Changelog
What shipped, by version.
Factual entries of what changed in the product. One entry per release, capability by capability. No roadmap promises, no marketing language.
v0.8
- Public blog with an in-app editor: write, schedule, and publish posts.
- Embeddable website capture form: a buyer states requirements on the producer's own site and it becomes a drafted, rep-reviewed inquiry. The buyer never sees the recommendation.
- Approval attestation: an immutable record of who approved every reply, and an exportable report confirming none were sent without a person.
- Adoption view for a customer's own managers: who is using GRADE, by week, with a named list of who has gone quiet.
- Separated experiences for reps, customer admins, and GRADE-internal operations.
- Public try-it page: run a real inquiry against a demo catalogue and see every product evaluated, each elimination with its source line, and the unknown column.
- Screening fix: a composition stated as PFAS-free is no longer mistaken for one that contains PTFE.
v0.7
- Inbound email intake: forward a customer's email to a unique address and it becomes a review-queue inquiry. GRADE never emails the customer back.
- In-app email-forwarding setup guide (Outlook and Gmail) with a one-click self-test.
- Public system status page with live status and uptime.
- Bounded in-app support assistant: answers how-to-use questions from our own documentation only; refuses chemistry and suitability questions.
- First-run onboarding checklist for new users.
- Admin: in-app user management and email-intake controls; refreshed navigation.
- Secret scanning moved to a pinned, self-contained binary so a provider outage can't block a release.
v0.6
- Multiple business units within a producer: shared or split catalogue, per-division reporting, tenancy still absolute at the producer.
- Reports: portfolio penetration (which grades never surface), demand intelligence (constraints nothing satisfies), and a response-time tracker.
- Expert-knowledge product notes, attributed and kept separate from data-sheet values and never used in screening.
- Live run-state streaming with real product counts and honest failure states.
- Daily off-site, encrypted database backups with an in-CI restore check.
- Deterministic screening hardened: 1,000,000 generated cases run through the screen with zero invariant violations, and — as a separate check — the paid model eval fails the build on a single false pass, in CI and nightly.
v0.5
- Match a competitor product (cross-reference mode).
- Upload a PDF, DOCX, or TXT inquiry.
- Report a problem and flag a wrong answer from any run.
- Team management, invites, and per-seat access.
- Versioned API with a request ID in every error.
v0.4
- Home dashboard and left navigation.
- The Queue: every inquiry tracked from drafted to resolved.
- Record what a customer bought — measured accuracy on Proof.
- Screening invariants proven across a million generated cases.
- Optional dark mode.