LMS caption analytics and compliance reporting: how to measure caption coverage, learner engagement, and build the documentation layer that protects you in an OCR investigation

Why analytics matter three months after ADA Title II enforcement

ADA Title II’s digital accessibility requirements became enforceable on April 24, 2026 for the largest covered entities. The months since have not been quiet. OCR complaint investigations have opened; several voluntary resolution agreements have been negotiated and published; and the DOJ has issued supplemental guidance on what “good faith compliance efforts” means in enforcement practice. The critical lesson from the first OCR cycle is that good faith is evidenced by documentation, not by intent.

An L&D team that can say “we believe our caption coverage is around 85%” is in a different position from one that can say “our caption coverage as of June 30 was 87.3% by video count and 91.2% weighted by completion count; we have a documented remediation plan targeting 97% by September 30; our accuracy sampling programme confirmed 89 of 97 sampled videos passing the DCMP 99% threshold in Q2 2026; and our vendor SLA includes a remediation trigger at below 97% accuracy with a documented escalation protocol.” The second statement describes a compliance programme. The first describes an aspiration. OCR treats these differently.

For L&D teams that are already using data to manage their caption programme, this post is about structuring that data into the four-layer framework that makes it producible in an investigation. For teams that have been relying on good intentions and a vendor invoice, this post is about building the infrastructure that turns caption management from a production task into a compliance programme.

The three questions every compliance officer needs to answer — what percentage of your library is captioned, are those captions accurate to WCAG standard, and are learners actually using them — require data from three separate systems that most organizations have never connected. The fourth question, can you prove it, requires a documentation layer that most organizations have never built. This post covers all four.

The measurement gap: why your LMS coverage report is not a compliance report

Every major LMS platform has a caption coverage report of some kind. In Cornerstone, it is a field in the Learning Objects report. In Workday, it is the Media Captions column in the Content Library. In Docebo, it is a filter in the advanced reporting module. In all of these platforms, the field is binary: a caption file is either attached or it is not.

This binary tells you almost nothing about compliance. The following are all counted as “captioned” in a standard LMS coverage report:

• A machine-generated caption file at 68% WER on a pharmaceutical compliance module where drug names are consistently wrong.
• A human-reviewed caption file that was accurate at delivery three years ago but has not been updated since the product was rebranded and fourteen key terms changed.
• A caption file that is attached to the video record in the LMS but renders as a blank track in the iOS Safari player because of an encoding incompatibility that was never caught in QA.
• A caption file in the wrong format (SRT attached to a platform that requires VTT) that the LMS ingested without error but that the learner-facing player ignores.

The WCAG 2.1 AA standard under Success Criterion 1.2.2 requires captions that are “accurate” — a term that the DCMP Captioning Key operationalizes as a Word Error Rate at or below one percent (99%+ accuracy). An L&D team reporting 91% caption coverage to their VP is almost certainly reporting caption file presence, not WCAG-compliant caption accuracy. The delta between those two numbers — the gap between captions-present and captions-compliant — is typically twenty to forty percentage points on a mixed content library where some material was auto-captioned without review.

The measurement gap has a second dimension beyond accuracy: engagement. An organization that has captioned 90% of its library and has zero data on whether learners are using those captions cannot make an effectiveness argument, cannot identify whether specific learner populations are experiencing accessibility barriers, and cannot answer the question that most DEI officers eventually ask: is the programme actually serving the employees who need it, or is it serving the audit spreadsheet?

Closing the measurement gap requires building four data layers. The platforms provide data for three of them, though not always in the place you expect. The fourth layer — documentation — has to be built intentionally, because it does not emerge from platform reporting alone.

The four data layers in a mature caption programme

A mature caption analytics programme operates across four data layers, each answering a different question:

Layer 1: Coverage

What percentage of your training video library has a caption file attached, broken down by content type, date range, platform, and priority tier? This is the question your LMS can partially answer. It cannot answer the accuracy sub-question, but it can tell you the scope of the coverage obligation.

Layer 2: Accuracy

Of the content that is captioned, what percentage meets the WCAG 99% accuracy threshold when measured by an independent DCMP-protocol sampling programme? This is the question your LMS cannot answer at all. It requires a separate QA programme with a defined sampling methodology, a measurement standard, and a remediation trigger.

Layer 3: Engagement

Are learners actually using captions? What percentage of sessions include active captions? Do caption usage rates differ by device type, learner segment, or content type? Is there a measurable relationship between caption access and module completion? This question is partially answerable from LMS analytics and more fully answerable from video platform analytics. It is the question that closes the ROI loop.

Layer 4: Documentation

Can you prove, with dated records, that you ran a compliance programme? Do you have a written policy, a vendor contract with accuracy provisions, a remediation plan, an accommodation request log, and QA sign-off records? This is the layer that matters most in an OCR investigation and the layer that most L&D teams have not built. It does not exist in the LMS reporting interface. It has to be assembled from multiple systems and maintained as a compliance record, not a project management artefact.

