Blog
Writing on captions, compliance, and the L&D workflow
Long-form posts from the GlossCap team. We write about WCAG 2.1 AA compliance, ADA Title II and EAA enforcement, LMS caption workflows, and the decisions L&D teams face when picking a captioning vendor. One post a week, no newsletter filler.
Latest
Live Training Operations · 2026-06-15
Live training captioning playbook: CART providers, real-time ASR, the vILT platform workflow, and the post-event recording that needs its own caption track
Live ILT and vILT sessions produce two compliance obligations: captions for the live audience (WCAG SC 1.2.4) and a caption file for the recording (WCAG SC 1.2.2 at 99% accuracy). Platform ASR in Zoom, Teams, Google Meet, and BigBlueButton delivers 72–85% accuracy on technical training content — below both thresholds. This guide covers CART provider sourcing and the 3-day briefing protocol that closes the vocabulary accuracy gap, platform-by-platform ASR setup, the multi-speaker Q&A protocols that prevent caption failure during open discussion, and the post-event recording timestamp-alignment problem that is the most commonly missed step in vILT programmes.
AI Video Captioning · 2026-06-15
Captioning AI-generated training video: the TTS→STT accuracy paradox, timing drift in synthetic speech, and the caption workflow for Synthesia, HeyGen, Descript, and Lumen5
AI avatar video platforms eliminate the human narrator — and accidentally eliminate the acoustic signal that makes speech recognition reliable. The TTS→STT round-trip degrades caption accuracy to 73–84% on engineering, compliance, and medical training vocabulary. Timing drift in videos over 10 minutes regularly exceeds the WCAG ±2-second synchronization threshold. HeyGen's built-in captions are STT-generated from TTS audio (not script-derived). The correct five-step workflow: export both video and SRT, correct timing drift, apply organizational glossary, upload to LMS, verify in the learner player.
Compliance Operations · 2026-06-14
Captioning purchased compliance training content: who owns the obligation when you use ComplianceWire, Navex One, Skillsoft, or OpenSesame
When you license compliance training from ComplianceWire, Navex One, Skillsoft, or OpenSesame, your ADA Title I and Section 508 caption obligation doesn't transfer to the vendor. The employer remains the responsible party. Vendor-by-vendor caption accuracy analysis (ComplianceWire 83–91%, Navex One 84–92%, Skillsoft 85–93%, OpenSesame varies by publisher), SRT file access by platform, five remediation workflows when vendor files are missing or below 99% WCAG, audit methodology for your third-party content library, and the six contract provisions to require before signing.
Budget Planning · 2026-06-14
Three-year caption programme budget model: per-minute vs subscription cost curves, internal labour allocation, and how to build the year-1 budget request
The ROI business case wins Finance approval. The operational budget model is what you use the approval for. This covers the three budget buckets every complete caption programme request must include (external vendor, internal labour, LMS infrastructure), the per-minute vs subscription cost curve comparison over three years, the glossary accuracy compounding effect that holds correction labour flat through volume growth, the seven cost levers L&D directors control, and year-1 through year-3 budget templates in three org-size tiers.
Vendor Evaluation · 2026-06-14
How to test a captioning vendor's accuracy before signing: test corpus design, DCMP-protocol scoring, and the red flags that predict live-environment failure
The RFP identifies which vendors qualify. The contract review covers what you sign. This covers the accuracy evaluation that happens between shortlist and signature: how to design a diagnostic test corpus (not a representative one), prepare reference transcripts for DCMP word-level scoring, identify vendor red flags that predict live failure, and connect evaluation data to an enforceable accuracy guarantee clause.
Compliance Guidance · 2026-06-13
Are AI-generated auto-captions ADA and WCAG-compliant? The 2026 regulatory position, platform accuracy reality, and when human review is legally required
WCAG 1.2.2 requires "accurate" captions but doesn't specify a percentage — that 99% threshold comes from the DCMP Captioning Key, which OCR and courts treat as the operative benchmark. YouTube, Teams, and Zoom auto-captions score 76–89% on technical training content. Platform-by-platform accuracy data, LMS caption system analysis, and the five conditions that determine when auto-captions are legally sufficient.
Employee Communications Operations · 2026-06-13
Captioning employee communications video: all-hands meetings, town halls, exec recordings, and the ADA Title I gap most L&D teams don't audit
All-hands recordings, town halls, executive video messages, and policy announcements are employer communications under ADA Title I — but most organizations have no caption workflow for them. This is the operational guide to the ownership gap, live-versus-recorded compliance, platform-by-platform workflows for Zoom/Teams/Vimeo/Wistia, and eight failure modes that leave organizations silently exposed.
Higher Education Operations · 2026-06-12
Captioning university lecture capture: Panopto, Echo360, Canvas Studio, Blackboard Ultra, and the academic-calendar compliance problem
University lecture capture has a compliance problem that corporate training video does not: the academic calendar. A new cohort of uncaptioned content arrives every sixteen weeks, the back-catalogue spans five to ten years of recordings, and the proper-noun failure rate is higher than any other training-video category because lecture audio concentrates faculty names, course codes, cutting-edge research terminology, and institutional acronyms that ASR models do not see in their training data. This practitioner's guide covers the platform-by-platform caption workflows for Panopto, Echo360, Canvas Studio, Blackboard Ultra, and Kaltura REACH; the institutional glossary architecture (three tiers: institution-wide, department, course) that makes the proper-noun problem manageable before batch processing begins; the academic-calendar compliance problem in detail (back-catalogue depth, new-content rate, guest lecture problem, course-closure timing); the semester-based compliance workflow (pre-semester setup, in-semester production, end-of-semester review, back-catalogue remediation); proper-noun failure modes specific to academic content (faculty names, course codes, cutting-edge research vocabulary, institutional acronyms); eight failure modes in university lecture-capture caption programmes; and a seven-question FAQ covering FERPA interactions, OCR investigation documentation, ADA Title II phased back-catalogue compliance, and platform-migration caption transfer.
