Changelog

What’s new in SkillCort

Everything here is live in the product. We ship fast and write it down.

56 improvements shipped in 2026

Shipping activity

JanFebMarAprMayJunJulAugSepOctNovDec

LessMore

Early August 2026

AI-assisted proctoring review, a real scoring desk, and blueprint-aware role fit

Reviewers got AI eyes and ears — with a human always deciding — scoring became a purpose-built workspace, forced-choice SJT items joined the bank, and the platform's privacy and retention commitments went from words to running code.

IntegrityAI-assisted proctoring review, staff-triggered and never automatic: webcam snapshots are scanned frame by frame with an expanded set of observation flags, screen captures are checked for outside-help windows, and the microphone recording gets a transcript-level triage. Every flag lands in front of a human reviewer — nothing becomes a verdict on its own.
IntegrityCohort-level similarity analysis surfaces possible collusion signals across candidates who sat the same assessment, a unified Logs timeline reconstructs each session from stored facts, and the authenticity report exports as a fully branded PDF — your organization's logo, colour-coded findings, and gated until the checks have actually run.
IntegrityProctoring and scoring are now two separate desks: the Proctoring tab is purely about session integrity, and all scoring happens on Results — so integrity review never blurs into grading.
ScoringA dedicated scoring workspace: an evaluator grading queue, questions in section order with per-question score pills, rubric-anchored entry with strict range validation, optional human override of auto-scored answers (the auto score is never mutated), a blind-scoring toggle for written work, and an explicit Ready-to-finalize → Finalized gate that Results waits for. Overall scores read 0–100 everywhere.
ReportsA standard candidate report with visible entry points, and an answer sheet: the candidate's exam paper — session, participant, and result blocks with a section-by-section breakdown — as a tab and a downloadable PDF.
AuthoringForced-choice situational judgment (SJT) items: statement pools with per-statement competency mapping, option use-caps, and weighted or threshold scoring — with per-competency roll-up into role fit. Open-text answers gained a hard server-enforced word limit.
AuthoringPDFs embed directly inside questions: candidates open a document card into an inline viewer, and large files open fast through on-demand page loading. The editor's candidate preview now renders the real exam card, type by type — what you see while authoring is what candidates get.
ItemsThe item bank reads its own quality signals inline: separate difficulty and discrimination chips per item, and the “Needs attention” reason spelled out on an amber chip. Rubrics gained a manager — search, bulk actions, standalone creation, and rename/edit/delete from the bank, plus AI-drafted rubrics from the question content.
Blueprints49 ready-made role blueprint templates seeded into every workspace, AI assist that drafts a blueprint or calibrates target levels — suggestions with stated reasons, always applied by a human — and a blueprint-aware fit model: the headline score reflects the blueprint, critical competencies act as a hard gate, and coverage is visible wherever fit is read.
InterviewsLive realtime interviews now offer a presentation choice per interview: a talking avatar or voice-only with an audio-reactive orb, with automatic fallback from avatar to voice and an eight-face interviewer gallery. Live simulations get the same choice, and previews run the real live experience.
CandidatesCandidate email got a paper trail: a searchable, paginated history of every send, per-assessment email template selection, bulk sends from a chosen saved template, and one-click invitation resends — with skipped rows explained instead of silently reported as sent.
TrustThe legal policy set went into force — an evidence-based privacy notice, cookie policy, and terms, published only once every claim matched deployed reality. Candidate data retention is now executable, not aspirational: a 24-month default (organization-configurable) enforced by daily sweeps, full data purge on organization offboarding, and the candidate AI disclosure names the realtime third parties involved.
PlatformThe dashboard moved onto the SkillCort design language: a consistent branded loading state, one breadcrumb pattern product-wide, a responsive sweep across tables and grids, and a simpler assessment lifecycle — draft → live → archived.

