Caption file lifecycle management: how to handle captions when training video is updated, rebranded, or archived — staleness triggers, re-captioning vs. glossary reprocess, LMS propagation, and audit archive

The four trigger types that make caption files go stale

Caption staleness is not a random event. It follows predictable patterns that correspond to specific categories of content change. Understanding the trigger type — not just the fact that a video was updated — determines the urgency of the remediation response, the right remediation path, and the audit exposure created by the gap between video update and caption update. Four trigger types account for the overwhelming majority of caption staleness in enterprise L&D content libraries.

Trigger type 1: Rebrand and rename events

Product acquisition, company merger, and brand refresh are among the most common sources of caption staleness in large content libraries, and they are uniquely difficult to detect through routine content management because the change often happens at the brand or product level rather than at the content level. When a product that appears in training audio gets renamed — “Acme Platform” becomes “GlobalCorp Workspace” after acquisition, for instance — every training video that references the old product name becomes inaccurate. The caption file repeats the old name wherever it appears in the audio. A learner watching the updated course slides (which have been refreshed to show the new product name) while reading captions that still say “Acme Platform” experiences a visible, jarring discrepancy between the visual content and the text they are reading.

In a large LMS, a product rename can create caption staleness across dozens or hundreds of courses simultaneously: onboarding modules that introduce the product, product training videos that demonstrate features by name, sales enablement content that uses the product name in role-play scenarios, HR benefit enrollment guides that reference the product line, and compliance training that cites the product’s regulatory classification. The breadth of the staleness is the challenge: a rebrand event is not a change to one video, it is a change to every video that contains an audio reference to the old product name or brand term, which in a mature content library may be very large.

Rebrand events are particularly high-risk for caption programmes because the inaccuracy is visible and self-evident. Unlike a regulatory change (where a learner may not know the current rule well enough to notice that the captioned version describes a superseded one), a brand mismatch is immediately apparent to any learner who is aware of the rebrand. If the learner is a Deaf or hard-of-hearing employee who relies on captions as their primary access to training audio, they are also likely to report the discrepancy as an accommodation failure rather than a mere content inconsistency.

The vocabulary-level fix — updating the term in the caption glossary and reprocessing the affected files — is usually the right remediation path for rebrand events, because the underlying audio change is specific to named terms and the instruction structure of the video is otherwise unchanged. This is one of the scenarios where GlossCap’s glossary reprocess capability makes the largest operational difference: a brand rename that would require manual correction of hundreds of SRT files becomes a single glossary update and a batch reprocessing job. The caption glossary maintenance workflow covers how to structure the glossary update so that brand terms, product names, and renamed entities are consistently handled across the full content library.

Trigger type 2: Feature deprecation and UI change

When training content demonstrates a product feature or UI element that no longer exists, the audio references something the learner cannot locate. The caption file faithfully transcribes that reference. This creates a different kind of staleness than a rebrand event: the vocabulary may be entirely unchanged, but the instructional content is wrong. A caption file that accurately says “click the Settings gear in the top right corner” is inaccurate the moment the settings UI is redesigned and the gear icon moves or disappears, because the instruction is directing the learner to a location that no longer exists in the product interface.

Feature-deprecation staleness is most common in software training content — product training, systems training, IT onboarding, and any content that demonstrates a UI flow step by step. The content scope matters materially when prioritising remediation: a 45-minute onboarding video that demonstrates a deprecated feature flow throughout its runtime is more urgent to remediate than a 3-minute supplemental micro-lesson that mentions a deprecated feature in a single segment. But both require remediation, because a Deaf learner who relies on the captioned version of the onboarding video to complete the task the video is teaching has been given incorrect instructions in the most accessible format of that training.

Feature deprecation typically requires full re-captioning rather than a glossary reprocess, because the underlying audio is usually being re-recorded to reflect the UI change. The new audio describes a different workflow with different steps, different button names, and potentially different vocabulary. A glossary update does not address structural changes to the instruction content; only re-captioning from the new audio produces an accurate file. For content teams that use a consistent recording workflow, the feature-deprecation update is also an opportunity to apply any glossary improvements made since the original captioning job — the new caption file will be generated with the current glossary applied, so any term improvements accumulate automatically.

From a lifecycle management perspective, the best prevention for feature-deprecation staleness is integrating a caption review trigger into the software release process or the content update intake form. When the product team notifies the L&D team that a UI change is coming and training content will need to be updated, the intake process should include a caption review flag that gets added to the update project alongside the script rewrite and the re-recording schedule. The governance integration points in Section 8 of this post cover how to embed this trigger into the standard content update workflow.

Trigger type 3: Regulatory and policy change

Training content about regulatory requirements, compliance obligations, and internal policy carries a built-in staleness risk: regulations change, enforcement guidance is updated, and internal policies are revised on cycles that may or may not align with the content update schedule. HIPAA training content from 2022 may reference OCR enforcement guidance that has since been superseded by new rulemaking. FINRA compliance training may cite rules that have been amended after an industry comment period. An internal code of conduct training may reference a leave policy that HR updated after a labour relations review. When the underlying regulation or policy changes, the training content needs to be updated — and the caption file for that content needs to be updated in lockstep, not as a downstream task.

