Method · pass of 30 Oct 2026
How this index is built, and where it stops.
Most directories ask you to trust a list. This one shows its work: the number of pages it tried to read, the ones that refused, the candidates it threw out, and the exact rule that decides whether a page is allowed into search.
The pipeline, in six steps
01 · Discover
Candidates come from a hand-curated seed of named products and from public registries: the GitHub search API and the Hugging Face hub. This pass: 321 candidates from 0 curated names and 0 registry repos.
02 · Read
For each candidate we fetch the official page: robots.txt first, then one request at a time per domain, with a crawl-delay floor. We store the HTML text, the title, the meta description, the JSON-LD and any price string we can see.
03 · Classify
The capability is computed from the product's own words with a documented keyword rubric. A low-confidence match is flagged for review instead of being published into a category.
04 · Score
The score is a fixed rubric, documented below. It measures how much we could verify, then adoption. It is not a user rating.
05 · Publish
A row enters the sitemap only if it passes the indexability rule in step 6. Everything else is reachable but noindex.
06 · Re-check
The daily crawler re-reads every URL, stores a hash of the readable text, and flags price or positioning changes rather than assuming they did not happen.
Four statuses, never one badge
A single “verified” tick is a marketing device. Every claim on a product page carries the status of that claim and the date it was checked.
- verified
- We read it ourselves on the operator's own page, and the quote is kept next to the claim.
- provider_claim
- The operator says it, we quote it, and we have not independently tested it. A one-liner is always this.
- unknown
- We looked and did not find it. Shown as n/v — never filled with a guess.
- blocked
- robots.txt or a bot wall stopped our checker. The row stays, labelled, and is not indexed unless another source carries the facts.
The score, in full
Published so you can disagree with it. A high score means “we could verify a lot about this product”, not “this will work for you”. Nothing in the formula can be bought: there is no affiliate term anywhere in the codebase.
| Component | Points | Why it is there |
|---|---|---|
| Base | 30 | Every published row starts from the same floor; the score differentiates evidence, not existence. |
| Readable official page | 12 | We could read the page ourselves rather than relying on a registry blurb. |
| Sourced one-liner | 10 | The maker published a description we can quote. |
| Price we could read | 8 | A price is the most-checked fact in any directory and the easiest to fake. |
| Platforms (≥2) | 8 | Two stated platforms outrank one; one outranks none by 4. |
| Launch date | 4 | The product says when it started — a proxy for a maintained page. |
| Free tier stated | 6 | A reader can try before paying. |
| Features (≥3) | 6 | Discrete capabilities we could name; one or two earns 3. |
| Adoption | 0–16 | log10 of GitHub stars × 2, capped at 10 points, then weighted ×1.6. Only where a registry number exists. |
Capped at 99. Rows whose page could not be read start lower and are labelled — that is the point.
Coverage report
- Products published
- 185
- Allowed to index
- 119
- Pages read
- 287
- Refused / unreachable
- 34
Sources that refused our checker (34)
Listed by name, because “we cover everything” is a claim we cannot make honestly. Blocked pages are not scraped around: we record the refusal and move on.
- chatgpt chatgpt.comblocked
- claude claude.aiunreachable 403
- gemini gemini.google.comunreachable
- openai-codex openai.comunreachable 403
- lovable lovable.devunreachable 403
- genspark genspark.aiunreachable 403
- midjourney midjourney.comunreachable 403
- ideogram ideogram.aiunreachable 403
- gamma gamma.appunreachable 403
- quizlet quizlet.comunreachable 403
- scispace scispace.comunreachable 403
- whoop whoop.comunreachable 403
- withings withings.comunreachable 403
- ray-ban-meta ray-ban.comunreachable 403
- halo-neuro haloneuro.comunreachable
- timekpr rewind.aiunreachable 403
- character-ai character.aiunreachable 403
- sana sana.aiblocked
- sanctum sanctum.aiunreachable
- magic-notes magic.xyzunreachable
- tradingagents-gh arxiv.orgunreachable 200
- ai-agents-for-beginners-gh aka.msblocked
- tts-gh coqui.aiunreachable 404
- claude-code-best-practice-gh linkedin.comunreachable 999
- milvus-gh milvus.iounreachable
- cosyvoice-gh funaudiollm.github.iounreachable 404
- amphion-gh openhlt.github.iounreachable 404
- memobase-gh memobase.iounreachable 404
- meta-ai meta.aiblocked
- le-chat chat.mistral.aiunreachable 403
- scite scite.aiunreachable 403
- galaxy-ring samsung.comunreachable 404
- oakley-meta oakley.comunreachable 403
- marek-health marekhealth.comunreachable 403
Candidates we excluded (54)
Real, harvested candidates that are not human augmentation: SDKs, vector databases, prompt packs, tutorials and code samples. The reason for each is committed in data/seeds/exclude.json.