Compliance Operations · 2026-06-12
Caption compliance KPIs and leadership reporting: what to measure monthly, the audit-readiness dashboard, and the metrics that survive a WCAG investigation
A complete caption compliance measurement framework has five KPI categories: library coverage rate (the primary outcome metric — what fraction of the obligation is discharged), new-content submission rate (the leading indicator that predicts whether the programme is gaining or losing ground), accuracy pass rate (the quality gate — what percentage of spot-checked videos meet the 99% WCAG 2.1 AA threshold), remediation velocity (the operational health metric — whether the programme can close its own findings within the target window), and exception programme health (the governance metric — whether the exception procedure is being administered or ignored). These five categories generate twelve reportable metrics, composing a monthly dashboard that takes roughly two hours to populate when the measurement protocol is defined in advance. The post covers the complete monthly measurement protocol in eight steps, the twelve-row dashboard structure with RAG thresholds and a worked illustrative example, the one-slide leadership model and the 200-word email format, the specific metrics and documentation that hold up under a WCAG or ADA Title II investigation (the four questions investigators consistently ask, the documentation pattern that changes the outcome), LMS-specific reporting notes for eight platforms (Kaltura, Docebo, TalentLMS, Absorb, Cornerstone, Workday Learning, Panopto, Canvas), and eight failure modes in compliance reporting including the coverage-rate fallacy (reporting only library coverage rate without submission rate), the undocumented methodology gap, and the single-point-of-failure documentation storage problem.
Change Management Operations · 2026-06-12
Rolling out a captioning programme to your L&D team: workflow adoption, producer training, and the resistance patterns that stall programmes
Most captioning programmes stall not at the design phase but at the producer adoption phase. The three structural gaps that cause rollout failure are workflow fragmentation (the submission step is not integrated into the normal publishing workflow), scope ambiguity (the content gate is not precise enough for producers to self-serve), and accountability absence (there is no monitoring mechanism that makes non-compliance visible). This practitioner's guide covers all three phases of a change management rollout — the foundation phase (governance policy, baseline audit, LMS integration, glossary seeding, before any producer is trained), the producer training phase (the three capabilities every producer needs, 30-45 minute session design, champion producer recruitment, phased department sequencing), and the live-period monitoring phase (monthly submission review, first non-compliance follow-up, exception handling in practice, workflow calibration). Also covers the four resistance patterns (extra work, content exemption, accountability absence, vendor blame) and their structural counters; LMS-specific rollout notes for Kaltura, Docebo, Cornerstone OnDemand, TalentLMS, Panopto, Workday Learning, Absorb, WorkRamp, Brightspace, and Canvas; the eight change management metrics including the four leading indicators (submission rate, on-time submission, review completion, first-time pass rate) and four lagging indicators; and eight failure modes that stall programmes that had adequate technical foundations.
Programme Governance · 2026-06-12
Writing the internal captioning policy: approval workflows, exception handling, new-content gates, and the four clauses that stop the drift
There is a meaningful difference between having a captioning process and having a captioning programme. A process is the workflow your team follows today. A programme is the durable governance structure that makes the process survive staff turnover, handle edge cases consistently, and withstand audit scrutiny. The bridge between them is a written policy document. This guide covers the four clauses that appear in every functioning captioning policy: the content gate (what triggers the captioning requirement), the accuracy standard (what "compliant" means in measurable terms), the exception procedure (what happens when a video can't meet the standard on schedule), and the audit trigger (what forces a compliance review). Also covers approval workflow architecture, the new-content gate decision logic for original/revised/republished/third-party content, exception handling in practice, and the seven-section policy document structure.
Vendor Management · 2026-06-11
How to switch captioning vendors: porting your glossary, managing the backlog, and the accuracy reset problem
Switching captioning vendors means risking the accuracy investment you have been compounding for 12–24 months: the per-customer glossary, the correction history, the LMS delivery configuration. This practitioner's guide covers the full transition lifecycle — when to consider switching (accuracy stagnation, renewal restructure, vendor-side change, capability gap), what you legally own and what the vendor retains (glossary is your IP if the contract says so; the fine-tuned model is not), the accuracy reset problem and how to size it before the switch (the difference between 98.5% and 93% is ~120 additional corrections per 10-minute video), porting the glossary before giving notice, the parallel-run period and how to set go/no-go criteria before the pilot starts, per-LMS delivery during the transition (Kaltura, Docebo, Cornerstone, TalentLMS, Panopto, Workday), contract termination mechanics including BAA data-return obligations, eight failure modes that turn a well-planned switch into a compliance regression, and the stay-vs-switch decision checklist.
Finance · 2026-06-11
Making the caption ROI argument to your VP of Finance: exposure math, vendor cost comparison, and the half-FTE payback model
Your VP of Finance has not read the WCAG specification — but they are very good at "what is the downside of not doing this, and what does it cost?" This guide builds the three budget-bucket components of the captioning business case: (1) risk exposure from ADA Title II / Title III / EAA non-compliance, with an expected-value table for a representative 150-employee org; (2) the absorbed correction labour — the hidden half-FTE already embedded in your L&D calendar, quantified at $24,000–$72,000/year depending on video volume; (3) the vendor cost comparison showing why per-minute pricing (Rev human review at $1,350/month for 15 hrs/month) costs 13.6× more than flat-monthly software at the same volume. Includes the one-slide business case template, the seven finance objections and their counters, and the full ROI calculation at three org configurations with conservative and aggressive scenarios. The conservative ROI at the 15-hour/month median configuration is 49×; the payback period is under 30 days.
