Kamin changelog

30 September 2026 · Task loop, daily experience, governed catalog and verifiable credentials

• Task → evidence → review loop: six tasks for three pathways, rubrics, saved statuses, and explicit separation of completion, evaluation and verification.
• Today strip on the dashboard: my goal, most important missing evidence, what I do now. Compare two or three pathways without a readiness percentage.
• Governance for every catalog relation (source, version, review date, owner, validity scope), explicit ESCO/O*NET/SSCO roles, and a local-opportunity schema shipped empty.
• A local capability suggester measured on a frozen labelled set, with the supporting sentence and abstention.
• Signed-credential import verified locally (structure, issuer, eddsa-jcs-2022 signature, validity, status). The trusted-issuer registry is intentionally empty.
• The operable advisor-review design is published in the repository with separation tests; it is not a working feature.

30 September 2026 · Trust level per field and document signals

• Provenance per field: extracted values are kept; edits and raised grades are named and turn the capability into self-declared on cards, shared snapshots and exports.
• Local document-integrity signals: digital signature and its coverage, later saves, producing software, modification date, and a QR code with a verify-at-source link.
• From gap to evidence: record a project, certificate or internship against a gap as self-declared applied evidence awaiting review, without changing any judgment.
• Unsigned portable export in the CLR 2.0 / Open Badges 3.0 shape from Privacy.
• Transcript parser fix: a title containing a number is no longer split and lost. New red-team checks for upload and the SPARQL lab.

30 September 2026 · Recommendation review and local contests

Added per-recommendation acceptance checks, local decisions under a name or alias, withdrawal and history, and fresh review after context changes or restoration. Encrypted backups include reviews; contests export test candidates without profile data. Decision time and escaped-error measures are study targets, not results.

30 September 2026 · Pre-pilot audit fixes

• One audited (light) theme across public pages and the student workspace; the incomplete dark theme that hid headings under an operating-system dark preference was removed. An automated WCAG 2.1 AA contrast gate now runs in both schemes.
• Dated release notes appear only on the changelog, not inside the privacy policy, trust center, guide or services pages.
• A security contact is published per RFC 9116 at /.well-known/security.txt, and the installed app opens in the language the student last used instead of forcing Arabic.
• Fixed the gold “Evidence” label contrast in skills and removed a preload for a logo the landing page never renders.

30 September 2026 · Team and hosting readiness

Updated live service descriptions, published the capstone and evaluation plan, and prepared document routes and search metadata for Cloudflare Pages. Hosting remains on Render; preparation is not a completed migration.

Research and sharing tools · 30 September 2026

SPARQL lab: error locations when available, readable labels, JSON/CSV downloads and Copy as CSV. Export the dataset as a Turtle union graph, or TriG to preserve named graphs and inference provenance.

Your interests: opt-in multilingual semantic recall runs locally after a model download. A suggestion adds no skill and changes no evidence. Inspect each pathway through the graph and rules.

Privacy: select capabilities or a goal, preview the snapshot, then create an encrypted link. The link includes the key; its holder can open and copy it, and it cannot be revoked after sending. This is manual sharing, not institutional integration or a verified credential.

29 September 2026 · Usability improvements

• Consistent navigation, guide and services on every page.
• Visible interest and goal choices, plus a relationship-network tour.
• Recommendations lead with context and an actionable next step; comparison has a useful starting point.
• Explicit saving failures and recovery guidance.
• Shareable Arabic and English pages and clearer team and help information.

27 September 2026 · Relationship network

Multi-step connections between people, interests, evidence, capabilities, work and learning, with synthetic data, sources and inference limits distinguished.

Privacy in this update

This update does not enable usage analytics or central collection of student profiles. Each purpose keeps separate consent. Review the policy and its history before adding information.