Each layer informs the next. Coverage data drives the remediation backlog. Accuracy data informs vendor conversations and QA resource allocation. Engagement data closes the ROI loop and surfaces learner populations whose compliance experience differs from the aggregate. Documentation is the layer that makes all the others legible to an outside reviewer, whether that reviewer is an OCR investigator, a FINRA examiner, or a board member asking what the L&D team has done to reduce accessibility liability since the April 2026 deadline.

Layer 1: Caption coverage metrics

Coverage measurement has five components, and most L&D teams only track the first one.

Total coverage rate

Total coverage rate is the simplest metric: captioned videos divided by total videos, expressed as a percentage. It is useful for setting a baseline and tracking aggregate progress over time. It is not useful for compliance reporting without the three disaggregations below, because it treats a mandatory quarterly compliance module that every employee in the organization must complete the same as a ten-minute optional professional development video that was viewed by three people in 2022.

New-content coverage rate

The new-content coverage rate measures what percentage of videos published in the past thirty days were captioned within your policy SLA. If your caption policy commits to captioned delivery within three business days of video publication, the new-content coverage rate is the percentage of videos published in the past thirty days that received a caption file within three business days. This is the forward-looking control metric. Total coverage rate tells you where you are; new-content coverage rate tells you whether your programme is keeping up with production.

Organizations that focus exclusively on back-catalogue remediation while continuing to publish uncaptioned content are running a programme that structurally cannot reach 100% compliance. If your production team publishes fifteen videos per month and your caption workflow has a five-day backlog, you are creating fifteen new compliance gaps every thirty days. The new-content coverage rate surfaces this gap immediately; total coverage rate buries it in the aggregate.

Back-catalogue remediation progress by priority tier

Back-catalogue remediation should be organized by priority tier, not by publication date. The standard LMS caption audit methodology organizes the remediation backlog into four tiers based on risk: Tier 1 (required for active enrollment within sixty days, mandatory compliance modules with current assignment), Tier 2 (role-required content in the six-month enforcement window), Tier 3 (general catalogue, optional content), and Tier 4 (archived content, assess before investing). Coverage reporting by tier tells you whether you have completed the most legally exposed content first, which is the remediation sequencing that OCR resolution agreements consistently require. A 95% total coverage rate that has not yet addressed Tier 1 mandatory compliance content is a worse compliance position than a 70% total coverage rate that has remediated 100% of Tier 1.

Age-weighted coverage rate

The age-weighted coverage rate adjusts the coverage calculation to reflect learner exposure rather than content count. For each video in the library, multiply its captioning status (1 or 0) by its completion count, sum across all videos, and divide by total completion count. This gives you coverage weighted by where learners are actually spending time, not where content happens to exist in the library catalog.

An organization with a ten-thousand-item catalog may have three hundred videos — three percent of the library — that account for sixty percent of all completions. If those three hundred videos are captioned, the organization’s compliance exposure is much lower than its raw coverage rate suggests. Conversely, an organization with 95% total coverage rate may have left uncaptioned its most-accessed new hire onboarding sequence and its annual harassment prevention module — the two programs every employee touches every year. Age-weighting makes this visible.

Platform-level disaggregation

Multi-platform organizations — those using Cornerstone for compliance training, Panopto for lecture capture, Teams for informal learning, and perhaps a separate external academy on Skilljar — have different coverage rates per platform. Aggregating these into a single coverage figure conceals the platform-level risk profile. A 94% coverage rate in Cornerstone combined with 45% coverage in Panopto and 8% coverage in Teams recordings represents very different compliance exposure depending on what content lives where. The platform-level disaggregation should be a standard row in the quarterly coverage report.

Layer 2: Caption accuracy measurement

Accuracy measurement is the layer that most L&D teams have not built, and it is the layer that creates the largest gap between perceived and actual compliance. The LMS cannot measure it. The vendor should not be trusted to self-report it. The only number that protects you in an enforcement investigation is an accuracy figure produced by a defined measurement protocol that you can explain to an investigator.

The measurement standard

The DCMP Captioning Key is the measurement standard for caption accuracy in L&D contexts. It defines accuracy as a Word Error Rate (WER) calculation: (Substitutions + Deletions + Insertions + Timing Errors) / Total Reference Words × 100. WCAG 2.1 AA compliance requires WER at or below one percent (99%+ accuracy) for synchronized captions. The DCMP standard counts errors in four categories, each with different root causes: substitutions (wrong word transcribed), deletions (word present in speech but absent from caption), insertions (word in caption not spoken), and timing errors (caption display offset by more than two seconds from the corresponding speech). To understand more about the error-counting methodology and what each error type reveals about your captioning workflow, the caption quality error rate calculator post walks through the DCMP formula with worked examples at different content types and difficulty levels.