Procurement · 2026-06-10
Caption vendor contract review checklist: SLA terms, accuracy guarantees, BAA provisions, and the clauses that matter
The contract phase of captioning vendor selection is where most L&D teams leave money, rights, and compliance protection on the table. This practitioner's guide covers the eight clause categories that determine whether a captioning engagement works as expected over its full term: accuracy guarantees (how to specify measurement protocol, sample methodology, and remediation triggers — not just the 99% floor), SLA terms (turnaround end-to-end, back-catalogue retrofit timeline, breach consequences), HIPAA BAA provisions (when a BAA is required, what a HIPAA-sound BAA must contain, five red flags in vendor BAA drafts), data retention and deletion (30-day audio retention, automatic deletion on termination, certificate of deletion), glossary ownership and portability (per-customer glossary as buyer IP, self-service export rights, model training exclusion), integration and format SLAs (output format guarantees, 90-day deprecation notice, LMS integration uptime), pricing and volume terms (overage rate as multiple of committed rate, rollover policy, rate lock with renewal cap), and termination and transition rights (termination for cause, transition assistance SLA, data return on exit). Includes the 42-point contract review checklist to run before signing any captioning vendor contract above $10K annual value, and the seven most common contract failures and how each one plays out operationally.
Customer Education Operations · 2026-06-10
Captioning customer education and partner training academies: ADA Title III, Skilljar, Thought Industries, LearnUpon, and the public-academy compliance difference
Customer education academies — hosted on Skilljar, Thought Industries, LearnUpon external portals, Gainsight CE, and Docebo — face ADA Title III (public accommodation), not ADA Title I (employees). This guide covers the legal framework difference, why authenticated academies are still covered, platform-by-platform caption workflows for the five most common customer education platforms, partner training legal classification, product-vocabulary accuracy challenges unique to customer education content (product names, UI labels, API identifiers, version numbers), the caption production workflow for CS and CX teams with no existing L&D captioning infrastructure, and eight failure modes that leave external academies non-compliant even when the internal training library is fully captioned.
Compliance Operations · 2026-06-10
The L&D accessibility coordinator playbook: building the role, managing caption compliance at scale, and what to track on a dashboard
The operational playbook for the L&D professional who owns caption compliance. Covers role scope definition and what the accessibility coordinator owns versus what others own, how the role scales across small (50–150 employee), mid-size (150–500), and large (500+) organizations, the full RACI for caption operations across content producer/accessibility coordinator/LMS admin/caption vendor roles, platform-by-platform LMS caption management for TalentLMS, Docebo, Absorb LMS, Cornerstone OnDemand, Workday Learning, Kaltura, Panopto, LearnUpon, 360Learning, Bridge, and Rippling, the seven-metric compliance dashboard with data sources, caption governance policy and procedure documentation, batch captioning workflows and the prioritization framework for the remediation backlog, vendor management from the accessibility coordinator seat including SLA monitoring and RFP evaluation criteria, and the eight failure modes that stall caption compliance programs at 70–80% library coverage.
Remote Workforce Operations · 2026-06-06
Captioning async training video for remote and hybrid teams: Loom, Zoom recordings, Microsoft Teams, and the home-office audio problem
The operational guide to captioning async training video in remote and hybrid workforce environments. Home-office audio degrades ASR baseline accuracy by 7–12 percentage points before any domain vocabulary is encountered — HVAC noise, omnidirectional laptop microphones, room reverb, and recording compression artifacts combine to push typical home-office recordings to 78–88% baseline accuracy rather than the 95–97% of controlled-environment production. Covers platform-specific caption workflows for Loom, Zoom cloud recordings, Microsoft Teams recordings, Microsoft Stream, Vimeo Business, and Wistia; glossary architecture for multi-speaker distributed libraries including speaker accent variation and vocabulary inconsistency across 20+ producers; WCAG 2.1 AA SC 1.2.2 and ADA Title I compliance for remote employee training; the distributed production model failure (non-L&D producers create training recordings with no caption workflow triggered); and eight failure modes including platform auto-captions falsely treated as WCAG-compliant, home-office audio causing systematic accuracy failures for specific speakers, and Teams personal-meeting recordings not routed to the central caption workflow.
LMS Operations · 2026-06-06
The L&D caption checklist for LMS migrations: what caption data survives the move, what breaks, and how to audit before cutover
The operational guide for L&D teams managing an LMS migration. Covers the five caption data types that require explicit migration (caption files, track metadata, timing integrity, glossary configuration, accuracy history), a migration capability matrix for ten common source-to-destination platform pairs (TalentLMS→Docebo, Cornerstone→Workday, Panopto→Kaltura, and more), platform-by-platform export checklists for eleven source LMS platforms, per-destination import checklists with format requirements and auto-generation clobber risks, the caption freeze window (the compliance gap no migration plan addresses), the post-migration five-step validation protocol, and eight failure modes that cause post-migration caption loss — most of which produce no error in the migration log.
Frontline Training Operations · 2026-06-05
Microlearning captions for frontline workers: TalentCards, EdApp, Axonify, and OSHA-vocabulary accuracy
The operational guide to captioning microlearning for frontline workers. Factory floor and construction site ambient noise exceeds 80–90 dB — captions are the primary content delivery channel, not an accommodation. Covers per-card SRT upload in TalentCards (not per-deck), the EdApp Creator-role requirement that silently blocks caption upload for Author-role users, Axonify's spaced-repetition caption-drift problem on content updates, OSHA-vocabulary accuracy benchmarks by content type (LOTO 84.6% baseline → 98.9% at 60 terms; HazCom 82.3% → 99.2%; respiratory protection, fall protection, transportation), the compliance framework under OSHA 1910/1926, ADA Title I, Section 508, MSHA, and DOT FMCSA, volume math and RACI for 200-module back-catalogue retrofits, and the eight failure modes that produce uncaptioned content on workers' devices despite appearing complete in the admin panel.