Caption staleness in regulatory and policy training creates a specific compliance risk that goes beyond the general accuracy obligation: if a Deaf or hard-of-hearing employee completes a captioned version of a compliance training module and the captions reference superseded regulatory guidance, the question of whether accessible-format delivery of the current compliance obligation occurred becomes contested. The employee received accessible content — a captioned video — but the captioned content described a version of the regulation that is no longer in force. Whether that constitutes effective accessible delivery of the current compliance training depends on the specific facts, but the risk is real and the documentation trail for how it happened — a caption file that was not updated when the policy was — is unflattering.

Organisations under ADA Title II, Section 508, or the European Accessibility Act need to maintain current, accurate captions for all compliance training content specifically because these are the training categories most likely to be examined in an OCR complaint investigation or audit. A caption file for a HIPAA training video that references a 2019 version of the enforcement guidance is not a technical inaccuracy in a peripheral training module — it is an accessibility failure in the highest-visibility, highest-risk content category in the organisation’s L&D portfolio. The broader enforcement landscape is covered in ADA Title II enforcement reality check 2026, which documents exactly what OCR investigators examine when they request caption files for compliance training content specifically.

Regulatory staleness is also the trigger type most likely to create caption staleness in purchased compliance content from third-party vendors. An external vendor’s OSHA training video that was captioned when it was produced in 2021 may not have been re-captioned when the vendor updated the script to reflect 2023 OSHA guidance. If your LMS is serving that video with a caption file that references the 2021 version of the standard, you have regulatory staleness in vendor-supplied content that you did not produce and may not have visibility into. The vendor content captioning question is addressed in the FAQ section of this post and in purchased compliance content captioning.

Trigger type 4: Safety content update

When a safety procedure is revised — after an incident investigation, after an OSHA citation, after equipment change, after process redesign — the training content describing the old procedure becomes not just inaccurate but potentially hazardous. For L&D teams producing safety training in manufacturing, construction, healthcare, and laboratory environments, safety content updates represent the highest-urgency caption update scenario. Unlike a rebrand event (where the caption file is wrong but the learner can still complete the task) or a UI change (where the learner is misdirected), a stale caption file for a revised safety procedure may direct a Deaf or hard-of-hearing worker to follow a step sequence that the safety update was explicitly designed to prevent. The accessible-format version of the training is actively giving wrong instructions.

The severity of this risk means that safety content caption updates must be treated differently from other trigger types: the caption file attached to a revised safety video must be updated before the video is published, not after. In every other trigger type, the caption update is remediation of an existing gap. In safety content, allowing the updated video to go live with the old caption file — even for a day — means that the caption-dependent learner receives instructions from the superseded procedure. The practical implication for L&D operations is that safety content updates need to include captioning as a blocking dependency in the project plan: the video cannot be published until the caption file has been updated and validated against the new audio.

The audit trail for safety content is also the most demanding of any trigger type. Regulatory agencies (OSHA, state OSHA equivalents), insurers, and litigation opponents may request the complete modification history for safety training content as evidence of when a revised procedure was communicated to workers and in what form. For Deaf and hard-of-hearing workers, the accessible form of that communication is the caption file. An organisation that cannot produce the caption file for the revised safety video — specifically the version that was in use on the date a particular worker completed the training — has a documentation gap in its safety communication record that can be exploited in post-incident litigation. The archive framework in Section 7 of this post is particularly important for safety content precisely because this audit trail demand is highest in that category.

The manufacturing safety equipment video captioning post covers the captioning production requirements for safety content specifically, including the accuracy standard for safety-critical terminology, the review workflow that ensures the caption file accurately reflects the procedure as described in the audio, and the distribution verification step that confirms Deaf and hard-of-hearing workers have received the updated captioned version before returning to the job task the training covers.

Staleness detection: identifying which caption files need attention

Knowing that caption staleness exists is different from knowing which specific files in a content library of hundreds or thousands of videos are affected. Before any remediation work can begin, L&D teams need a method for identifying the population of caption files that are likely to be out of sync with their current audio. Three systematic detection methods address this problem at different levels of automation, and a prospective workflow integration approach prevents the accumulation of new staleness going forward.

The 10% audio-change threshold

The practical question that staleness detection must answer is not “has this video changed?” but “has this video changed enough to require a caption update?” Minor edits to a video — a title card revised, a lower-third text corrected, a colour grading adjustment — do not affect the audio and therefore do not create caption staleness. The threshold for required caption remediation is substantive audio change: a change that alters what the learner hears and therefore what the caption must accurately represent.

The standard content-revision trigger used in training content management is the 10% substantive change threshold: if more than 10% of the audio content has changed in substance — not just cosmetically — a caption update is required. This threshold is derived from accessibility policy frameworks that distinguish between minor editorial revisions (fixing typos in a slide, reordering sentences that do not change meaning, adjusting pacing without changing words) and substantive revisions (changing instructions, updating terminology, adding or removing steps, re-recording segments for accuracy). Applying the 10% threshold to caption lifecycle decisions provides a defensible, consistent standard that can be documented in the caption governance policy and referenced when an OCR investigator asks how the organisation determines when caption updates are required.