Sampling methodology

You do not need to QA every caption file — a statistically defensible sample is sufficient for both operational monitoring and OCR response. OCR resolution agreements typically call for quarterly review covering a random sample of at least ten percent of the active captioned library, stratified by content type (compliance/regulatory, technical, soft skills), by production method (AI-generated without review, AI-generated with human review, human-generated), and by vendor or production source. Stratification ensures that your sample reflects the full accuracy distribution of your library, not just the easy-to-produce content types that tend to perform well.

For each sampled video, the QA process requires a reference transcript — a word-for-word transcription of the audio produced by a human reviewer before the caption file is scored. This reference transcript is the benchmark against which WER is calculated. Producing the reference transcript is the step most organizations skip, and skipping it makes the resulting accuracy measurement meaningless. A vendor’s claim that their captions are 99% accurate on your content is only verifiable if you have the reference transcript to compare against. The vendor accuracy evaluation methodology post covers reference transcript preparation and the pre-contract evaluation process; the same methodology applies to ongoing quarterly sampling in a live programme.

What to track in the accuracy spreadsheet

For each sampled video, your accuracy record should include: video ID, title, content type (compliance, technical, soft skills), production method, vendor or generation source, original caption delivery date, QA sample date, WER score as calculated by DCMP protocol, error type breakdown (substitution %, deletion %, insertion %, timing %), pass/fail result at the 99% DCMP threshold, and, where applicable, re-submission date and post-remediation WER. This record should be maintained as a compliance document with version history, not as a transient project management task.

Accuracy trending and vendor performance monitoring

Run the quarterly sample consistently and plot accuracy trend lines by vendor and by content type. A vendor whose accuracy on technical content has declined from 98.4% to 95.2% over six quarters is showing a trend that requires intervention before contract renewal, not after it. The feedback loop post describes how to use accuracy trend data to drive vendor improvement conversations and glossary maintenance decisions; the accuracy measurement programme described here is the data source that makes those conversations possible. Without a consistent, methodology-defined accuracy record, you are negotiating accuracy improvement based on anecdote, and vendors have no incentive to respond.

Error type distribution as a diagnostic tool

The distribution of error types tells you more than the aggregate WER does. A high substitution error rate on a single content type — say, 4.1% substitutions on pharmaceutical regulatory content vs. 0.7% on general HR content — points directly to a glossary gap or a training data mismatch. A high timing error rate across all content types from a single vendor points to a delivery workflow problem in the encoding step, not a transcription problem. A high deletion error rate on content recorded with a specific microphone model points to an audio quality issue that affects ASR input. Using error type distribution diagnostically, rather than just tracking aggregate WER, lets you direct remediation investment to the root cause rather than to re-captioning content that will fail again for the same reason.

Layer 3: Learner caption engagement analytics

Engagement analytics for captions answer a set of questions that coverage and accuracy analytics cannot: are learners actually using captions, which learners are using them, and does caption access produce a measurable change in learning outcomes? These questions matter for compliance (to demonstrate that your programme is accessible in practice, not just on paper) and for budget (to close the ROI loop that coverage and accuracy data alone cannot close).

Caption toggle rate

Caption toggle rate is the percentage of learner video sessions during which captions were active. It is the most direct measure of caption usage and the most difficult to obtain from standard LMS reporting. Platforms that track caption toggle events natively: Kaltura (Advanced Analytics module), Panopto (Viewer Analytics module), Canvas Studio (formerly Arc). Most LMS platforms — Cornerstone, Workday Learning, Docebo, TalentLMS — do not track caption toggle events in their standard reporting. If your LMS hosts video directly rather than embedding from Kaltura or Panopto, you may not have access to caption toggle data at all.

Where toggle data is available, a caption-on rate of eight to fifteen percent is typical for general corporate training content. Technical training content at organizations with significant ESL populations commonly shows caption-on rates of twenty-five to forty percent. Mandatory compliance training (which employees are required to complete, often under time pressure and often in non-ideal listening environments) shows the highest caption-on rates, typically fifteen to thirty percent, for the same module. If your caption-on rate for a critical compliance module is below five percent, the problem may not be lack of need — it may be that the caption control is not discoverable in the player UI, or that captions are not loading correctly on the most common device types in your learner population.

Completion rate differential

If you have learner groups with meaningfully different caption access situations — for example, a group using an LMS player where captions render correctly and a group using an embedded player where captions do not render on iOS — comparing completion rates across those groups on the same module gives you a measurable proxy for caption impact. More commonly, compare completion rates between learners who have formally requested caption accommodations and the general population on the same modules. If the accommodation-request population shows lower completion rates, that is a signal that either captions are not being delivered at the right quality, or the delivery mechanism (a separate captioned version, a link to an external file) introduces friction that reduces completion. The learning outcomes research post provides the academic framework for understanding why caption access should correlate with completion rates even for hearing learners, particularly on technical content and for ESL populations.