Sales Enablement Operations · 2026-06-05
Captioning sales enablement video at scale: WorkRamp, Highspot, Seismic, and the SKU-name problem
The operational guide to captioning sales enablement video at scale: how to keep caption vocabulary synchronized with quarterly product releases, per-platform workflows for WorkRamp, Highspot, Seismic, Allego, and Bigtincan, the SKU-name glossary architecture that prevents accuracy cliffs at every product launch, accuracy benchmarks for sales training content (16.1% baseline WER on new-SKU product briefings, dropping to 1.1% with a 70-term sales glossary), WCAG compliance obligations under ADA Title I for employee training, the two-tier production model for planned SKO recordings and continuous coaching recordings, and the eight failure modes that cause sales caption quality to degrade silently at the most commercially important terms.
Compliance Operations · 2026-06-05
Multi-language caption workflow for global L&D teams: translation pipeline, LMS delivery, EAA and AODA compliance
The operational guide to building a multi-language caption workflow for L&D teams serving EU and Canadian employees. Covers the four-stage translation pipeline (source caption lock, method selection, timing adaptation, target-language QA), translation method selection by content type and regulatory exposure, per-language glossary architecture and term freeze discipline, LMS delivery workflows for Docebo, Kaltura, TalentLMS, Workday, Cornerstone, Panopto, SAP Litmos, and 360Learning, EAA/EN 301 549 compliance obligations for multi-language digital content, AODA bilingual requirements for Ontario employers, the adapted DCMP spot-check for translated captions including reading-speed adjustment for text-expansion languages, and the eight failure modes that leave teams with non-compliant translated caption tracks.
Quality Operations · 2026-06-04
Caption QA for training video production: how to run spot-checks, set pass/fail thresholds, and fix systematic errors
The operational guide to building a defensible caption QA process. Covers the DCMP spot-check protocol, the four error types and what each signals about root cause, pass/fail thresholds by content category, LMS-specific QA workflows for eight platforms, the five-step systematic error triage protocol, QA roles and RACI, and the eight failure modes that keep teams at 94–96% accuracy.
Product Operations · 2026-06-04
The caption feedback loop: how training teams compound glossary accuracy from 91% to 99% over 6 months
How L&D teams run a structured correction feedback loop that makes caption accuracy improve with every captioned hour. Covers the six signal types, three phases of compounding, vertical-specific patterns (engineering, healthcare, sales, manufacturing), and the switching cost that accumulates after 200+ hours.
Compliance Operations · 2026-06-03
How to audit your LMS caption library for WCAG compliance: methodology, tooling, and a 5-day sprint plan
A seven-dimension framework for L&D operators who need to know whether their existing caption library is WCAG 2.1 AA–compliant — covering coverage, format validity, accuracy (word error rate and proper-noun WER), synchronisation, metadata and accessibility labelling, LMS delivery across web/mobile/SCORM/offline contexts, and documentation. Includes LMS-specific audit playbooks for TalentLMS, Docebo, Absorb, Cornerstone OnDemand, Workday Learning, Kaltura, Panopto, WorkRamp, Allego, Bigtincan, and TalentCards. A five-day sprint plan for a single L&D operator covers a 300-asset Tier 1 catalogue across all seven dimensions, producing a remediation-prioritised export with effort taxonomy (format conversion, metadata correction, recaption from transcript, full recaption from audio) and a vendor-sizing input for the back-catalogue retrofit.
Technical Architecture · 2026-06-03
How to build a customer glossary for AI captions: architecture, ingestion, term sourcing, and the compound accuracy effect
The definitive technical guide to building a domain-specific vocabulary artifact for AI caption systems. Covers why flat word lists underperform structured glossary entries (canonical form, phonetic variants, context signals, priority weights); eight sourcing methods including Notion/Confluence sweeps, training script analysis, past correction file extraction, SME interviews, and employee directory sweeps; glossary taxonomy by vertical (engineering 90+ terms, healthcare 48 terms, EHS 72 terms, sales 50–70 terms, legal 30–50 terms); the full ingestion architecture from phoneme alignment to decoder-side beam search biasing; the initial prompt injection method for teams building on the Whisper API; glossary sizing with per-vertical diminishing-returns curves and a practical sizing table; the compound accuracy effect — how each captioned hour improves the glossary through the feedback loop, documented from 92.5% at session 1 to 99.1% after 100 hours; glossary maintenance cadence (continuous, quarterly, event-triggered); conflict resolution for homophone collisions and temporal renames; and a 30-day build plan from audit to first production version to maintenance infrastructure.
Technical Research · 2026-06-03
Whisper accuracy benchmarks by vertical: engineering, healthcare, financial services, manufacturing, legal, and sales training
Whisper large-v3 word-error-rate benchmarks across eight L&D content verticals, showing why "94% accuracy" vendor claims are meaningless without knowing which vertical your content lives in. Engineering onboarding content: 88.8% baseline → 99.2% with a 94-term glossary. Healthcare/clinical: 86.9% → 99.4% with 48 terms. Manufacturing/EHS: 85.8% → 99.1% with 72 terms — the hardest vertical at baseline, driven by IUPAC systematic chemical names and OSHA regulatory codes. Financial services: 89.2% → 99.1% (SOFR → "soccer," CECL → "Cecil," the systematic initialism-to-common-word mapping problem). Legal/compliance, sales enablement, government, and academic results included. Per-vertical error-category taxonomy. Model-size effect table (tiny through large-v3): the glossary investment outperforms 3 model tiers worth of scale. Glossary-size versus accuracy-gain curves per vertical, showing when you hit diminishing returns and why healthcare gets most of the benefit from the first 20 terms while engineering needs 90+. Compliance implications for WCAG 2.1 AA SC 1.2.2: which verticals can reach 99% with a light-touch review pass versus which require a systematic correction workflow. Instructions for running your own content-specific accuracy test using the methodology described.