- nginx-ui web server admin UI, unrelated to human augmentation
- unity-mcp MCP bridge for a game engine — infrastructure
- transformers model library, not a product a person uses
- unsloth training/fine-tuning framework
- copilotkit React framework for embedding copilots
- anti-slop prompt rules for coding agents, not a product
- a11y-md rules file for coding agents
- shadscan developer audit CLI
- rocketsimapp-gh iOS simulator tooling for developers
- docmost enterprise wiki — knowledge tooling, not augmentation
- kalosm-gh Rust library for local LLM workflows
- opensquilla-gh unclear product; page and repo describe different things
- claude-real-video-gh unofficial clone, not the maker's product
- ponytail-gh prompt/skill pack
- oh-my-openagent-gh prompt skill pack
- tapq-gh voice layer for developers' agents — infrastructure
- opendroid-gh Android automation agent; page is a demo, category unclear
- ghost-os-gh computer-use driver for agents — infrastructure
- beav-gh Chinese social-media asset workbench; scope unverifiable
- lime-gh Chinese writing tool; no readable English product page
- jev-chat-jarvis-gh reply-suggestion utility with unclear data handling
- holaos-gh service-business agent platform — B2B infrastructure
- agent-gh generic name, page describes a different product than the repo
- noi-gh landing page has no description of what it does
- codealmanac-gh codebase documentation tool for teams — infrastructure
- clawd-on-desk-gh novelty desktop pet
- paseo-gh launcher for other coding agents — infrastructure
- codecompanion-nvim-gh Neovim plugin, library-shaped
- basic-pitch-gh audio-to-MIDI model, a library
- mempalace-gh memory library with benchmark claims we cannot verify
- agentmemory-gh persistent memory for coding agents — infrastructure
- gentle-ai-gh stack for coding agents — infrastructure
- deer-flow-gh agent framework (LangChain-based)
- deepseek-harness-gh agent harness — infrastructure
- firecrawl scraping API for agents — infrastructure, not a person-facing tool
- rea-gh reverse-engineering agent; page and repo disagree on scope
- mimiclaw-gh novelty firmware demo
- ai-ethos-gh nonprofit learning initiative, not a product
- baocut-gh video agent with no pricing or platform information
- halliday the URL we had resolves to an unrelated crypto company; glasses product not verified
- canva page served a bot wall; no usable description
- espnet-gh research toolkit documentation
- llms-from-scratch-gh book companion repo
- hello-agents-gh placeholder repository description
- tradingagents-gh non-HTML page, no readable description
- edge-tts-gh TTS library, not a product
- ollama-gh A local model runner: it serves weights, it does not augment a human capability. Inference infrastructure is out of scope by rule.
- chattts-gh A TTS model implementation. Model weights and inference code are infrastructure by our own rule, not a product a person adopts.
- mlx-audio-gh Audio model runtime/toolkit (MLX). Inference infrastructure.
- openvoice-gh Voice-cloning model implementation. Research weights, not an adopted product.
- sherpa-onnx-gh Speech recognition/synthesis runtime. Inference infrastructure.
- vits-gh End-to-end TTS research model. Academic repo, not an adopted product.
- react-email-editor-gh A React component library. Component code is infrastructure by our own rule.
- openclawchinesetranslation-gh A translation fork of another project, not a distinct product. Listing it would double-count the upstream row.
Corrections
If a fact here is wrong, the fix is a source, not an opinion. Send the URL that shows the correct price, platform or description and we will re-read it, update the row and bump its check date. Products cannot be removed by their owners for being unfavourably scored; they can be corrected, or marked as shut down, with a source.
185 products · 119 indexed · pass 30 Oct 2026 · rebuilt from raw artefacts by scripts/coverage.mjs