Badges, labels & markers

Every card carries a compact row of typed markers. This page is the reference for what each one means and the conventions behind it. The code is authoritative — the free-tier limitations live in src/data/generosity.ts and the quality labels in src/data/quality.ts; this document mirrors them and defines the rendering rules.

Rendering conventions

* after a number = CALC. A trailing asterisk means the number was computed by the pipeline, not observed. RAW numbers render plain and carry a proof URL. See How ratings are computed for the RAW / CALC / PARAM principle.
Access bucket — free, freemium, or paid. Derived from pricingModel plus the typed limitations, never hand-set. The derivation rule is in How ratings are computed §3.
Achievement badges are short headline strings prefixed by a sentiment glyph: green (positive), orange (caution), red (negative). Examples: a green “Free forever”, a green “Frontier models”, an orange “Op fees begin Oct 2026”.
Commercial-use badge is mandatory and always shown: green when commercial use is allowed, red when it is forbidden.

Free-tier limitations (infrastructure)

Source: src/data/generosity.ts (limitationTypes). Each limitation carries a class that drives the access bucket: barrier and finite move an offer into freemium; quality stays in-bucket as a caveat; renewable is normal and low-impact. The full prose definitions and the bucket rule are in Free-tier conditions & limitations.

Icon	Type	Label	Class	Severity
⏳	`time_limited`	Time-limited	finite	high
🔋	`total_volume_cap`	Finite pool	finite	high
🔄	`quota_renewable`	Renewable quota	renewable	low
🔁	`rate_limit`	Rate-limited	renewable	low
💳	`requires_card`	Card required	barrier	medium
💰	`requires_min_balance`	Minimum balance	barrier	medium
🔒	`requires_eligibility`	Eligibility gate	barrier	high
🧩	`feature_restricted`	Feature-limited	quality	medium
🐢	`performance_throttled`	Throttled	quality	medium
💤	`cold_start_sleep`	Sleeps when idle	quality	medium
1️⃣	`single_resource_cap`	Single instance	quality	low
🚫	`no_commercial_use`		quality	high
🏷️	`attribution_required`	Attribution	quality	low

Quality labels (software / APIs / agent-tools)

Source: src/data/quality.ts (qualityLabelTypes). Outside infrastructure, the software, APIs, and agent-tools sections are governed by a separate quality model, not the value model. These labels are factual flags, not penalties.

Icon	Type	Label	Sections	Severity
📦	`archived`	Archived	software, agent-tools	high
🕸️	`inactive`	Inactive	software, agent-tools	high
🌱	`young`	New project	software, agent-tools	low
⚠️	`unlicensed`	No license	software, agent-tools	medium
🅒	`copyleft`	Copyleft	software, agent-tools	low
🏠	`self_host`	Self-hosted	software	info
☁️	`managed_service`	Managed option	software	info
🏋️	`heavyweight`	Heavyweight	software	low
🔓	`no_auth`	No auth	apis	info
🔑	`key_required`	Key required	apis	low
🪪	`oauth`	OAuth	apis	medium
💳	`paid_only`	Paid	apis	high
⏳	`trial_only`	Trial only	apis	medium
✅	`official`	Official	agent-tools	info
👥	`community`	Community	agent-tools	info

Per-category rating axes

The value model produces one $-denominated headline rating per bucket (see How ratings are computed). Categories also surface secondary display axes — these are shown for context, not as separate top-line ratings:

LLM providers — the headline is the Overall value rating; the breakdown surfaces Volume (free tokens/mo), Artificial Analysis (best free model), and Throughput (tokens/min). Concurrency is a secondary “how many agents in parallel” signal, not a volume multiplier.
Software / agent-tools — quality axes: adoption, activity, maturity, community, capability (qualityComponents).
APIs — quality axes: free access, agent relevance, reliability, developer experience, openness.