Technical Guidance · 2026-06-02
Live session captions vs recorded training video: why the accuracy requirements are completely different
WCAG SC 1.2.4 (live captions) and SC 1.2.2 (prerecorded captions) are different standards with different accuracy expectations. When a Zoom recording is uploaded to an LMS, the live caption track at 80–88% accuracy no longer meets the prerecorded standard at 99%+. This post explains why the two standards differ, how Zoom, Teams, Webex, Google Meet, Panopto, and Mediasite behave in practice, what the accuracy data looks like for each platform, and how to build the two-stage workflow that closes the archival gap without creating heavy manual burden. The two-stage requirement: Stage 1 — live captions enabled during the session (SC 1.2.4 compliance). Stage 2 — glossary-corrected reprocessing before LMS publication (SC 1.2.2 compliance). Both required. Neither replaces the other.
Program Management · 2026-06-02
Building a caption compliance program from scratch: from policy to audit trail in 90 days
Most L&D teams cross the threshold from "we need captions" to "we have captions on some things" and then stall. The back catalogue stays partially captioned. New video ships without captions until someone complains. This post is not about how to caption a video — it is about how to build a program: a policy, a workflow, a vendor relationship, a quality-control gate, and an audit-documentation package that makes caption compliance self-sustaining. The 90-day plan covers five phases: inventory and gap analysis (days 1–14), policy and governance (days 15–28), vendor and tooling selection (days 29–45), production workflow and QC (days 46–70), back-catalogue retrofit (days 60–90). Includes the captioning policy template, QC rubric, vendor-evaluation criteria, compliance-urgency triage matrix, and the audit-documentation package your legal team will need if you face an OCR complaint or DOJ investigation.
Compliance Guide · 2026-06-01
Section 508 vs ADA Title II vs WCAG 2.1 vs state laws: the training-video compliance matrix with decision tree
The US accessibility compliance landscape for training video is not one law — it is five overlapping federal frameworks plus a growing layer of state laws, each with its own covered entities, technical standards, enforcement agencies, and risk profiles. This post maps every framework to the organization types it covers: Section 508 for federal agencies and contractors (WCAG 2.0 AA via 36 CFR Part 1194); ADA Title II for every state and local government entity including public universities, with WCAG 2.1 AA effective April 24, 2026 for large jurisdictions; Section 504 for any organization receiving federal financial assistance — every hospital accepting Medicare/Medicaid, every college accepting Title IV student aid — including the HHS May 2024 rule adopting WCAG 2.1 AA for healthcare entities; ADA Title I for private employers with 15+ employees and their mandatory employee training; ADA Title III for publicly accessible customer academies, open certification programs, and public training portals. State laws covered in detail: California AB 434 and Unruh Civil Rights Act (§4,000 statutory damages per Title III violation), Colorado HB 21-1110 (WCAG 2.1 AA for state agencies), Virginia VITA standards, Washington State OCIO Policy 188, Texas DIR standards, New York Human Rights Law, Illinois ITAA, Maryland COMAR. Includes: the full applicability matrix (10 organization types × 5 frameworks); a six-branch decision tree to identify every framework that applies to your org; enforcement risk profile by framework (who enforces, what triggers it, typical outcome, practical risk level); the cross-framework accuracy question (WCAG 2.0 vs 2.1 — does it matter for captions?); the Section 1557 and CVAA overlays; the public university multi-framework case study; a 30-day compliance sprint plan from inventory to audit trail; and a 7-question FAQ covering private employers, auto-caption adequacy, legacy content, multi-framework overlap, and cross-border applicability.
Format Guide · 2026-06-01
SRT vs VTT vs TTML vs STL: the L&D operator's guide to caption formats, LMS compatibility, and conversion pitfalls
Four caption file formats have survived long enough to matter in the training video world: SRT (the plain-text universal), VTT (the HTML5 native), TTML (the XML-structured broadcast heritage), and STL (the binary broadcast legacy). Every L&D operator eventually faces a platform that refuses one of them — and the rejection message rarely explains why. This post covers the full LMS compatibility matrix (21 platforms), the five silent corruption bugs that break captions during format conversion before a single frame plays (BOM, CRLF, comma-vs-period timing separator, zero-indexed sequence numbers, sequence reset across multi-segment video), a per-platform deep dive for TalentLMS (SRT only, comma required), Docebo (SRT/VTT with mandatory BCP-47 locale), Absorb LMS (SRT with no-BOM and LF-only requirements), Kaltura (SRT/VTT/TTML, DFXP profile, caption asset API), Panopto (VTT preferred, voice spans, transcript search), Vimeo API (raw VTT body, not multipart), Wistia (ISO 639-2/T three-letter codes), Canvas, Brightspace, Workday, Cornerstone, LinkedIn Learning, SAP Enable Now, Salesforce Trailhead, 360Learning, Schoology, Loom, Camtasia, and Microsoft Stream. Why WCAG 2.1 AA success criterion 1.2.2 says nothing about format — only accuracy ≥99% and synchrony ±2 seconds. The GlossCap export workflow (SRT + VTT simultaneously, BOM-free, LF endings, TTML-on-request). Validation tools (W3C Nu Checker for VTT, VLC and SubtitleEdit for SRT). A Python byte-level hygiene script for BOM and CRLF detection. Format selection decision tree: when to pick SRT, when VTT beats SRT, and when TTML is actually required. How to generate STL without hand-authoring binary files. Vertical notes for healthcare (HealthStream, Relias), federal/Section 508, university lecture capture, and multi-LMS corporate stacks.