The 10% threshold is a minimum bar, not a ceiling for caption update decisions. In safety content and regulatory training, a 2% audio change that revises a critical procedure step or a specific regulatory citation requires an immediate caption update regardless of the overall percentage. The trigger type determines whether the threshold applies as an absolute bar or as a conservative minimum, with the trigger type hierarchy described in the previous section determining urgency: safety content updates are always caption-blocking regardless of percentage; regulatory updates require assessment of whether the changed section is material to compliance; rebrand and feature deprecation updates can apply the 10% threshold as a practical minimum bar.

The change trigger inventory

Before remediating a set of caption files, L&D teams need to identify which files are affected by a given change trigger. Three inventory sources are useful for this identification work, and the most effective approach combines all three rather than relying on any single source.

The first inventory source is LMS version history. Most enterprise LMS platforms maintain a version history for SCORM, xAPI, and video content. If a course has been revised, the LMS version history shows when the revision was published. Comparing the revision date against the caption file’s last-modified date identifies candidate files where the video was updated after the caption file was created. In Cornerstone OnDemand, the learning object version history is accessible through the Content Management interface; in Docebo and TalentLMS, course material version records are available to administrators through the course management panel; in Canvas, media object revision history is available through the Media Management interface. The LMS version history comparison is the fastest inventory method for teams that have not been maintaining a separate caption version log, but it has a limitation: if the LMS version history records only the most recent update event (not a complete history of all update events), a course that has been updated multiple times may show only the most recent update date, missing earlier updates that also created caption staleness.

The second inventory source is content management system tags. Teams that use a CMS — SharePoint, Confluence, an internal wiki, or a content development tracking tool — to manage training content status can query for content tagged with the rebrand, policy-change, or safety-update tag and cross-reference the resulting list against the caption library. This approach is most effective for teams that already use a CMS to track content change events, because the tag-based query produces a focused candidate list rather than requiring a comprehensive timestamp comparison across the full library. The limitation is that CMS tags are only as reliable as the tagging practice: if content updates are not consistently logged in the CMS with the appropriate tag, the tag-based query misses the updates that were not tagged.

The third inventory source is vendor caption metadata. Caption vendors and SaaS tools — including GlossCap — maintain metadata about when each caption file was generated. This generated-at timestamp is the authoritative record of when the caption was last produced or reprocessed from the source audio. If the video file has a newer last-modified timestamp at the storage layer than the caption file’s generated-at timestamp, the caption file predates the most recent video version and is a staleness candidate. GlossCap’s project dashboard surfaces this comparison for all captioned content in a project, making it the starting point for any staleness detection sweep in a library where GlossCap was used for the original captioning. The enterprise LMS caption audit methodology covers how to structure the metadata comparison at scale for libraries of several hundred to several thousand videos.

Manual spot-check protocol

For large content libraries where automated timestamp comparison is not feasible — because the LMS does not expose file-level timestamps, the caption files were produced by multiple vendors without consistent metadata, or the content management infrastructure does not support tag-based queries — a manual spot-check protocol provides a defensible alternative that can identify the most likely staleness candidates without reviewing the entire library.

The protocol: select 10 to 15 percent of the content library at random, focused specifically on three high-risk populations: (a) content that references named products, product features, or specific UI elements; (b) content in subject matter areas that are known to have changed in the past 12 months — this includes any content category where a regulation, policy, or product line has been updated; (c) content in the highest-risk domains defined by the trigger type hierarchy: safety procedures, regulatory and compliance training, onboarding content, and product training. For each sampled item in the focused selection, play the current video while reading the caption file at the same timestamp. Discrepancies of more than 5% of lines — captions that do not match the current audio — indicate a file that requires full review. A sample where more than one-third of the sampled items show discrepancies above the 5% threshold indicates systemic staleness in that content category, warranting a full review of the category rather than just the sampled items.

The manual spot-check protocol is not a replacement for systematic metadata comparison in large libraries, but it is a practical starting point for teams that need to assess the scope of the staleness problem before committing to a full remediation programme. The results of the spot-check — documented as a brief written protocol with sample list, review methodology, and findings summary — are also a useful artefact for the OCR document request if a complaint investigation is already underway. An organisation that can show OCR that it conducted a systematic spot-check to identify staleness candidates, found a specific scope of issues, and initiated remediation is demonstrating the same kind of deliberate, documented compliance effort that OCR investigators distinguish from organisations that had no detection process at all.

Embedding staleness detection in the content update workflow

The most reliable staleness detection is not retrospective audit but prospective workflow integration: catching caption staleness at the point where the content change is initiated, rather than discovering it weeks or months later in a periodic review. This means embedding a caption review checkpoint into the content update workflow — the same workflow that triggers script review, recording scheduling, and accessibility review before new content is published.

The practical implementation is a mandatory checkbox or field on the content update intake form that every L&D professional completes when initiating a video update request. The field asks: “Does this update change audio content? (Yes / No). If yes, attach the current caption file and indicate whether the change is vocabulary-level (rebrand, terminology update), structural (added or removed segments, re-recorded sections), or full re-record.” This single intake step creates the trigger for the caption remediation workflow without requiring the accessibility coordinator to monitor every content project in parallel. The person initiating the update has the information needed to answer the question; the intake form routes that information to the caption programme at the right point in the project lifecycle.

Prospective workflow integration also ensures that the caption update is scheduled alongside the content update rather than after it — preventing the window of time during which the updated video is live in the LMS with the outdated caption file still attached. The caption programme governance policy template includes a content update intake form template with the caption review checkpoint built in, and the caption programme change management rollout post covers how to train content producers to use the intake form consistently as part of their standard update process.