Late July 2026

Drag-and-drop question types, a calmer proctored exam, and a sharper editor

Two new auto-scored interaction formats joined the item bank, the proctored exam stopped interrupting candidates, and authoring got a faster type picker plus a stronger writing surface.

AuthoringTwo new auto-scored question types: Drag & drop categories (sort items into labeled buckets) and Fill in the blanks (drag words from a word bank into gaps in a passage, with optional distractors). Both are drag-first with a tap-to-place path for keyboard, mobile, and screen-reader users; answer keys stay server-side and the item pool / word bank shuffles per candidate.
AuthoringCreating an item now starts from a searchable type library grouped by what the candidate does — Selection & Choice, Written & Input, File & Media, Evaluation — with scoring-mode filters and Auto-scored / Human review / AI-assisted badges on every card.
ItemsItem tags became name: value pairs (e.g. Difficulty: Hard, Subject: Math). A question can carry values from several tag names regardless of its folder, filters read “Name: Value” everywhere, and a bulk action applies or removes values across selected questions — including minting a new value on the spot.
Exam experienceProctored exams now ask for every camera and microphone permission up front on the start screen — never mid-exam. Browser permission prompts no longer count as leaving a locked-down exam, lockdown notices moved from a banner above the questions to a quiet corner toast, and every capture device is released the moment the assessment is submitted.
AuthoringScenario / case pages: build an item set around a shared stimulus — the case material stays pinned beside its questions for the candidate.
EditorA stronger rich-text editor: curated font choices, heading and alignment dropdowns, tables with merge/split and unmistakable cell selection, syntax-highlighted code blocks, a full-screen writing mode, and a compact icon toolbar that stays on one row.

July 2026

Access control, self-registration, and defensible validity

Deliveries gained real access control — public sign-up, codes, and credential matching — and the blueprint page turned scores into reliability-qualified validity evidence.

AccessPublic and restricted access per delivery: a shareable link with optional self-registration, single-use invitation codes, or credential matching against a guest list (e.g. employee or student ID) — no per-candidate email required. Every access decision is made server-side, with no candidate enumeration.
AccessAttempt rules per delivery: multiple attempts, resume-in-progress, time-boxed attempts, and one-click retake grants from the candidate list.
AccessA per-delivery access form to capture custom fields at entry, plus a reviewable access log — filters, candidate linkage, and CSV export — of every pre-exam access event.
BlueprintsRole-fit evidence on the blueprint page: how the candidate pool measures against each role competency, with reliability stated per competency (ICC, McDonald's ω) and an honest “n/a” when there isn't enough data to trust a number.
BlueprintsGroup-disparity checks: four-fifths impact ratio and Cohen's d across any candidate group you choose — framed as a signal to investigate, never a legal adverse-impact ruling.
BlueprintsCriterion validity: record real outcomes and see the observed role-fit-to-outcome correlation, clearly labeled preliminary and uncorrected, with optional range-restriction and attenuation what-ifs.
ReportsOne-click validation-evidence export: reliability, criterion validity, group disparity, and a written methodology (what's claimed and what isn't) in a single defensible JSON pack.
EvidenceEvidence Wallet: a per-candidate, auditable dossier — task outputs, scores, evaluator notes, and integrity signals — viewable and exportable as one record.
PreviewPreview as candidate now runs the real AI interview and simulation conversations, not a stub, so authors experience exactly what candidates will before publishing.
BuilderLive assessments are now editable: add, edit, or reorder items without taking the assessment offline. Candidates who already submitted keep their exact results — scores come from what they answered. A staff-only re-score then recomputes objective scores across every candidate after you change an answer key, so your edits and their results stay consistent. Closed and archived assessments stay locked.
AuthoringSelection limits on multi-select and matching questions: set a minimum and maximum number of picks or matched pairs. Candidates see the rule under the question, extra picks lock at the cap, and the limits are enforced server-side.
AuthoringMultiple-choice answer layouts: besides the standard list and column layouts, options can render as inline chips or — for single-select questions — a dropdown menu.
AuthoringGiven cells in table questions: mark any cell as pre-filled content and candidates see it read-only — hand out a partially completed table and ask only for the blanks.
AuthoringTable questions grew up: row and column headers can now be asked as scored blanks, and neighboring cells can be merged into larger regions — enough to model real worksheets, not just flat grids.
CandidateExtended-time accommodations: grant a candidate a per-attempt time multiplier (1×–3×), and every timer in the exam — whole-assessment, section, and page — scales automatically.
CandidateDownloadable resources during the exam: attach reference materials — formula sheets, case documents, data files — that candidates open and download while they work, served through short-lived signed URLs.
CandidateAn on-screen calculator candidates can open during the exam, enabled per assessment — for the roles where the tool, not mental arithmetic, is the point.
DecisionA cohort analytics overview on the Decision Board: score distributions and role-fit spread across everyone who took the assessment, so a batch of candidates reads as a story you can act on, not just a table of names.