Engineering · 2026-06-01
The LMS caption ingestion workflow: bulk retrofit across TalentLMS, Docebo, Absorb, Kaltura
Bulk caption retrofit across the four most common mid-market LMS platforms is harder than it looks. Each platform has a different upload surface, format expectation, BCP-47 language-code convention, and API depth. This post walks the full engineering workflow — inventory, triage, generate, normalise, upload, verify — and documents the format normalisation failures that corrupt captions silently in each platform's player: UTF-8 BOM (corrupts first cue in TalentLMS and Absorb), Windows CRLF line endings (breaks cue boundaries in Absorb), SRT comma-vs-VTT-period timing separator (TalentLMS rejects VTT timing syntax even with .srt extension), Docebo's rejection of bare ISO 639-1 language codes (422 on en; requires en-US), Wistia's requirement for ISO 639-2/T three-letter codes (eng, not en), and Vimeo's pre-signed URL upload requiring a raw VTT body rather than multipart form data. Platform-specific API walkthroughs for TalentLMS, Docebo, Absorb, Panopto, Vimeo, and Wistia. Kaltura caption asset API in full — captionAsset.add, uploadToken, captionAsset.setContent, captionAsset.get status polling, and REACH bulk submission for general-vocabulary content. xAPI caption-viewed event structure for compliance documentation in Kaltura + Docebo LRS. A worked example: 340-asset mixed estate (Docebo + Panopto + Wistia), Tier 1 retrofit of 62 assets in 7 hours from first API call to compliance documentation package. Seven-day engineering sprint plan. Common failure mode diagnostic table. FAQ: SCORM repack workflow (Articulate Storyline 360), Panopto-via-LTI vs LMS-layer caption management, Docebo vs Vimeo caption ownership for embedded video, Kaltura REACH batch submission, Section 508 VPAT scope vs content obligations, LMS migration caption transfer pattern.
Technical · 2026-05-31
The proper noun problem in training video captions: 15 categories of words that break auto-captions, and what to do about each one
Every captioning post on this blog is, at some level, about the same underlying problem: ASR models were trained on general English, and training video is not general English. This post is the systematic version of that argument. We catalogued every category of word that reliably fails automatic speech recognition in domain-specific training content — 15 categories in total — and for each one explain why it fails, what the failure looks like in practice, and what glossary injection can and cannot do about it. The 15 categories: software and platform product names (Kubernetes → "cube ernetus," PagerDuty → "pager duty"), proprietary acronyms and initialisms (JSON → "Jason," SCIM → "skim," RBAC → "are back"), pharmaceutical INNs and brand names (pembrolizumab, adalimumab, Eliquis), IUPAC chemical names (2,4-dichlorophenoxyacetic acid with embedded locant numerals), regulatory citations (29 CFR § 1910.147 spoken as cardinal numbers but formatted as dotted-decimal notation), financial codes and instrument names (SOFR → "soccer," CECL → "Cecil," CCAR → "scar"), medical eponyms and clinical nomenclature (Creutzfeldt-Jakob, Dupuytren's contracture, CHA₂DS₂-VASc), person names (Jensen Huang → "Jensen Wang," Satya Nadella → "Satya Nadela"), geographic and institutional names, operator codes and keyboard shortcuts (Ctrl+Alt+Del, EACCES), part numbers and SKUs (R740xd, m5.xlarge), certification codes (CISSP, CompTIA, CKA), CLI commands (kubectl apply -f, userAuthToken → "user auth token"), regulatory body names (FinCEN, OFAC, PCAOB), and non-English loanwords (heijunka → "hedgehog," jidoka → "judoka," DSGVO). Per-vertical breakdown table mapping each of the 15 categories to the dominant verticals (engineering, medical, financial services, manufacturing, construction safety, cybersecurity, government, FDA-regulated, sales enablement, compliance). Three-stage intervention pipeline: decoder-side glossary injection (Stage 1, closes 85–95% of proper-noun errors in most categories), post-processing text normalisation (Stage 2, for citation formats and code tokens), and human review of low-confidence segments (Stage 3, backstop for genuinely out-of-vocabulary terms). Glossary size, token budget, what to include and exclude, and maintenance cadence. Seven-question FAQ including whether model size (medium → large-v3) fixes proper nouns, phonetic respelling as a workaround, which categories resist glossary injection, live vs. recorded captioning differences, and how to diagnose which categories apply to your content without guessing.
EHS Compliance · 2026-05-31
Captioning HazCom training: why SDS chemical names break ASR, and the OSHA 1910.1200(h) documentation rule
OSHA 29 CFR 1910.1200(h) requires documented, effective training on every hazardous chemical in the workplace. SDS training carries the highest IUPAC systematic-name density of any EHS content type — and Whisper-large at default settings produced 231 errors in 1,874 words (87.7% accuracy) on a 15-minute HazCom module covering 12 chemicals. Of those errors, 83% clustered in five HazCom-specific proper-noun categories: IUPAC systematic names (every one incorrect without glossary), chemical trade names and abbreviations, GHS hazard-category designations (H-codes and P-codes), exposure-limit codes (NIOSH REL, OSHA PEL, TLV-TWA, IDLH), and PPE abbreviations (PAPR, SCBA). The same audio with a 52-term chemical glossary closed 186 of 193 proper-noun errors and brought accuracy to 99.1%. Covers: why IUPAC names fail worse than drug INNs (longer, embedded numerals, near-zero training-data coverage), the four structural IUPAC failure patterns (substituted alkanes with locant numerals, isocyanates and diisocyanates, reactive peroxides, regulated carcinogens), the H-code and CAS-number format-conversion problem and its post-processing fix, the OSHA § 1910.1200(h) "effective training" standard and how garbled captions create a documented-training compliance gap, the OSHA inspection sampling pattern (chemical inventory → SDS verification → training records → worker interview), the ADA Title I and workers' compensation exposure angles, and a five-day EHS remediation sprint plan.