Device type distribution

What percentage of your learner sessions come from mobile devices? Research consistently shows that forty to sixty percent of mobile video sessions are completed in audio-absent environments — commuting, open-plan offices, anywhere where headphone use involves a social or logistical cost. In those environments, captions are the primary verbal channel, not a supplement to audio. An L&D programme that has captioned all content but whose LMS mobile player renders captions in a six-point white font on a white background is not, in practice, serving the mobile learner population with captions at all. Mobile device share in your completion data tells you the maximum size of the population for whom caption rendering quality is the determining factor in module accessibility.

Accommodation request volume as a floor, not a ceiling

Formal accommodation requests tell you the number of learners who self-advocated. They do not tell you the number of learners who needed captions and did not ask. Research on disability self-advocacy in employment contexts consistently finds that formal request rates understate the population with accessibility needs by a factor of five to ten. An organization with fifty accommodation requests for captions in a calendar year has, as a reasonable estimate, two hundred fifty to five hundred employees for whom caption access meaningfully affects their ability to engage with training content.

Accommodation request volume trends are useful even in the absence of absolute population data. A rising request volume after a caption programme launch suggests growing awareness of the programme and growing confidence that accommodation requests will be honored — both positive signals. A sustained zero accommodation request volume in an organization with a significant DHH employee population or a multilingual workforce is a red flag: it suggests the accommodation request process itself is a barrier, not that the population does not need captions. Understanding the difference between these two signals requires tracking accommodation request volume against programme launch timing, communication events, and workforce demographics — not just counting requests in isolation.

For the documentation requirements around accommodation requests, see the accommodation request timeline post, which covers the interactive process documentation obligation and three-year retention requirements in detail.

ESL learner segment and language-based disaggregation

If your LMS or HR system includes learner language preference or country of employment, disaggregating caption usage by this variable consistently reveals the largest caption user population: employees for whom English is not a first language and who are engaging with technical or compliance content in English. The multilingual caption workflow post covers the organizational infrastructure for serving multilingual learner populations; the analytics framework here gives you the measurement side — the data that tells you whether your captioning investment is reaching the ESL population and at what quality level.

Where LMS or HR systems do not include language data, a proxy approach: segment completion rates and time-in-module by country of employment or time zone. Employees in non-English-speaking markets completing English-language compliance training typically show higher time-in-module (more replays, more segment navigation) and lower completion rates on uncaptioned content than their English-native peers on the same module. Adding captions reliably closes both gaps, and the closing of the gap is measurable in your LMS completion analytics without requiring explicit language data.

LMS-specific reporting: where to find each metric

The platforms below cover the majority of the L&D video ecosystem. For each platform, the coverage field location, the accuracy limitations, and the engagement analytics availability are documented as of mid-2026. Platform reporting interfaces change with product releases; specific field names may differ from your instance configuration. The approach described is correct even if a field is renamed.

Cornerstone OnDemand

Coverage: In Cornerstone Edge (the current reporting interface, replacing classic Crystal reports), run a Learning Objects report or a Curriculum Objects report filtered by Object Type = Online Course or Media. Add the “Accessibility” or “Closed Captions” column from the Media category in the column selector — the exact label varies by instance configuration. A value of “Yes” or “1” indicates a caption file is attached to the learning object. Export to CSV, group by caption status, and calculate coverage rate. For new-content coverage rate, add a Creation Date filter for the past thirty days.

Accuracy: Cornerstone OnDemand does not have a native accuracy field for caption files. If Cornerstone auto-generates captions via its AI Caption feature (available in Cornerstone Learning), the generated captions are typically in the 78–86% WER range on technical content — well below WCAG compliance. The coverage report will mark these as captioned without any accuracy indicator. Your accuracy data must come from an independent sampling programme maintained outside Cornerstone.

Engagement: Cornerstone’s standard Transcript Report and Training Summary Report do not include caption engagement fields. The Learning Assignment Completion report captures completion date, score, and time-in-module, which you can use to compare completion rates across learner cohorts. Caption toggle event tracking requires Cornerstone’s Media Advanced or Cornerstone Analytics Premium packages; on standard licensing, this data is not available.

Documentation: Accommodation requests should be logged in Cornerstone’s HR Case Management module (if licensed as part of Cornerstone HR) or in a separate accommodation log maintained outside the LMS. The critical documentation fields are: request date, learner ID, content ID, request description, response date, action taken, and outcome. These records require three-year minimum retention.