Early July 2026

Voice interviews, page-based exams, and granular integrity

A fast month: the AI interviewer learned to talk, exams gained a page structure with per-page timers, and proctoring became precisely configurable.

InterviewsTurn-based voice AI interviews: candidates answer by speaking, transcripts are stored as reviewable evidence, and an adaptive follow-up question is asked back out loud. A typing fallback is always available.
InterviewsThe AI interviewer now appears as a disclosed talking-avatar persona for the opening question — always labeled as an AI interviewer, never imitating a person.
EvaluationTranscript-based AI score suggestions: criterion-level draft scores from interview and simulation transcripts, with model/version provenance. A named evaluator confirms or overrides every suggestion.
ExamsPages inside sections: assessments now structure as sections → pages → blocks, with shuffle available at all three levels (seeded per attempt) for fair parallel forms.
ExamsPer-page time limits: when a page's time runs out the candidate moves on and the page locks, alongside the overall assessment timer.
ExamsA redesigned candidate exam view: sticky progress bar, a question navigator with answered/marked states, page-by-page navigation, mark-for-review, and a pre-submit review summary.
IntegrityGranular proctoring controls per delivery: fullscreen lockdown with configurable violation limits, copy/paste blocking, and second-screen detection — all disclosed to candidates at the consent gate, all reviewed by a human.
ScoringSection scoring weights now flow into role-fit, and multiple-choice options support per-option credit multipliers — including negative marking for high-stakes exams.
BuilderA redesigned builder workspace: outline sidebar, focused panels, and a publish checklist computed from your real assessment data that blocks publishing until required setup is complete.
BuilderBulk authoring: import from the item library through a filterable picker, quick-paste multiple-choice questions, and AI generation with an explicit count — up to 100 questions per run, optionally grounded in an uploaded source document.
ReportsAudit pack export: one self-contained decision file with the assessment structure and weights, plain-language methodology, every score with its named evaluator and timestamp, integrity events with resolutions, and the decision log.
RubricsLevel-based matrix rubrics (criteria × levels with anchored behavior descriptions), criterion weights, and an option to show the rubric to candidates during the assessment.
PlatformOrganization settings (default language, timezone, reply-to email), assessment folders and labels, and side-by-side stimulus layouts for case-style items.

First half of 2026

The foundation

The core assessment operating system, built and verified end to end before we opened the doors.

CoreRole blueprints, the skill & task bank with folders and tags, and the assessment builder connecting them.
TasksWork-sample task types across the board: written responses, file uploads, code, tables, audio and video recording, photo capture, and classic question formats.
EvaluationRubric scoring with multiple evaluators, evaluator notes, and transparent weighted scoring math — evaluators average within a criterion, criteria combine by weight, tasks roll up by skill and section weight.
DecisionsThe Decision Board: candidates compared on weighted role-fit with drill-down from any score into the underlying evidence, plus candidate evidence reports with white-label PDF export.
DeliveryDeliveries with time windows, email invitations, a candidate consent and preflight gate, and proportional integrity signals reviewed by humans.
AIThe AI copilot for drafting tasks, questions, and rubrics — with the founding rule already in place: AI assists, a named person decides.

