Laravel Word Refiner Laravel Package

Product Decisions This Supports

• Shipping a content-moderation v1 or deferring it. A word-level refiner lets you answer the "can users post this?" question for comments, reviews, display names, support tickets, and marketplace listings without signing a vendor contract. The PM decision is scoped: do we gate UGC at launch with a word-list approach, or delay until we can afford contextual moderation?
• Build vs. buy for text sanitization. This package is the "buy-cheap" middle path — MIT-licensed, self-hosted, no per-call cost, no data leaving your infrastructure. It establishes a baseline that either (a) satisfies the requirement permanently for a low-risk surface, or (b) buys you time until usage justifies Perspective API, AWS/Azure content moderation, or a trust-and-safety vendor.
• Feature-level decisions it directly unlocks: a "clean language" or "professional tone" toggle, username/pseudonym validation, pre-publish review screening, support-inbox triage and tagging, marketing copy linting, and word normalization for search/SEO. Each is a discrete backlog item you can estimate and A/B against.
• Risk posture and vendor-dependency strategy. With 25 stars and a last release of April 2023, the real PM decision is not "does it work" but "who owns it in 18 months." That drives whether you wrap it behind an internal TextRefiner interface (recommended) so it's swappable, and whether you budget for a maintained fork.
• Compliance and store-review readiness. For app-store review, age-gated communities, or enterprise procurement questionnaires, being able to say "we apply automated language filtering at ingest" is a checkable box. This package makes that claim cheap to make true — but only if you can also describe the word list and its update cadence.

When to Consider This Package

Consider it when:

• You're on Laravel and the package's supported PHP/Laravel range matches your stack — verify this first; a 2023 release may not cover Laravel 11/12 or PHP 8.3+.
• Moderation needs are word-list driven and English-first: obvious profanity, banned terms, brand-safety lists. No contextual nuance required.
• Data residency or self-hosting is a constraint (health, finance, EU-only, air-gapped demos), which rules out SaaS moderation APIs.
• Volume is low-to-medium and latency budget is generous — in-process filtering beats a network round-trip.
• You want a 2–5 day spike to validate demand before committing engineering or procurement effort.
• You're comfortable owning a fork: the MIT license means you can patch, extend the word list, and pin it internally without legal friction.

Look elsewhere when:

• You need contextual or semantic moderation — sarcasm, implicit harassment, threats, intent. Word lists don't do this and will produce false positives ("Scunthorpe problem") that erode user trust.
• Multilingual coverage is required. Word-list packages age badly across languages and scripts.
• Adversaries are motivated: leetspeak, homoglyphs, spacing tricks, and image/emoji abuse all defeat naive matching. Public-facing teen or gaming communities should not rely on this alone.
• You need audit trails, appeal workflows, reviewer queues, or SLAs — that's a platform, not a package.
• Dependency hygiene is non-negotiable (SOC 2, regulated environments). An unmaintained package with 25 stars is a finding in a security review unless vendored and reviewed.
• Maintenance is dead. Check commit activity, open-issue response, and whether the maintainer responds to Laravel major-version PRs. If not, treat it as a code sample, not a dependency.

How to Pitch It (Stakeholders)

For executives (one paragraph): We can add basic automated language filtering to user-generated content in about a week using an MIT-licensed, self-hosted Laravel package — no vendor contract, no per-request cost, and no user text leaving our systems. It covers the "obvious bad words" case that satisfies store-review and procurement questions, and it lets us measure how often users actually trigger filtering before we spend on a real moderation platform. The caveat is maintenance: the package is small (25 stars) and last released in April 2023, so we'd wrap it behind our own interface, own the word list, and treat it as a one-quarter bet with a clear exit to a vendor if abuse volume or nuance demands it. Ask: approve a one-week spike, not a permanent commitment.

For engineering (the decision that matters): Don't call the package directly from controllers or models. Put it behind a TextRefiner contract with one adapter for this package and one for a future vendor/no-op. Pin the version, review the source (it's small), verify PHP/Laravel compatibility against our matrix, and add a config-driven word list we control in-repo so tuning doesn't require a package release. Log every filter hit with the matched rule and surface — that telemetry is the input to the next buy/build decision.

The honest framing: this is a cheap, reversible experiment with a known maintenance tail, not a moderation strategy. Value comes from the option it creates — ship something defensible this quarter and buy real signal on how much moderation we actually need.