Workday Learning

Coverage: Workday Learning’s Content report (run from the Workday reporting task, search for “Learning Content”) includes a Media Captions field for each content item when captions are attached. The field reflects caption status as recorded at the content level in Workday Learning; it does not aggregate sub-content caption status (a learning program containing ten individual videos with mixed caption status will show the program-level flag, not a per-video count). For a true library-wide coverage count, you need to run the report at the learning content item level, not the program level, and filter for media type = video or media.

Accuracy: Workday Learning does not have a native caption accuracy field. Captions in Workday Learning are typically delivered from the source video platform (Brightcove, Kaltura, Panopto, or Vimeo) and Workday Learning does not inspect caption file quality. Your accuracy data must come from the source platform or from an independent sampling programme.

Engagement: Workday Learning’s Learner Activity report provides completion date, completion status, and time-in-module. It does not track caption toggle events. For caption engagement data for Workday Learning content hosted on Kaltura or Panopto, pull the engagement analytics from those platforms and join on video ID. This integration is manual unless you have a data pipeline that connects Workday Learning activity data with Kaltura or Panopto event data.

Docebo

Coverage: Docebo does not have a platform-wide caption coverage report in the standard interface. In the Course Management area, each course’s Content tab shows whether individual media items have caption files attached. For library-wide coverage reporting, use either (1) the Docebo API (v1 /manage/v1/courses endpoint returns course content metadata, including caption presence for content uploaded directly to Docebo with captions attached), or (2) the Advanced Reporting module (requires the Advanced Reporting add-on license) to build a custom report filtering learning objects by media type and caption status. The Advanced Reporting approach is more sustainable for ongoing monitoring; the API approach gives more granular data but requires technical implementation.

Accuracy: Docebo does not have a native caption accuracy measurement capability. If content is hosted in Docebo’s native video player, caption accuracy measurement must be conducted outside the platform. If content is hosted in Kaltura integrated with Docebo, use Kaltura’s caption accuracy tier as an initial proxy (machine tier, enhanced tier, professional tier) and run DCMP sampling on the machine-tier population.

Engagement: Docebo’s Learning Impact module (where licensed) provides learner engagement at the individual content-item level, including play events and completion events. Docebo’s native player does not natively record caption toggle events. If you use Docebo with Kaltura Video Portal, Kaltura Advanced Analytics captures caption-on events at the player level and can be queried separately from the Docebo engagement data via the Kaltura Analytics API.

TalentLMS

Coverage: TalentLMS does not have a native caption coverage report. To obtain library-wide caption status, use the TalentLMS API (GET /api/v1/courses returns course metadata; GET /api/v1/courses/:id returns unit-level content including caption file status for video units where captions were uploaded to TalentLMS directly). A Python or JavaScript script that calls the courses API and returns a flat list of video units with caption status is the standard approach; this can be scheduled to run weekly and write results to a Google Sheet or similar tracking tool. Coverage rate is then calculated in the sheet.

Accuracy: TalentLMS does not track caption accuracy. The platform accepts SRT and VTT caption uploads without validation. Caption accuracy measurement must be conducted via an independent sampling programme.

Engagement: TalentLMS does not track caption toggle events in its standard reporting. The best available proxy is comparison of completion rates and time-in-module across learner groups with different caption access profiles — for example, comparing a group in a market where your caption delivery is consistent against a group in a market where a technical issue has affected caption rendering. Without group-level variation in caption access, completion rate comparison is not a clean signal for caption engagement.

Kaltura (KMC / KMS and Kaltura LMS)

Kaltura functions both as a video management platform embedded in other LMSs and as a standalone LMS. Reporting capabilities differ by deployment mode.

Coverage (as video platform): In the Kaltura Management Console (KMC), select the Content tab and use the filter for “With Captions” to get a filtered list of entries with caption tracks. For library-wide coverage, use the Kaltura API (captionAsset.list filter with the entryIdIn parameter) to retrieve caption asset status across the full entry library. The Kaltura KMS Caption Report (under Analytics > Reports > Caption in KMS) provides an aggregated view of captioned vs. uncaptioned entries.

Accuracy (REACH integration): If you use Kaltura REACH for caption generation, the caption asset metadata includes the REACH accuracy tier: Machine (typically 73–86% WER on technical content), Enhanced (typically 93–97% WER), or Professional (99%+ when human-reviewed to DCMP standard). Use tier as an initial classifier for the accuracy sampling plan: Machine-tier content should be sampled at 25–30% quarterly; Professional-tier content at 5–10%. Note that REACH tier labels reflect the production method, not an independent WER measurement of the delivered caption file — they are a proxy, not a measurement. Professional-tier captions can still fail DCMP 99% if the vendor’s process had quality problems. Independent sampling is still required. For more detail on building an accuracy evaluation programme for Kaltura REACH output, the QA methodology post covers the spot-check protocol and error taxonomy in depth.