The remediation decision tree — which path to choose

Once a caption file has been identified as stale, the next decision is which remediation path to take. Three paths are available: re-caption from scratch using the current audio, update the caption glossary and reprocess the existing audio through the updated glossary, or apply manual corrections to the existing SRT or VTT file. Each path has a different cost profile, a different audit trail character, and a different suitability profile based on the nature of the audio change. Choosing the wrong path is one of the most common operational errors in caption lifecycle management: applying manual corrections to a file that needed a full re-caption leaves the underlying accuracy problem partially addressed, while re-captioning a file that only needed a glossary update wastes time and budget that could have been directed to other remediation work.

Path 1: Re-caption from scratch

Re-captioning from scratch means submitting the current version of the audio to the captioning workflow — whether that is a human vendor, a SaaS tool, or an internal captioning team — and generating a new caption file that reflects the audio as it exists today. (For teams deciding whether to handle update re-captioning in-house or through a vendor, the in-house vs. vendor caption team decision post covers the break-even analysis and the compliance documentation implications of each model.) The existing caption file is set aside and archived; the new file becomes the production version. This path is appropriate when the audio has changed substantially (more than 30% of the content revised in substance), when the original audio quality was poor and re-captioning will produce a cleaner transcript than correcting the existing file, or when the content type requires the clearest possible audit trail — safety procedures and regulatory compliance training being the primary examples.

Re-captioning from scratch produces the cleanest audit trail of any remediation path. There is no ambiguity about whether the caption file reflects the current audio, because it was generated from the current audio using the current date’s production workflow. For GlossCap users, re-captioning with the updated glossary applied means that the new file immediately incorporates any terminology changes from the rebrand or policy update — the new product names, the revised regulatory citations, the updated safety terminology — without requiring a separate glossary update step. The new file is generated accurate from the first pass rather than requiring post-production corrections for terminology that the old glossary did not handle correctly.

The cost driver for re-captioning from scratch is proportional to the video duration: re-captioning costs approximately the same as the initial captioning job, because the captioning workflow processes the full audio regardless of how much of it has changed. For a 45-minute product training video, a re-caption represents approximately 3 to 4 hours of captioner time in a human vendor workflow or a single GlossCap processing job at the per-minute rate. For teams managing a large library of stale content, re-captioning every affected file from scratch may not be economically feasible, which is why the decision matrix below distinguishes the cases where re-captioning is specifically required from the cases where a less expensive path is adequate.

Re-captioning is also the path of choice when the original captioning quality was poor and the accumulated corrections needed in the existing file would effectively require rewriting the entire transcript anyway. A file with a 15% word error rate from the original job is not a good candidate for manual correction; the correction effort approaches the cost of re-captioning while producing a modified file with a complicated edit history rather than a clean new file. In this situation, re-captioning from scratch is both more economical and more defensible from an audit perspective.

Path 2: Update glossary and reprocess

Glossary reprocessing means updating the term mapping in the caption glossary — adding a new entry, correcting an existing one, or replacing an old term with a new one — and submitting the original audio through the updated glossary to generate a corrected caption file. The underlying transcript structure (sentence boundaries, speaker identification, timing offsets) is preserved; only the vocabulary corrections from the glossary update are applied. This path is the fastest remediation option for rebrand events and terminology changes, and it is appropriate when the audio has changed primarily in vocabulary (a product name changed, a feature was renamed, a regulatory term was updated) but the underlying instructional content and audio structure are unchanged.

Glossary reprocess is particularly powerful for rebrand events because the change pattern is exactly what the glossary is designed to handle: a finite set of old terms that need to be replaced with new terms throughout the caption file, applied consistently. A brand rename that would require manually locating and correcting every instance of the old product name across dozens of SRT files becomes a single glossary update (add the new mapping: “Acme Platform” → “GlobalCorp Workspace”) followed by a batch reprocessing job that applies the correction to all affected files simultaneously. The caption glossary maintenance workflow covers the glossary update procedures for exactly this type of systematic term change.

This path requires a critical prerequisite: that the audio change is genuinely vocabulary-level. If the audio references the old product name but also contains deprecated UI element references, updated regulatory citations, or revised safety procedure steps, a glossary reprocess will fix the terminology errors but will not address the structural accuracy issues. A spot-check after reprocessing — playing the updated audio while reading the reprocessed caption file — should verify that the corrected file accurately reflects all of the current audio content, not just the specific terminology that the glossary update addressed. If the spot-check reveals structural discrepancies beyond the vocabulary corrections, the file needs to move to Path 1 (re-caption from scratch) or Path 3 (manual corrections) for the remaining issues.

The audit trail for a glossary reprocess is distinct from both re-captioning and manual correction: the reprocess event is logged with a timestamp and the specific glossary changes applied, and the resulting file is generated from the original audio with the updated glossary — making it traceable to both the original audio and the updated term set. For OCR document requests that ask how specific terminology changes were applied to caption files after a policy or product update, the glossary reprocess log provides a clear, systematic answer rather than requiring an explanation of dozens of individual manual corrections. GlossCap’s glossary reprocess workflow generates a reprocess event record that includes the old term, the new term, the number of instances corrected, and the timestamp of the reprocess — exactly the format useful for an OCR document request or VRA monitoring report.

