Decision workflow

The six steps

The strip above lives on every selection surface (/select, /compare/build, every use-case detail page) so the workflow is consistent wherever a reader lands.

Why no ranking

Rankings invite bad decisions. Different workloads weight context window, output limits, modality, pricing, hosted availability, and freshness completely differently. A score that mixes these into one number imports an opinion the catalogue does not have — and could not justify with primary sources.

WebmasterID Models therefore exposes verified fields per model and lets the reader weight them. The shortlist order on /select is deterministic and documented (verified field count → active lifecycle → source count → name) — it is intentionally not described as a ranking.

Why use cases come first

Filters only mean something inside a use case. A 1M-token context window does not matter if the workload is a chat assistant. A hosted-pricing reference does not matter if the buyer needs a first-party support contract. /use-cases names the verified fields that matter per workflow and links straight into the selection workspace.

How shortlist order works

1. Verified field count desc — well-sourced records first.
2. Active lifecycle first — deprecated/retired records last.
3. Source count desc — more citations beats fewer.
4. Model name asc — stable tiebreak.

The ordering is deterministic at build time. There is no score, no rank function, and no weight on opinion. A reader who wants a different ordering should apply use-case filters; the ordering does not change in response to unrelated signals.

How the comparison builder works

The builder at /compare/build renders 2–4 selected models side by side using verified fields straight from the typed local data layer. It is a workspace, not a separate indexable page — generated query URLs carry noindex, follow so each comparison is reproducible by URL without spamming the index. Static, curated comparison pages still live under /compare and follow a higher editorial bar (two-sided verified before indexing).

Selected models render in the order provided by the query string. Unknown values render the canonical unverified-data label; nothing is inferred or derived. The data-gaps matrix makes missing fields explicit per model.

How data gaps affect decisions

A data gap is a field the catalogue records as unverified — either the vendor does not publish it, or automated retrieval is blocked (the OpenAI 403 case). Gaps are not guesses; the catalogue refuses to invent them.

A reader should treat gaps as part of the decision. If a model is otherwise attractive but has an unverified output limit, the safe move is to open the vendor docs and confirm the limit before running a workload that depends on it. The reverification queue surfaces every record currently due for manual re-check.

How to use sources and freshness

Every verified field is paired with a primary-source URL and a retrievedAt date. Freshness states (fresh / review due / stale / blocked / unknown) are computed deterministically against siteConfig.buildDate. A row marked stale is a row whose source has not been confirmed for longer than the policy window — not a claim the value is wrong.

The conservative read: re-open the source URL before any decision that depends on the value. Pricing references in particular are intentionally not framed as live quotes; re-verify against the vendor pricing page before procurement. Cost projections need workload-specific arithmetic that the catalogue does not perform.

What the platform does not decide for you

• Which model is "best" for your workload.
• Which provider is "cheapest" — pricing is a reference, not a ranking.
• Which model is "fastest" — latency / throughput claims are not asserted.
• Which comparison declares a winner — comparisons render verified fields, never deltas or rank.
• Whether a model meets a specific compliance regime — verification status describes citations, not regulatory posture.

The reader runs the workload, weighs the data gaps, opens the vendor pages, and decides what to test externally. The catalogue is the evidence base, not the verdict.