Engagement (Advanced Analytics): Kaltura Advanced Analytics (requires Advanced Analytics add-on or Kaltura Backstage access) tracks the Caption engagement event — specifically, the percentage of session time during which captions were active. This is the most direct caption toggle data available in the mainstream LMS ecosystem. The Kaltura Analytics API allows you to query caption-on rate by media entry, by user segment, by device type, and by date range. For organizations that want to report on caption usage across their learner population without manual data assembly, Kaltura Advanced Analytics is the most capable native option available.

Panopto

Coverage: Panopto’s Admin Console includes an Accessibility Report (Admin > Sessions > Accessibility or Site Admin > Sessions, filtered by Caption Status). The report shows all sessions with caption provider (Panopto ASR, Rev.ai integration, Amara import, or manually uploaded SRT/VTT), caption status (enabled or disabled), and session folder path. Export to CSV for library-wide coverage calculation. Coverage rate = captioned sessions / total sessions × 100, segmented by folder (which typically corresponds to course, department, or content owner).

Accuracy: Panopto generates ASR captions for all sessions by default. Panopto ASR accuracy on technical content is typically 78–87% — not WCAG compliant without human review. In the Caption Management area, sessions show a Last Modified date for the caption track. Sessions where the caption track was last modified significantly after the ASR generation timestamp have likely received human review. Sessions where Last Modified equals or closely follows the creation timestamp are likely ASR-only. Use this date differential as an initial classifier for the human-reviewed vs. ASR-only populations, then run DCMP sampling on each population at different rates. The auto-captions compliance status post covers in detail why ASR-default captions at Panopto’s typical accuracy range do not meet the WCAG 1.2.2 standard, even when the LMS marks them as captioned.

Engagement: Panopto’s Viewer Analytics module (Admin > Reports > Viewer Activity) tracks per-session viewer events including play, pause, seek, and caption toggle. The caption toggle event (captionsEnabled = true in the session timeline) allows you to calculate caption-on rate per session and per viewer over any date range. The Panopto API (statistics endpoint) allows programmatic extraction of viewer analytics including caption events at scale for large libraries.

Canvas LMS

Coverage: Canvas LMS does not have a native caption coverage report for videos embedded in course content. Caption coverage in Canvas depends entirely on the video hosting platform. For Canvas Studio (formerly Arc): caption status is visible in the Studio Admin interface and accessible via the Studio API (GET /api/v1/media_objects returns media with caption status). For Kaltura-Canvas integration: use Kaltura KMC reporting as described above. For Panopto-Canvas integration: use Panopto’s Accessibility Report. For videos linked from YouTube or Vimeo: caption status must be verified at the source platform. A Canvas course with four different video hosting sources has four separate coverage reporting workflows, which is the main reason multi-platform Canvas environments tend to have highly variable caption coverage across courses.

Engagement: Canvas does not track caption engagement events natively for embedded video. The engagement analytics live in the hosting platform: Canvas Studio (Studio engagement report includes caption usage), Kaltura (Advanced Analytics), or Panopto (Viewer Analytics). For Canvas Data 2 customers, the Canvas Data 2 schema includes an attachments table with a caption_status field for media elements in Canvas Studio — this is the most direct programmatic path to library-wide caption coverage data for Studio-hosted content.

Building the compliance dashboard

A compliance dashboard for caption analytics has two audiences with different information needs, and designing for both in a single report is the most effective way to get the programme the visibility it needs.

The legal and compliance audience needs: coverage rate (with accuracy filter applied), remediation trajectory, accommodation request status, vendor SLA performance, and a pre-flight assessment of what documentation would be available if a complaint arrived today. The learning effectiveness audience needs: completion rate differential, caption toggle rate by content type, mobile learner engagement, and ESL learner segment performance. These audiences often cannot share a report because the information density overwhelms the non-technical reader; the solution is a one-page traffic-light summary that links to the detailed report for each audience.

The eight metrics for the compliance summary

1. Total compliant coverage % — captioned AND accuracy-verified videos as a percentage of total library (not just captions-present)
2. New-content coverage rate — % of videos published in the past 30 days captioned within the policy SLA window
3. Back-catalogue remediation progress by tier — Tier 1 % complete, Tier 2 % complete, estimated completion dates
4. Accuracy pass rate — % of sampled captions passing DCMP 99% threshold in the most recent quarterly sample
5. Caption engagement rate — caption-on rate for the most-accessed content type (where data is available from Kaltura or Panopto)
6. Open accommodation requests — count of open requests, % resolved within policy SLA, median resolution time in business days
7. Vendor accuracy trend — 6-quarter accuracy trend line by primary vendor (improving / stable / declining, with most recent WER)
8. Remediation backlog projection — at current remediation pace, months to 100% Tier 1 completion; months to 100% Tier 2 completion

Traffic-light thresholds for each metric should be set in your caption policy and published to the team. An example threshold set: Green = new-content coverage ≥99% within SLA, accuracy pass rate ≥90%, zero open accommodations outside SLA. Yellow = new-content coverage 90–98%, accuracy pass rate 75–89%, accommodations open but all within SLA. Red = new-content coverage <90%, accuracy pass rate <75%, any accommodation outside SLA, or vendor accuracy trend declining for three or more consecutive quarters.