Path 3: Manual corrections to the existing file

Manual corrections to the existing SRT or VTT file mean editing the caption file directly — locating the specific lines that need to be changed, making the corrections, and saving the updated file for upload to the LMS. This is the fastest and least expensive path for small, localised changes: a date or version number updated in one segment, a contact name changed in a single line, a sentence revised without changing the surrounding content or timing. Manual corrections are appropriate when the audio change is minor (under 10% of content, affecting one or two specific segments), when the change does not involve vocabulary complexity, and when the existing file is otherwise accurate enough that the correction scope is genuinely limited.

Manual corrections have one critical operational requirement that teams frequently fail to satisfy: every correction must be logged in the modification record before the corrected file is uploaded to the LMS. An untracked manual edit to a caption file — opening the SRT in a text editor, making the change, saving the file, uploading it — leaves no audit trail. The file timestamp on the saved version may change, but there is no record of what changed, who changed it, why, or when. If that file is later reviewed in an OCR document request or produced in e-discovery, the absence of a modification record creates a compliance documentation gap: the organisation cannot explain what changed between the original file and the file currently in use. This is one of the eight failure modes described in Section 9, and it is the most common version-control failure in caption libraries that rely on manual correction as a primary update method.

The modification log entry for a manual correction should include: the course ID and video ID the file is attached to; the date of the correction; the name or role of the person who made the correction; a brief description of what was changed and why (for example, “Updated policy effective date from January 1, 2024 to January 1, 2025 per HR policy revision dated 2025-02-15”); and the version number assigned to the corrected file per the naming convention described in Section 7. This log entry takes less than two minutes to complete and is the difference between a defensible audit trail and a gap that cannot be explained under scrutiny. The caption QA methodology for training video teams covers the correction logging format in detail as part of the broader quality assurance documentation framework.

The decision matrix

The three factors that determine the right remediation path — scope of audio change, vocabulary complexity of the change, and audit exposure of the content — combine to produce a decision matrix that can be applied consistently across your content team. “Audit exposure” in this context means the likelihood and consequence of the content being examined in an OCR document request, VRA monitoring review, or e-discovery hold: safety content and regulatory compliance training have the highest audit exposure; evergreen professional development content has the lowest. The matrix below covers the five most common change scenarios:

The matrix is a decision aid, not a rigid algorithm. Content characteristics that do not fit cleanly into the matrix cells — for example, a video with a 15% audio change that is almost entirely vocabulary-level in a high-audit-exposure domain — require judgement informed by the trigger type context: in high-audit-exposure domains, when in doubt, choose the path that produces the cleaner audit trail. A re-caption is never a wrong choice from an audit perspective; a glossary reprocess of a file that also had non-vocabulary accuracy gaps is a gap waiting to be found. For large-scale remediation programmes covering hundreds of stale files across multiple trigger types, the large-scale caption backlog remediation playbook covers how to apply the decision matrix at scale and batch similar change types together to maximise processing efficiency.

LMS propagation: updating caption files without losing learner records

The most operationally consequential step in the caption update workflow is getting the corrected file into the LMS without triggering an unintended version event that resets learner completion records. This is the step where teams make the most costly mistakes, and the correct approach is platform-specific. What works in Cornerstone OnDemand is different from what is safe in Docebo, and what is safe in TalentLMS is different from the Canvas Media Gallery workflow. The platform-specific guidance below covers the five enterprise LMS platforms most commonly used in corporate and higher-education L&D environments.

Before walking through the platform specifics, one principle applies universally: before replacing any content file in an LMS, confirm with the platform administrator or the LMS support team whether the specific replacement method you intend to use triggers a version event. Platform release notes change this behaviour over time, and the guidance below reflects current platform behaviour as of mid-2026 but may be superseded by platform updates. Test the replacement in a staging environment with a test learner completion record before updating the production course, particularly for safety training and compliance training where learner completions have legal significance.

Cornerstone OnDemand

In Cornerstone OnDemand, caption files for MP4 content uploaded via the Learning Object module can be replaced by updating the learning object’s associated files without creating a new course version. Navigate to the Learning Object via Edge → Learning → Learning Objects, select the specific learning object, and enter the content file management section. From there, the associated caption file (SRT or VTT) can be replaced by deleting the existing subtitle file and uploading the corrected version while keeping the video file unchanged. Learner completion records in Cornerstone are attached to the curriculum or course assignment at the transcript level, not to the specific content file version, so a caption-file-only replacement at the learning object level does not affect completions. The completion record references the learning object ID, which does not change when the caption file is updated within the learning object.

If the video file itself is being replaced (not just the caption file), the correct approach is to create a new content version in Cornerstone using the Versioning function rather than overwriting the existing content file. Cornerstone’s content versioning allows new completions to reference the current version while learners who completed prior versions retain their completion status. The version management interface is accessible from the learning object record and generates a new version ID that the LMS uses to track which version each learner transcript references. Overwriting the video file directly without creating a new version can produce timestamp mismatches in the transcript record that may create issues with completion data integrity in subsequent reporting. For the detailed LMS-specific caption ingestion workflows including Cornerstone-specific file formats and subtitle track configuration, see LMS caption ingestion workflow engineering and the Cornerstone OnDemand captions reference.