Ontario Compliance · 2026-05-30
The AODA multi-year accessibility plan template for Ontario 50+ orgs, annotated section by section
The Integrated Accessibility Standards Regulation (IASR, O. Reg. 191/11) requires every Ontario organization with 50+ employees — and all designated public-sector organizations regardless of size — to maintain a public multi-year accessibility plan. The next large-org compliance-reporting window closes 2026-12-31. This annotated template covers all five AODA standards with particular depth on the Information and Communications Standard training-video caption obligation: WCAG 2.0 AA Section 17 scope (new and significantly refreshed web content including LMS-hosted training video), the substantive accuracy floor the Accessibility Compliance Authority applies in audits (≥99% on DCMP scoring criteria — not auto-captioning at 80–90%), the proper-noun failure mode in Ontario-specific content (PHIPA/MFIPPA/WSIB regulatory vocabulary, bilingual French proper nouns, Indigenous community names, provincial program citations), the ACA audit methodology and compliance-order sequence, and a cross-walk of AODA vs ADA Title II (WCAG 2.1 AA, 28 CFR § 35.200) vs EAA / EN 301 549 for cross-border organizations. Plus a captioning-specific annex that most organizations are missing from their plan, the back-catalogue remediation timeline template, the new-content enforcement workflow for Brightspace, Canvas, and Moodle, and a 30-day sprint plan to get your plan updated before the 2026 filing window closes.
Procurement · 2026-05-01
How we ran a captioning vendor RFP: scoring sheets, vendor responses, and what we'd do differently
A first-person walkthrough of a captioning-vendor RFP for a 280-employee mid-market healthcare-adjacent SaaS with a 1,400-hour training-video back catalogue. The 14-question RFP template across four weighted sections (Accuracy 35 / Format-Integration-Workflow 20 / Security-Privacy-Compliance 25 / SLA-Pricing-Contract 20). Six anonymised vendor responses (Rev-shaped, 3Play-shaped, Verbit-shaped, AI-Media-shaped, Otter-shaped, GlossCap-shaped) with the actual numerical scoring per sub-question — including the Q4 supplied-audio-and-supplied-glossary sample where the spread on a 17-named-entity 60-second clinical-content excerpt ranged from 0 to 13 proper-noun substitution errors. The full 100-point scoring sheet, the Section 3 BAA-and-SOC-2 deal-breaker filter that walked two vendors out, the Section 4 pricing math where year-1 quotes ranged from $38.6K to $264K, the SOC 2 Type II risk-acceptance trade we documented for the winning vendor, the 6-week procurement timeline, and the four mistakes we made — sample request too late, vendor-supplied demo audio, BAA not on call 1, integration weighted as hygiene. Plus eight things we'd do differently next time, the per-section weighting if your content is not proper-noun-heavy, the engineer-on-evaluation-team load-bearing input, and a 6-month status update on how the winning vendor is performing.
EU Compliance · 2026-04-30
The EU accessibility statement template for an EAA-bound SMB, annotated clause by clause
The 12-clause EU accessibility statement template (Implementing Decision (EU) 2018/1523) reused under the EAA, annotated clause by clause. The two EAA-specific add-on clauses — Article 14 disproportionate-burden assessment with the Annex VI 3-limb template, and Article 4(5) microenterprise carve-out documentation with the AWU / linked-and-partner / two-financial-years rules. Member-state variations across Germany (BFSG section 4 BFSGV easy-language and DGS sign-language obligations, MLBF €100,000 fine ceiling), France (Loi n° 2023-171 + Décret n° 2019-768 multi-year accessibility plan and DGCCRF 7-day acknowledgement clock), Italy (Decreto Legislativo 82/2022 AgID JSON format at /.well-known/accessibility-statement.json + quarterly review minimum + AgID/AGCOM jurisdiction split), Spain (Real Decreto 193/2023 autonomous-community contacts + Catalan/Galician/Basque/Valencian language obligations), the Netherlands (Implementatiewet article 2 Dutch-language obligation + €900,000 / 1% of worldwide turnover ceiling), plus quick-reference for Austria, Belgium, Denmark, Finland, Ireland, Poland, Portugal, Sweden. The feedback-mechanism operational checklist, the periodic-review cadence (annual / quarterly / 30-days-from-trigger), the OCR Title II / Section 504 / EAA cross-walk for an EU-and-US-operating SMB, and a downloadable annotated template you can copy verbatim. 30-day publication plan.
Healthcare Compliance · 2026-04-30
Captioning under a Joint Commission triennial survey: HR.01.05.03, NPSG.03.05.01, IC.02.04.01 walk-through
The hospital captioning playbook for the triennial unannounced survey. The three Joint Commission standards that put training video on the hot seat — HR.01.05.03 (ongoing in-service education), NPSG.03.05.01 (anticoagulant therapy), IC.02.04.01 (infection prevention education) — with the EP-by-EP mapping to caption-quality evidence. How tracer methodology actually works: the surveyor follows a real patient back to the staff who cared for them, and from there to the training each one completed. The HR file review and what an evaluator literally pulls out of the folder. Five caption-related failure modes that recur in survey-finding documentation: missing captions for an employee with a documented accommodation, mangled drug INNs and pathogen names, out-of-sync caption files, captions that exist on the master but get silently dropped during LMS upload, and patient-facing video with no captions where the EP requires patient-and-family education. A 90-day pre-survey playbook covering the catalogue audit, the captioning RFP, the bulk back-catalogue retrofit, the HR-file-review prep, and the documentation lock. Plus a section on why the Joint Commission tracer lens differs materially from the OCR HIPAA investigation lens, and where vendor selection lands when both are in scope.