See it live

The best way to read a changelog is a demo

Bring a role and see the newest features working on your own use case.

Book a demo

AI-assisted proctoring review, a real scoring desk, and blueprint-aware role fit

IntegrityAI-assisted proctoring review, staff-triggered and never automatic: webcam snapshots are scanned frame by frame with an expanded set of observation flags, screen captures are checked for outside-help windows, and the microphone recording gets a transcript-level triage. Every flag lands in front of a human reviewer — nothing becomes a verdict on its own.

IntegrityCohort-level similarity analysis surfaces possible collusion signals across candidates who sat the same assessment, a unified Logs timeline reconstructs each session from stored facts, and the authenticity report exports as a fully branded PDF — your organization's logo, colour-coded findings, and gated until the checks have actually run.

IntegrityProctoring and scoring are now two separate desks: the Proctoring tab is purely about session integrity, and all scoring happens on Results — so integrity review never blurs into grading.

ScoringA dedicated scoring workspace: an evaluator grading queue, questions in section order with per-question score pills, rubric-anchored entry with strict range validation, optional human override of auto-scored answers (the auto score is never mutated), a blind-scoring toggle for written work, and an explicit Ready-to-finalize → Finalized gate that Results waits for. Overall scores read 0–100 everywhere.

ReportsA standard candidate report with visible entry points, and an answer sheet: the candidate's exam paper — session, participant, and result blocks with a section-by-section breakdown — as a tab and a downloadable PDF.

AuthoringForced-choice situational judgment (SJT) items: statement pools with per-statement competency mapping, option use-caps, and weighted or threshold scoring — with per-competency roll-up into role fit. Open-text answers gained a hard server-enforced word limit.

AuthoringPDFs embed directly inside questions: candidates open a document card into an inline viewer, and large files open fast through on-demand page loading. The editor's candidate preview now renders the real exam card, type by type — what you see while authoring is what candidates get.

ItemsThe item bank reads its own quality signals inline: separate difficulty and discrimination chips per item, and the “Needs attention” reason spelled out on an amber chip. Rubrics gained a manager — search, bulk actions, standalone creation, and rename/edit/delete from the bank, plus AI-drafted rubrics from the question content.

Blueprints49 ready-made role blueprint templates seeded into every workspace, AI assist that drafts a blueprint or calibrates target levels — suggestions with stated reasons, always applied by a human — and a blueprint-aware fit model: the headline score reflects the blueprint, critical competencies act as a hard gate, and coverage is visible wherever fit is read.

InterviewsLive realtime interviews now offer a presentation choice per interview: a talking avatar or voice-only with an audio-reactive orb, with automatic fallback from avatar to voice and an eight-face interviewer gallery. Live simulations get the same choice, and previews run the real live experience.

CandidatesCandidate email got a paper trail: a searchable, paginated history of every send, per-assessment email template selection, bulk sends from a chosen saved template, and one-click invitation resends — with skipped rows explained instead of silently reported as sent.

TrustThe legal policy set went into force — an evidence-based privacy notice, cookie policy, and terms, published only once every claim matched deployed reality. Candidate data retention is now executable, not aspirational: a 24-month default (organization-configurable) enforced by daily sweeps, full data purge on organization offboarding, and the candidate AI disclosure names the realtime third parties involved.

PlatformThe dashboard moved onto the SkillCort design language: a consistent branded loading state, one breadcrumb pattern product-wide, a responsive sweep across tables and grids, and a simpler assessment lifecycle — draft → live → archived.

Drag-and-drop question types, a calmer proctored exam, and a sharper editor

Two new auto-scored interaction formats joined the item bank, the proctored exam stopped interrupting candidates, and authoring got a faster type picker plus a stronger writing surface.