Docebo

In Docebo, caption files for learning materials can be replaced via Course Management → the specific course → Materials. Select the video material, navigate to the accessibility or subtitle section, and upload the replacement caption file. Docebo associates caption files with the material at the file level independently of completion records, so a caption file replacement within an existing material does not affect course completion tracking. The completion record is attached to the course enrolment and the material’s activity record, not to the subtitle file version. This makes Docebo one of the more straightforward platforms for caption-only updates: the workflow is direct and the completion record impact is reliably zero for file-level subtitle replacement.

For SCORM content hosted in Docebo where the caption file is embedded in the SCORM package — a common configuration for content produced in Articulate Storyline or Rise, where captions are embedded as a subtitle track in the SCORM output — a full SCORM package update is required to update the caption file. Docebo’s SCORM package update workflow preserves existing completion records for the prior SCORM version while directing new completions to the updated package. The versioning behaviour is controlled by the SCORM package revision field in the content properties. Confirm with your Docebo administrator that the package update method you are using is the versioned update path rather than the replace-in-place path before proceeding. The Docebo captions reference covers the subtitle configuration interface and supported formats in detail.

TalentLMS

In TalentLMS, caption files for video content that is not SCORM-packaged are uploaded as a separate subtitle track on the video resource. Caption replacement requires navigating to the specific unit in the course editor, selecting the video content item, accessing the subtitle settings, deleting the existing subtitle track, and uploading the corrected file as a new track. This process does not affect course completion records in TalentLMS: the completion record is tied to the unit and course, not to the subtitle file associated with the video resource. Learner progress and completion status are unaffected by caption file replacement at the subtitle track level.

For SCORM content on TalentLMS, the same pattern applies as Docebo: updating the SCORM package is required when the caption file is embedded in the package, and TalentLMS preserves completion records for the prior SCORM version while directing new completions to the updated package. One TalentLMS-specific caution: if the course uses the “Reset progress when re-taking course” setting, a SCORM package update may interact with that setting in ways that affect existing learner progress. Review the course reset settings with your TalentLMS administrator before updating any SCORM package in a course where learner completions are legally significant. The TalentLMS captions reference documents the subtitle track configuration and SCORM caption embedding approaches supported by the platform.

Canvas LMS

In Canvas, caption files uploaded as subtitle tracks on media items — whether uploaded via the Rich Content Editor, the Course Files media library, or the dedicated Media Gallery — can be replaced or updated without affecting course completion status. The most direct path for caption file replacement is through the Media Management or Studio interface: navigate to Media Gallery → select the specific media item → Edit → Captions. From the Captions tab, the existing caption track can be deleted and the corrected file uploaded as a replacement. This action is available to instructors with course design permissions and to administrators with account-level media management access. Course completion status in Canvas is tracked through the course completion rules (which are based on module requirements and assignment submissions), not through the specific caption file version associated with a media item.

For Canvas courses that use external video hosting — Kaltura, Panopto, or Vimeo — with captions managed at the hosting platform level rather than through Canvas’s native caption track, the caption update must be made in the hosting platform rather than through Canvas. A corrected SRT file uploaded to Kaltura replaces the caption track on the Kaltura media item and propagates through all Canvas embeds of that Kaltura item automatically; no Canvas-side action is needed. For Panopto-hosted content in Canvas, the caption update follows the Panopto caption edit workflow. The Canvas LMS captions reference covers the native Canvas caption track management and the integration patterns for Kaltura and Panopto. The LMS migration caption checklist covers the caption track handling requirements when moving content between hosting platforms or between LMS environments.

Workday Learning

Workday Learning manages caption files as part of the learning content record associated with each lesson in the Content Library. Updating caption files requires access to the Content Library: navigate to the specific lesson, open the Edit Content view, and access the content attributes section where the caption file is associated with the lesson. From this view, the existing caption file can be replaced with the corrected version. Workday’s completion records are tied to the learning assignment and learner transcript record at the lesson level, not to the specific content file version, so caption file replacement within an existing lesson record does not affect completion history. Learner transcripts that show completion of the lesson before and after the caption update both remain valid.

For Workday content where the video file and caption file are bundled as a single content asset — a configuration common in older Workday Learning implementations where content was imported as a packaged media file rather than configured with separate video and subtitle tracks — updating the caption file may require creating a new content version rather than replacing the file within the existing lesson. Workday’s content versioning workflow in this scenario is release-specific: the version management interface changed significantly between Workday 2023R2 and 2024R1. Consult the Workday Learning documentation for your specific release before attempting a bundled content asset update, and test the update in a Workday sandbox environment before applying it to the production tenant if learner completions on the affected lesson are in any way legally or operationally significant.

Common propagation risk: the version reset

The single most common and costly LMS propagation error is triggering a version reset by replacing the course package or the underlying course structure rather than the caption file specifically. When a version reset occurs, the LMS may treat the course as a new course registration requirement for all learners, invalidating existing completion records and requiring learners who already completed the course to complete it again — or, in some platform configurations, showing previously completed learners as incomplete in reporting dashboards, which creates audit trail problems for compliance training where completion records are used to demonstrate that all employees received required training.

