Next test: Accessibality Drift (a11y-drift)

Prepared 2026-09-18. Claim tsk_20260918034715_ja0rj. Task kind: research-rollout-2026-09-17:next-test.

Source: ventures/a11y-drift/screening.md and screening.json (dated 2026-09-18, decision revise — not proceed, not stop) plus research/keyword-research-2026-09-07/REPORT.md and brief.md. No new paid research calls were made for this task; it draws only on evidence already on disk.

Riskiest assumption (from the screening)

The screening's decision is revise, not stop, so this is a next-test plan, not a stop recommendation. The screening identifies two blocking unknowns, but they are not equally cheap to resolve. Priority:

**"A B2B SaaS team that has just been asked an accessibility question by a

customer (VPAT / EN 301 549 / procurement questionnaire) wants an ongoing

CI/PR-diff trend subscription, rather than a one-time completed answer to the

question in front of them right now."**

This is the assumption named as the idea-killer in both brief.md's 2026-09-04 correction and screening.md's decision section. It is cheaper to test than the other blocking unknown (the size of the reachable EU buyer population, which needs either a paid B2B database/list-building pass or a much longer organic-discovery period) because it can be tested directly with a handful of real conversations before any enumeration work, and a clear "no" here makes the market-sizing question moot.

The two named trigger search terms (accessibility questionnaire, accessibility procurement) have ~10/mo US search volume each — evidence that this buyer moment is not reliably reachable through SEO/content at all, which independently constrains what channel this test (and any future one) can use.

The test

Behavior being measured: does a real target-persona contact, shown a manually prepared example of the actual deliverable (a filled-out conformance record / VPAT-style answer for a plausible product), say they would pay an ongoing monthly fee for a trend/drift service — versus wanting only a one-time completed document?

This is a "sell before you build" pretotype per the charter's pretotyping policy: no GitHub App, no landing page, no ad spend. A manually prepared artifact and direct conversations.

Target customer: an accessibility-adjacent or engineering lead at a

20–80 person B2B SaaS company selling into EU enterprise, university, bank, or public-sector customers. Recency filter: has personally handled a VPAT, EN 301 549, or accessibility-related security-questionnaire line item in the last ~6–12 months (90 days is likely too narrow to source enough contacts; widen the recall window rather than inventing volume).

Channel: not cold outbound (charter no-go list) and not posting (agents

may not post per the charter's publish-vs-post distinction). Use the cssbed GitHub repo's existing audience/contributors and any warm, already-connected accessibility/dev-tools communities Yuval can reach directly, one-to-one, as a genuine question — not a pitch, not a public post. Concretely: a short list of specific named people or maintainers Yuval (or someone he asks) already has a real connection to, asked directly. If no warm list of at least ~8–10 qualifying contacts exists, that itself is a finding (see thresholds).

Price assumption carried into the test: $350/mo, exactly as recorded in

brief.md and screening.json — this is the existing recorded proposal under test, not a price Yuval has approved for display or quoting. The conversation should surface it as "if this were $X/mo" framing, not as an offer to transact; no payment should be collected or requested in this test.

Costs: researcher/Yuval time only to prepare one example artifact

(~2–4 hours) and hold 6–10 short conversations (~20–30 min each). No paid tooling, no ads, no lists purchased. This stays inside the $25 S0A research cap and does not require new spending authority — it is conversation labor, not a paid channel.

What "prepared" means concretely: one realistic, redacted example

conformance-record/VPAT answer (using a hypothetical or genuinely public open-source project's site, not a real prospect's confidential data), built by hand or with axe-core output, formatted the way the eventual product would present it.

Thresholds

Sample size: aim for 6–10 completed conversations with qualifying contacts (exact number depends on how many are reachable through the warm-channel constraint above — do not force volume by going cold to hit a number).

BUILD (promote toward S1 registration): at least 3 of the qualifying

contacts, after seeing the example artifact, say unprompted or when asked that they would want this kept current / updated over time (not just a point-in-time document), and at least 1 gives a specific reason tied to a recurring need (repeat questionnaires, ongoing customer audits, a compliance renewal cycle) rather than "nice to have."

ITERATE: responses are mixed or conditional — e.g., contacts want the

document but are unsure about paying monthly for it, or want it bundled into something else, or the $350/mo number gets pushback but a different price or packaging (one-time fee, annual, bundled with the free GitHub Action) gets interest. Revise the offer/price/packaging and re-test with a fresh small sample rather than building.

STOP: fewer than 8–10 qualifying contacts can be reached at all through

a non-cold, non-post channel within a reasonable window (this would confirm the search-volume evidence that the buyer moment is not discoverable/ reachable), or a clear majority say they only ever wanted the one-time answer and have no interest in tracking drift afterward — i.e., the offer and the trigger are confirmed misaligned, exactly as flagged in brief.md's 2026-09-04 correction.

Exact owner decision needed from Yuval

Approve running this specific conversational pretest (no page, no

spend, no App) — i.e., confirm the Scout/Steward may spend the ~2–4 hours of prep plus a handful of warm conversations against the existing $25/ 7-day S0A cap, since this is still pre-S1 research, not S1 spend.

Name or approve the warm-contact list — the plan depends on contacts

Yuval or someone he trusts already has a real relationship with (e.g. through cssbed); the system should not originate cold contact. Yuval needs to either supply names/introductions or confirm none exist yet (which would itself trigger the STOP threshold above).

**Confirm the $350/mo figure may be spoken aloud in a hypothetical framing

during these conversations** — it remains an internal planning assumption, not an approved public price; Yuval should decide whether it's fine to test verbally in this narrow context or whether a different number should be floated first.

No stage change, price change, spend, campaign, or outreach has occurred as part of writing this plan.