Technical Strategy · 2026-04-30
Prompting vs per-customer glossary models vs fine-tuning: how to actually decide for captioning in 2026
The decision-framework companion to the implementation post. Three real ways to specialise a captioning model: glossary-biased prompting (the floor — every team should pass), per-customer compounding glossary models (the durable middle path, where the glossary is the artifact and accumulates per-customer over hundreds of hours), and LoRA fine-tuning (the heaviest hammer, almost never the right answer for SMB). Cost, effort, and quality compared at three real volume tiers, with two worked examples — a 50-employee SaaS landing on path 2 at $1,400/year against $3,600 for hand-correction, and a 5,000-seat regulated-healthcare enterprise landing on path 2 + a narrow path-3 overlay on surgical narration. Plus a four-question buyer-evaluation flow that distinguishes vendors who run a real per-customer glossary architecture from those selling a feature as a product.
EU Compliance · 2026-04-30
Why the EAA is forcing EU SMBs to re-think captions by Q3 2026
The European Accessibility Act has been live since 28 June 2025. Why Q3 2026 is the inflection point — soft-enforcement window closing, first cross-border complaints accumulating, grace-period contracts running out, and member-state authorities staffing up to file their first periodic application reports under Article 30. The microenterprise carve-out under Article 4(5) is real, but only covers under-10-employee, under-€2M-turnover service businesses; the 10-49 employee tier is fully in scope. The technical floor is EN 301 549 V3.2.1 clause 7 (audiovisual services), referencing WCAG 2.1 AA SC 1.2.1, 1.2.2, 1.2.4 and 1.2.5. Five-step playbook for an EU SMB between now and Q3 2026, plus a member-state enforcement landscape walkthrough — Germany BFSG fine ceiling €100,000, France DGCCRF €75,000, Italy €40,000, Spain €600,000, Netherlands €900,000 or 1% of turnover.
Higher Education · 2026-04-29
How to pick a captioning vendor if you're a public university after ADA Title II
The buyer-journey for the procurement that follows the 2026-04-24 deadline. The five frameworks the contract has to satisfy at once (ADA Title II, Section 504, Section 508, IDEA Part D, FERPA), the higher-ed vendor field (Verbit, 3Play, Kaltura REACH, Panopto, AI-Media — plus Otter as the shadow-IT problem), the seven RFP questions, the realistic Year-1 budget shape ($80K–$300K at a mid-size public flagship, set against the $200K–$800K floor of a post-OCR consent decree), and where GlossCap fits as the back-catalogue + departmental-glossary layer the institution-wide MSA does not optimize for.
Procurement · 2026-04-29
Rev vs 3Play vs Verbit vs GlossCap: pricing breakdown for mid-market L&D teams
A side-by-side at three real volume tiers (5, 15, 30 hrs/mo). Per-vendor SKU detail with industry-typical rates, the synthesis table, the explicit break-even calculations (GlossCap Team beats Rev AI on dollars at 6.6 hrs/mo and any human-reviewed tier at under 1 hr/mo), the five cases where a competitor is actually the right answer, and the pre-answered procurement-objection chain you can forward to finance.
Healthcare · 2026-04-25
Captioning medical training video: why Whisper mangles drug names and how to fix it
A real audit of Whisper-large on a 12-minute pharmacology refresher: 200 substitution errors at default settings (87.6% accuracy), 79% of them clustered in five proper-noun categories — drug INNs, brand names, procedure terms, ICD-10/CPT codes, and uncommon anatomy. The same audio re-run with a 48-term glossary closed 154 of 158 proper-noun errors and brought accuracy to 99.4%. Includes a HIPAA workflow note, the four error classes glossary biasing won't fix, and a Mon-to-Fri sprint plan for medical L&D leads.
Operations · 2026-04-25
The hidden half-FTE in your L&D budget: video caption correction costs
The labour cost of in-house caption correction, run end-to-end at three real org sizes. A minute-by-minute audit of where the hour actually goes, why the second-order costs (time-to-publish drag, accessibility-statement risk, instructional-designer turnover) are usually larger than the labour line, and the break-even calculation that makes the buy-vs-keep-doing-this decision.
Technical · 2026-04-25
Glossary-biased captioning: how a Whisper prompt beats YouTube auto-captions on engineering terms
The implementation companion to the 99%-accuracy post. What Whisper's "previous-text" prompt actually does to beam search, how to fit a real engineering glossary into the 224-token context budget, when prompts help and when they cannot, and the minimal Python snippet that turned a 91.4% transcript into a 99.2% one.
Technical · 2026-04-25
Why 99% caption accuracy matters: the WCAG 2.1 AA threshold, with real training-video examples
The 99% accuracy figure cited in WCAG 2.1 AA audits comes from the DCMP Captioning Key, not from WCAG itself. We score the same nine-minute engineering onboarding clip two ways and find Whisper-default at 91.4% versus Whisper-with-glossary at 99.2%. The interesting part is which words make up the missing 7.8 points.
Compliance · 2026-04-24
ADA Title II just became enforceable — what training teams need to fix this week
The DOJ's Title II web-content rule went live 2026-04-24, for large state and local government entities. A 7-day sprint plan for the L&D lead at a public university or state agency: how to triage the back-catalog, where the audit actually starts, the vendor-call script, and the budget politics of the ask.
Reference pages
Shorter, reference-style pages on specific compliance standards, caption formats, LMS integrations, and vertical use cases:
- ADA Title II captions — the 2026-04-24 deadline
- WCAG 2.1 AA captions — the exact spec
- WCAG 2.2 AA captions — what changed
- SC 1.2.2 Captions (Prerecorded) explained
- WCAG captions vs transcripts
- EAA captions — the EU companion rule
- University lecture captions
- Compliance training video captions
- Medical training video captions
- Engineering onboarding video captions
Full site inventory in llms.txt.