The version reset risk is highest in SCORM-based content where the caption file is embedded in the SCORM package, because the most obvious update method — re-publishing the course from the authoring tool and uploading the new SCORM package — typically triggers a version event. Avoiding this risk requires either extracting the caption file from the SCORM package and uploading it as a standalone subtitle track (which is platform-specific and not always possible in all SCORM packaging configurations), or using the LMS’s native SCORM versioning workflow that explicitly preserves prior completion records. Both approaches require platform-specific knowledge and should be validated in a staging environment before applying to production courses. The LMS caption ingestion workflow engineering post covers the SCORM-specific caption embedding configurations for each major authoring platform and how to update them without triggering a version reset in the most common LMS environments.

Audit archive: preserving superseded caption files

When a caption file is updated — for any of the reasons described in this post — the LMS replaces the active caption file with the updated version. The prior version is no longer delivered to learners, and in many LMS configurations it is effectively invisible: the platform holds only one active caption track per video, and the replacement overwrites the display reference to the prior file. If no independent archive was maintained, the superseded file is gone. This is a compliance risk that most L&D teams do not recognise until they are in an OCR investigation and cannot produce the document OCR has asked for.

Why deleting the superseded caption file is a compliance risk

There are three distinct compliance scenarios in which an organisation needs to produce a superseded caption file — a file that is no longer active in the LMS but was in use during a specific prior period. Each scenario creates a different consequence if the file cannot be produced.

The first scenario is an OCR document request under ADA Title II. In an OCR complaint investigation, OCR may request caption files for specific training content that was in use during the period covered by the complaint. If the complaint covers content consumed between April 2025 and April 2026, OCR may request the caption file that was in use during that period — not the current updated file, but the specific version that was served to learners who accessed the content during the complaint window. If the superseded file was deleted when the updated file was uploaded, the organisation cannot produce it. This is a document production failure in a civil rights investigation, which is a materially worse outcome than producing the superseded file and explaining why it was subsequently updated. OCR investigators understand that caption files get updated; they do not accept “we deleted it” as a satisfactory response to a document request for a specific version.

The second scenario is VRA monitoring under a voluntary resolution agreement. VRAs commonly require the respondent organisation to demonstrate that caption accuracy has been maintained and improved over the monitoring period. Organisations under active VRA monitoring should retain prior caption versions so that accuracy review can be conducted against the historical file, not just the current one. A VRA monitoring report that shows the current caption file meets the 99% accuracy standard is useful, but a VRA monitoring report that shows the accuracy of the file that was in use at the start of the monitoring period and the improvement achieved over the monitoring period through successive updates is far more compelling evidence of the programme improvement that VRAs are designed to produce.

The third scenario is e-discovery in civil litigation. In ADA Title I and Section 504 disability discrimination lawsuits where the accessible-format delivery of training content is at issue — a Deaf employee alleging that the company’s training programme was inaccessible during a specific period of their employment, for example — the caption file served to the plaintiff during the period covered by the complaint is potentially discoverable. If that file has been replaced and deleted, the defendant organisation is in a substantially weaker position than one that has preserved the complete version history. A complete version history can demonstrate: what caption file was in use during the plaintiff’s employment period, whether that file was accurate, when it was updated and why, and the overall trajectory of the captioning programme’s accuracy over time. All of that context is lost when superseded files are deleted at replacement time. The caption vendor audit rights and examination evidence post covers the full scope of documentation that civil litigation discovery can surface and how to structure the archive to satisfy the broadest reasonable discovery scope.

Archive naming convention

The archive does not need to be a sophisticated document management system. A simple folder structure in a cloud storage location — SharePoint, Google Drive, Amazon S3, or an equivalent enterprise file storage platform — with a consistent naming convention is sufficient to satisfy OCR document requests, VRA monitoring, and civil litigation discovery in all but the most complex multi-entity situations. The naming convention needs to uniquely identify the file, the content item it belongs to, the file format, the version number, and the date the version became active (which is also approximately the date the prior version was superseded).

Recommended naming convention: [course-id]_[video-id]_[format]_v[version]_[YYYY-MM-DD].[ext]

Applying this convention to a compliance training video that has been updated twice since original captioning:

• COMP-001_vid-003_srt_v1_2024-03-15.srt — original file, generated March 15, 2024
• COMP-001_vid-003_srt_v2_2025-11-02.srt — post-rebrand update, generated November 2, 2025
• COMP-001_vid-003_srt_v3_2026-07-11.srt — post-policy-update revision, generated July 11, 2026

The version number sequence (v1, v2, v3) allows the complete history to be reconstructed in chronological order without needing to examine file timestamps. The date in the filename is the date the version was first made active in the LMS, not the date it was generated in the captioning workflow (these are usually close but not always identical). The course ID and video ID map to the LMS content identifiers, making it straightforward to locate all caption versions for a given course or video when a document request specifies a specific piece of content. The format field (srt, vtt) distinguishes files if both formats are maintained for a given video.

If your organisation uses both SRT and VTT formats for the same video — for example, SRT for LMS delivery and VTT for a public-facing course page — archive both format versions at each version point. OCR document requests typically specify the file format delivered to learners, which should be the LMS format, but having both formats archived avoids the need to explain the format discrepancy if a document request is worded differently than expected.