The OCR response pre-flight

The most valuable use of the compliance dashboard is as a pre-flight checklist for the document request scenario. OCR requests consistently ask for eight categories of evidence within thirty to forty-five days of complaint notice. Before your next quarterly review, run this checklist and note which documents you can and cannot produce:

1. Written caption policy with an effective date predating the complaint period
2. LMS coverage report as of the period covered by the complaint
3. Vendor contract with accuracy SLA, measurement protocol, and remediation clause
4. Accommodation request log with resolution dates and outcomes for the relevant period
5. Remediation plan and progress documentation
6. Accuracy sampling records (DCMP-protocol QA reports for the relevant quarters)
7. L&D team training records showing staff were trained on the caption policy
8. Back-catalogue audit results showing how the remediation backlog was identified and sequenced

An organization that can produce all eight is demonstrably running a compliance programme. An organization that can produce one or two is in the position of defending a complaint without evidence. The pre-flight does not require a compliance investigation to be useful — it identifies, right now, which documentation gaps to close before a complaint arrives. For the detailed framework on what OCR document requests look like in practice, the ADA Title II enforcement reality check post walks through the investigation process and the types of voluntary resolution agreements that have resulted from the first wave of post-April-2026 enforcement activity.

Using caption analytics to close the ROI loop

Caption analytics programmes that only produce the compliance dashboard leave three significant arguments on the table: learning effectiveness, HR inclusion, and financial ROI. Each argument requires a different cut of the same underlying data.

The compliance argument (legal/risk)

Coverage rate × your organization’s ADA Title II exposure estimate × probability-of-complaint multiplier = expected value of non-compliance. This is the risk-reduction framing that your General Counsel understands and that budget committees respond to. The ROI framing for finance audiences post builds this calculation through the three budget buckets (vendor cost, correction labour, infrastructure), the seven-row business case template, and the Finance objection responses. The caption analytics programme provides the numerator: the coverage rate and accuracy rate that let you calculate what percentage of the risk has been addressed and what percentage remains.

The learning effectiveness argument (CLO/Learning VP)

Caption toggle rate × completion rate differential × learner count by affected segment = measurable learning-outcome impact. If thirty-two percent of your learners in the ESL population use captions on technical onboarding content, and captioned completions in that population run sixteen percent higher than uncaptioned completions, you have a sixteen-percent completion uplift for thirty-two percent of your highest-growth employee segment — from a programme that was already being funded for compliance reasons. This is the argument that earns caption programme budget renewal in years two and three, when the legal urgency of year one has receded. The learning outcomes research post provides the academic framework (dual-coding theory, cognitive load, ESL outcome meta-analyses) that supports the completion-rate claim; the analytics framework here provides the organization-specific data that grounds the abstract research in your actual learner population.

The HR inclusion argument

Accommodation request close rate + ESL learner caption usage + disability population engagement = accessibility programme evidence. Organizations with strong DEI commitments increasingly want data that their L&D programme is accessible in practice — not just that a caption policy exists on a SharePoint site. The accommodation request log (closed within SLA for every request in the past year), the ESL learner completion rate differential (documented in LMS analytics), and the caption toggle data for populations with disclosed disabilities are the three datasets that convert a compliance programme into an inclusion programme for DEI reporting purposes.

The financial ROI argument (CFO)

Total programme cost (vendor + correction labour + infrastructure from the budget planning guide) divided by (EV reduction + learning-outcome value + HR equity programme value) = ROI multiple. A programme running at $38,000 per year in all-in costs that reduces a modelled $86,000 per year in compliance expected value, improves completion rates by eleven percent for a 340-person ESL population (worth approximately $18,000 in avoided re-training cost at two hours per incomplete), and produces a DEI reporting data point that supports the annual report has an ROI multiple well above one before the learning-effectiveness argument is fully monetized. The analytics layer is what makes the inputs to this calculation observable rather than theoretical. Without it, the CFO is asked to renew a budget line item based on a risk argument that becomes harder to make as time passes without an incident — which is exactly when the programme is working.