AuthoringTwo new auto-scored question types: Drag & drop categories (sort items into labeled buckets) and Fill in the blanks (drag words from a word bank into gaps in a passage, with optional distractors). Both are drag-first with a tap-to-place path for keyboard, mobile, and screen-reader users; answer keys stay server-side and the item pool / word bank shuffles per candidate.

AuthoringCreating an item now starts from a searchable type library grouped by what the candidate does — Selection & Choice, Written & Input, File & Media, Evaluation — with scoring-mode filters and Auto-scored / Human review / AI-assisted badges on every card.

ItemsItem tags became name: value pairs (e.g. Difficulty: Hard, Subject: Math). A question can carry values from several tag names regardless of its folder, filters read “Name: Value” everywhere, and a bulk action applies or removes values across selected questions — including minting a new value on the spot.

Exam experienceProctored exams now ask for every camera and microphone permission up front on the start screen — never mid-exam. Browser permission prompts no longer count as leaving a locked-down exam, lockdown notices moved from a banner above the questions to a quiet corner toast, and every capture device is released the moment the assessment is submitted.

AuthoringScenario / case pages: build an item set around a shared stimulus — the case material stays pinned beside its questions for the candidate.

EditorA stronger rich-text editor: curated font choices, heading and alignment dropdowns, tables with merge/split and unmistakable cell selection, syntax-highlighted code blocks, a full-screen writing mode, and a compact icon toolbar that stays on one row.

Access control, self-registration, and defensible validity

Deliveries gained real access control — public sign-up, codes, and credential matching — and the blueprint page turned scores into reliability-qualified validity evidence.

AccessPublic and restricted access per delivery: a shareable link with optional self-registration, single-use invitation codes, or credential matching against a guest list (e.g. employee or student ID) — no per-candidate email required. Every access decision is made server-side, with no candidate enumeration.

AccessAttempt rules per delivery: multiple attempts, resume-in-progress, time-boxed attempts, and one-click retake grants from the candidate list.

AccessA per-delivery access form to capture custom fields at entry, plus a reviewable access log — filters, candidate linkage, and CSV export — of every pre-exam access event.

BlueprintsRole-fit evidence on the blueprint page: how the candidate pool measures against each role competency, with reliability stated per competency (ICC, McDonald's ω) and an honest “n/a” when there isn't enough data to trust a number.

BlueprintsGroup-disparity checks: four-fifths impact ratio and Cohen's d across any candidate group you choose — framed as a signal to investigate, never a legal adverse-impact ruling.

BlueprintsCriterion validity: record real outcomes and see the observed role-fit-to-outcome correlation, clearly labeled preliminary and uncorrected, with optional range-restriction and attenuation what-ifs.

ReportsOne-click validation-evidence export: reliability, criterion validity, group disparity, and a written methodology (what's claimed and what isn't) in a single defensible JSON pack.

EvidenceEvidence Wallet: a per-candidate, auditable dossier — task outputs, scores, evaluator notes, and integrity signals — viewable and exportable as one record.

PreviewPreview as candidate now runs the real AI interview and simulation conversations, not a stub, so authors experience exactly what candidates will before publishing.

BuilderLive assessments are now editable: add, edit, or reorder items without taking the assessment offline. Candidates who already submitted keep their exact results — scores come from what they answered. A staff-only re-score then recomputes objective scores across every candidate after you change an answer key, so your edits and their results stay consistent. Closed and archived assessments stay locked.

AuthoringSelection limits on multi-select and matching questions: set a minimum and maximum number of picks or matched pairs. Candidates see the rule under the question, extra picks lock at the cap, and the limits are enforced server-side.

AuthoringMultiple-choice answer layouts: besides the standard list and column layouts, options can render as inline chips or — for single-select questions — a dropdown menu.

AuthoringGiven cells in table questions: mark any cell as pre-filled content and candidates see it read-only — hand out a partially completed table and ask only for the blanks.

