PUBLISHED RUBRIC
Methodology
A tool is scored per goal, because it can be a 90 for voiceovers and a 60 for voice agents. The number lives on the goal, not on the tool. This page is the method. The numbers on goal pages in v1 are provisional placeholders that use the method, not measurements from it.
What v1 actually computed
Method version seed-v0-provisional. Every published score has confidence low. None were hands-on tested. Subscores are editorial seed values so the ranking UI, the sort, and the sponsorship separation can be reviewed. They will be replaced when the scoring pipeline runs. A low-confidence score is shown as a band of ±8 around the point estimate. That band is a display convention, not a statistical interval. Order uses the point estimate.
Entry prices checked on Oct 16, 2026 are stored for display: free plan, cheapest monthly USD, billing basis, status, pricing URL, and the date checked. Rows that could not be read say Price not verified. Those dollars are not subscores, and the ranker does not import them. Launch dates and most changelog rows are still examples. Where a date exists so a badge can render, the row says Example. Affiliate research from Oct 14, 2026 is stored beside the catalog and is not passed into the ranker.
The score
score = Σ(weight × subscore) / 100, then subtract penalties
Each subscore is 0–100. Weights sum to 100. Penalties are shown on the row.
| Criterion | Weight | What it means |
|---|---|---|
| Goal fit | 25 | How fully the tool covers the stack role for this goal, against that role's must-have checklist. |
| Output quality | 20 | Quality of the result: code that runs, voice naturalness, video fidelity, mesh topology, or the equivalent for the role. |
| Ease for vibe coders | 15 | Time to a first useful result, setup steps, no-code options, docs, and whether an agent can drive it. |
| Pricing and free tier | 15 | Whether a free tier is useful, how predictable the entry price is, and what a reference workload costs. |
| Integrations and API | 10 | Public API, SDKs, webhooks, MCP, native links to other stack roles, and how hard it is to export. |
| Momentum | 10 | Shipping velocity over 90 days. Log-scaled and capped so a hype spike cannot dominate the score. |
| Community sentiment | 5 | What recent users say, after deduping and down-weighting brigading. Net sentiment, not a star average. |
Anchors for a subscore
- 0Misses the role. Unusable for this goal, or discontinued.
- 25Covers a sliver of the role. You would still need another tool to finish.
- 50Works, with clear gaps a vibe coder will hit in the first session.
- 75Covers the role well. Gaps are edge cases, not the main path.
- 100Covers the role end to end for the reference workflow.
Penalties, confidence, stack score
- Subtract 5 to 15 for an unresolved security incident, unclear commercial rights, a dark-pattern billing pattern, or an announced deprecation. The reason is printed on the row. The seed applies −8 on Veo for the preview shutdown window, and −12 on Udio because downloads are disabled during a licensing transition.
- Discontinued tools, Sora included, are hidden from rankings and marked Retired.
- Confidence is high, medium, or low from data coverage and whether anyone tested it. v1 is entirely low.
- The stack score is the mean of the role winners, plus a compatibility adjustment of −5 to +5. It is shown apart from tool scores. v1 compatibility notes are labeled examples.
- Weight changes get a new method version. Score changes get a changelog with a reason. On Oct 16, 2026 the pricing subscore moved for Cursor (66 to 62), Writesonic, Surfer, Unity, Superhuman, Attio, Clay, and Elicit after the price check. PlayHT is retired and Cartesia takes its place on the voiceover page. The formula did not change.
Sponsorship versus rankings
- Ranks are computed only from subscores and penalties. The ranker cannot import the commercial file. A test checks that.
- Affiliate status never sets the order. A tool with no program can be first. On the weekend SaaS page, Vercel leads hosting over Railway, which has a program.
- Paid listings are labeled Sponsored, at most one per goal page, visually distinct, and never inside the ranked list or the stack.
- The $49 fast track buys review speed only. It does not buy a score or a place in a stack. Payments are not connected in v1.
- No buyable badges. StackFor Pick and Verified would be earned. v1 does not award them.
- A newsletter sponsor is labeled Sponsor. Editorial picks in the same issue follow the scores.
- Rank changes are meant to be logged with their inputs. A quarterly note would put sponsor revenue next to rank stability. v1 has neither revenue nor a revision log.
- Vendor corrections use the public submit form. Facts can be fixed. Scores move only through the pipeline.
We may earn a commission if a tracking link is later connected and you buy. Every v1 tracking URL is empty, so /go/<tool> redirects to the vendor site. See also the submit page.
Freshness
100 × e^(−days since last substantive release / 45)
A major release multiplies that by 1.2, capped at 100. Freshness is shown next to the score and is not added into it. Badges, as of Oct 14, 2026:
- New tool: first seen under 30 days.
- Shipped this week: a non-deprecation, notable or major note under 7 days.
- Rising: momentum subscore up at least 10 over 30 days. v1 stores two example deltas, labeled as examples.
- Stale: a release on file older than 180 days. No release on file is not stale. A pricing page that changed and was not re-checked would also be stale once that tracker exists.
Only notable and major notes feed a goal's “what's new” module and the newsletter. A major release or a pricing change is supposed to re-queue that tool's scores within 48 hours. The pipeline that would do it is not running in v1.
What the pipeline is supposed to read
Goal fit from the checklist and vendor docs. Output quality from a fixed prompt pack and public benchmarks. Ease from a timed signup. Pricing from page snapshots. Integrations from API docs and automation catalogs. Momentum from changelogs and package or GitHub deltas, log-scaled. Sentiment from recent public posts, deduped. The scorer does not read affiliate tables, sponsored placements, or the internal fields that decide which goal to build first.
Prompt packs are not published in this seed because no test has been run. When they exist, they belong on this page next to the weights.