Eight failure modes in caption analytics

1. Treating the LMS “captioned” flag as an accuracy evidence

The LMS marks content as captioned when a caption file exists. It does not measure accuracy, cannot detect blank tracks, cannot identify encoding-incompatible files that the player ignores, and does not know whether the file was last updated before a major terminology change in the content area. A programme that reports LMS coverage rate as its compliance metric is reporting file presence, not WCAG compliance. The delta between those two numbers is typically twenty to forty percentage points on a mixed-production-method library. OCR investigators are increasingly aware of this delta and are asking, in document requests, not just for coverage reports but for accuracy measurement records. A coverage rate without an accuracy measurement programme attached to it is an incomplete compliance claim.

2. Not running an independent accuracy sampling programme

Vendor-reported accuracy is not independently verified accuracy. Vendors measure accuracy on their test sets, using their own methodology, on content that is often not representative of your library. The only accuracy figure that supports a compliance claim is one produced by a defined protocol (DCMP Captioning Key), on a random sample of your actual content, conducted by your QA team or an independent evaluator. Without this, you cannot answer the OCR investigator’s question “how do you know your captions are accurate?” with anything other than “our vendor says so” — which is not a satisfactory answer. The 99% accuracy post explains why the DCMP threshold is the operative standard and how it differs from vendor accuracy marketing claims.

3. Coverage rate that does not weight by learner exposure

A library where 7% of content is uncaptioned can be a minor remediation task or a critical compliance gap, depending on which 7%. If the uncaptioned content includes the three courses every new hire completes in their first month, the legal exposure is concentrated. Age-weighting the coverage rate by completion count surfaces this concentration. Reporting an unweighted coverage rate to leadership and legal counsel creates a false sense of the risk distribution.

4. No new-content coverage rate tracking

Organizations that track remediation progress but not the forward coverage rate for new content cannot determine whether their programme is making net progress or just running in place. At a production rate of fifteen videos per month with a three-week caption backlog, total coverage can remain flat or decline even while the remediation plan is nominally on track. The new-content coverage rate is the only metric that tells you whether your caption workflow is keeping up with content production. For the operational mechanics of building a caption workflow that maintains a high new-content coverage rate, the building a caption compliance programme post covers the intake process, vendor coordination, and LMS delivery workflow.

5. Accommodation requests tracked in email and Slack

An accommodation request that is processed through an informal email chain has been responded to operationally but not documented for compliance purposes. The interactive process documentation obligation under ADA Title II and Section 504 requires a dated record of: (a) the request, (b) the organization’s response, (c) any interim accommodation provided, (d) the accommodation provided, and (e) the outcome. Email chains buried in an L&D coordinator’s inbox do not meet this standard. When an OCR document request arrives and asks for accommodation request logs for a specific period, the organization that has been processing requests via email cannot produce what is being asked for, regardless of how responsively those requests were handled.

6. No documentation retention policy for caption records

Caption files, vendor delivery confirmations, QA sign-off records, accuracy audit results, and accommodation request logs are compliance records. ADA record-keeping guidance under 29 C.F.R. § 1602.14 requires retention of employment practice records for at least three years. When a complaint covers a period more than three years in the past, the exact scope of the retention obligation depends on the specific legal theory and jurisdiction — but an L&D team that routinely discards caption delivery records after project closeout has eliminated its ability to reconstruct the compliance picture for any period more than a year or two in the past. The caption records and e-discovery post covers the litigation hold and document retention scope in the civil litigation context, which sets the outer boundary for what your retention policy should cover even absent active litigation.

7. Analytics that do not disaggregate by learner segment

An aggregate caption analytics report that shows 91% completion on a mandatory compliance module is not useful for identifying whether a specific learner population — the customer support team in Manila, the warehouse operations staff, the recently acquired engineering team with significant non-native English speakers — is experiencing the module differently. Segment-level disaggregation is the mechanism that surfaces accessibility gaps before they become accommodation requests or, worse, complaints. A programme that tracks only aggregate completion cannot answer the question “is the caption programme serving the employees who most need it?” It can only answer “is the aggregate completion rate acceptable?” Those are different questions with different answers and different implications.

8. Reporting coverage without a remediation projection

Coverage at a point in time is less meaningful than coverage trajectory. An organization at 60% coverage with a documented ninety-day remediation plan to reach 95% is in a fundamentally different compliance position from one at 60% coverage with no plan. The compliance dashboard should always pair the current coverage rate with a projection line: at the current remediation pace, when does Tier 1 reach 100%? When does Tier 2 reach 100%? What is the expected completion date for the full library? This projection is the evidence of a programme, not just a state. It is what makes voluntary resolution agreements defensible — an OCR settlement that requires 90% coverage by a specific date is only possible if you can demonstrate that you have a plan and the capacity to execute it. For building and documenting the remediation plan, the caption compliance self-assessment checklist provides the scoring framework that gives the plan its starting point, and the maturity model post describes the five-level progression that gives the plan its target.