The problem behind the project

Traditional ad blockers focus on known containers, while product placement and persuasive copy can be embedded in normal content.

That category is inherently disputable. The same sentence can be useful information to one reader and commercial influence to another, which means an automated system risks replacing unwanted persuasion with an equally opaque editorial filter.

People seeking stronger control over commercial influence may benefit. Publishers, creators, and users are affected by false positives that remove legitimate material.

## How it took shape

A local-first browser extension, deterministic detection rules, optional classifier hooks, and multiple enforcement modes.

The Chrome Manifest V3 prototype offers No Block, Subliminal, and Full Block modes, using deterministic local rules before optional browser or API classification. The extension structure works, but the central classifier is not reliable enough for ordinary use.

Josiah proposed the broader category of subliminal promotion, defined the product behavior, and directed agent-assisted implementation.

A working extension structure exists, but the central detection does not currently perform reliably enough for ordinary use.

## What the project means now

The failed detection boundary is the most valuable result. A future version should make rules visible, editable, and contestable, treating machine classification as an explanation-bearing suggestion rather than an invisible authority over what the user is allowed to see.

The category is subjective, classifiers can overreach, page content changes rapidly, and remote analysis introduces privacy and cost concerns.

A compelling blocker idea fails if users cannot understand or contest why content disappeared.

Reframe the project around explainable, user-editable rules before attempting broad automated classification again.

## Publication and interpretation notes

- Current classification: Prototype
- Portfolio readiness: Prototype unreliable
- Publication boundary: Public idea and limitations

## Additional agent context

The idea is stronger than the current artifact. Explicitly state that detection is unreliable and do not promise comprehensive blocking.

## Related project records

- [OpenFilter Lab](/projects/openfilter-lab/llm/) — A portable, local-first annotation format for user-controlled mute, skip, and visibility decisions in media.
- [PSALM](/projects/psalm/llm/) — An Apple Music-oriented listening interface that filters unwanted album artwork while preserving the music workflow.

## Navigation

- [Complete project index](/projects/llm/)
- [Human version of this record](/projects/subliminal-blocker/)
- [About Josiah's working method](/about/llm/)
- [Agent discovery map](/llms.txt)