AuthoringTable questions grew up: row and column headers can now be asked as scored blanks, and neighboring cells can be merged into larger regions — enough to model real worksheets, not just flat grids.

CandidateExtended-time accommodations: grant a candidate a per-attempt time multiplier (1×–3×), and every timer in the exam — whole-assessment, section, and page — scales automatically.

CandidateDownloadable resources during the exam: attach reference materials — formula sheets, case documents, data files — that candidates open and download while they work, served through short-lived signed URLs.

CandidateAn on-screen calculator candidates can open during the exam, enabled per assessment — for the roles where the tool, not mental arithmetic, is the point.

DecisionA cohort analytics overview on the Decision Board: score distributions and role-fit spread across everyone who took the assessment, so a batch of candidates reads as a story you can act on, not just a table of names.

Voice interviews, page-based exams, and granular integrity

A fast month: the AI interviewer learned to talk, exams gained a page structure with per-page timers, and proctoring became precisely configurable.

InterviewsTurn-based voice AI interviews: candidates answer by speaking, transcripts are stored as reviewable evidence, and an adaptive follow-up question is asked back out loud. A typing fallback is always available.

InterviewsThe AI interviewer now appears as a disclosed talking-avatar persona for the opening question — always labeled as an AI interviewer, never imitating a person.

EvaluationTranscript-based AI score suggestions: criterion-level draft scores from interview and simulation transcripts, with model/version provenance. A named evaluator confirms or overrides every suggestion.

ExamsPages inside sections: assessments now structure as sections → pages → blocks, with shuffle available at all three levels (seeded per attempt) for fair parallel forms.

ExamsPer-page time limits: when a page's time runs out the candidate moves on and the page locks, alongside the overall assessment timer.

ExamsA redesigned candidate exam view: sticky progress bar, a question navigator with answered/marked states, page-by-page navigation, mark-for-review, and a pre-submit review summary.

IntegrityGranular proctoring controls per delivery: fullscreen lockdown with configurable violation limits, copy/paste blocking, and second-screen detection — all disclosed to candidates at the consent gate, all reviewed by a human.

ScoringSection scoring weights now flow into role-fit, and multiple-choice options support per-option credit multipliers — including negative marking for high-stakes exams.

BuilderA redesigned builder workspace: outline sidebar, focused panels, and a publish checklist computed from your real assessment data that blocks publishing until required setup is complete.

BuilderBulk authoring: import from the item library through a filterable picker, quick-paste multiple-choice questions, and AI generation with an explicit count — up to 100 questions per run, optionally grounded in an uploaded source document.

ReportsAudit pack export: one self-contained decision file with the assessment structure and weights, plain-language methodology, every score with its named evaluator and timestamp, integrity events with resolutions, and the decision log.

RubricsLevel-based matrix rubrics (criteria × levels with anchored behavior descriptions), criterion weights, and an option to show the rubric to candidates during the assessment.

PlatformOrganization settings (default language, timezone, reply-to email), assessment folders and labels, and side-by-side stimulus layouts for case-style items.

The foundation

The core assessment operating system, built and verified end to end before we opened the doors.

CoreRole blueprints, the skill & task bank with folders and tags, and the assessment builder connecting them.

TasksWork-sample task types across the board: written responses, file uploads, code, tables, audio and video recording, photo capture, and classic question formats.

EvaluationRubric scoring with multiple evaluators, evaluator notes, and transparent weighted scoring math — evaluators average within a criterion, criteria combine by weight, tasks roll up by skill and section weight.

DecisionsThe Decision Board: candidates compared on weighted role-fit with drill-down from any score into the underlying evidence, plus candidate evidence reports with white-label PDF export.

DeliveryDeliveries with time windows, email invitations, a candidate consent and preflight gate, and proportional integrity signals reviewed by humans.

AIThe AI copilot for drafting tasks, questions, and rubrics — with the founding rule already in place: AI assists, a named person decides.