Archive storage structure

The archive should be separate from the production caption library. Not in the LMS itself — which holds only the currently active file and has no version management for caption tracks in most platforms — but in a dedicated caption archive folder that mirrors the LMS course structure at the folder level. A top-level folder called caption-archive or caption-version-history with subfolders named by course ID, each containing the versioned caption files for all videos in that course, provides a structure that scales from a library of 50 videos to one of 5,000 without requiring reorganisation.

Access control for the archive folder is important: write access should be restricted to the caption programme manager and the accessibility coordinator. All other users should have read-only access at most. This restriction prevents accidental overwriting of archived files and ensures that the modification history of the archive itself is attributable to a small set of responsible parties. If an archived file is ever produced in a legal proceeding, the chain of custody is cleaner when access was restricted to designated individuals rather than open to the full L&D team.

The archive folder path and access control settings should be documented in the caption governance policy so that the archive can be located quickly during a document request without requiring institutional knowledge that resides with a specific individual. If the accessibility coordinator who set up the archive is no longer with the organisation when OCR makes its document request, the governance policy documentation ensures that the incoming coordinator or their manager can find and access the archive immediately. The caption programme governance policy template includes an archive documentation section that can be populated with the specific paths and access control settings for your organisation’s storage environment.

Archive retention period

The minimum retention period for superseded caption files is five years from the date the file was superseded — meaning from the date the updated file replaced it as the active version in the LMS. This five-year retention period is designed to cover the three compliance scenarios described above: it exceeds the OCR complaint filing window (180 days from the date of the access denial that triggered the complaint, but complaints can be filed years after a course was completed if the access denial was part of an ongoing employment relationship), it comfortably spans the VRA monitoring period (two to three years in most captioning VRAs), and it provides a reasonable buffer for civil litigation discovery timelines in employment discrimination cases (which can span three to five years from the alleged discriminatory act to discovery completion in complex cases).

Organisations in regulated industries — healthcare, financial services, government contracting, pharmaceutical manufacturing — should consult their records management policy and legal counsel for retention requirements that may exceed the five-year minimum. HIPAA business associate agreements may specify retention periods for training record documentation. FINRA requires certain records to be maintained for six years. Federal contractors subject to OFCCP regulations may have specific retention requirements for training records. In any of these contexts, the caption archive retention period should be aligned with the longest applicable records management requirement rather than defaulting to the five-year minimum.

At the end of the retention period, superseded files can be deleted pursuant to the organisation’s records retention schedule. The deletion should be documented in the records management log with the file identifiers, the reason for deletion (retention period expired), and the date of deletion. A clean records management deletion log is useful evidence that the organisation manages its records systematically rather than deleting documents selectively.

Embedding lifecycle management in governance

Caption lifecycle management is not a one-time remediation project — it is a recurring operational responsibility that accumulates over the lifetime of the content library. Without structural integration into the governance framework that surrounds the L&D content development process, lifecycle management defaults to periodic cleanup projects that catch staleness months after it has accumulated and produce remediation queues that overwhelm the available captioning capacity. Three governance integration points make lifecycle management sustainable as a concurrent operational practice rather than a downstream cleanup function.

Integration point 1: Content update intake form

When a content producer initiates a video update request, the intake form should include a mandatory caption review field that routes caption staleness assessment to the captioning workflow at the same point where the content update project begins. The field should ask: “Does this update change audio content? (Yes / No). If yes, attach the current caption file and indicate whether the change is vocabulary-level, structural (added or removed audio segments), or full re-record.” A “Yes” response to this field automatically creates a caption update task in the project plan alongside the content update tasks, ensuring that the caption update is resourced, scheduled, and tracked as part of the same project — not as a separate downstream action that gets deferred when the content update is complete and the project team has moved on.

The intake form integration is the most effective staleness prevention tool available to L&D operations because it catches the staleness trigger at the point of creation rather than in a retrospective audit. It also creates a documented record of every content change event and the associated caption assessment, which is a useful component of the audit trail if a document request ever asks how the organisation monitors its caption file accuracy over time. A content update intake log that shows every update request was evaluated for caption impact — even when the answer was “no audio change, no caption update needed” — demonstrates a systematic approach to lifecycle management that is qualitatively different from an organisation that has never asked the question. The caption compliance programme build guide includes the intake process design as part of the broader programme governance structure, and the caption programme change management rollout covers how to train content producers to complete the intake field accurately and consistently.

Integration point 2: Annual caption library audit

Even with a well-functioning intake process, some content updates will escape the intake process: a subject matter expert who updates a training module directly in the LMS without filing an intake request, a departmental coordinator who re-records a segment without notifying the L&D team, a vendor-supplied content update that does not trigger the internal intake workflow because the content was not internally produced. The annual caption library audit is the safety net that catches these missed updates before they accumulate into a backlog that represents significant compliance exposure.

The annual review process — covered in full in the caption programme annual review process — should include a staleness scan as a standard component: for every captioned item in the LMS, compare the video file’s last-modified date against the caption file’s generated-at or last-updated date. Items where the gap is more than 90 days warrant manual review: pull the current caption file, play a 10-minute sample of the current video at the most change-likely section (product demonstrations, policy citations, procedure steps), and confirm that the caption text matches the current audio. Items where a discrepancy is found during the manual review go into the remediation queue with the appropriate path designation from the decision matrix.

The 90-day gap threshold for flagging items is a practical balance between sensitivity (catching genuine staleness) and specificity (avoiding false positives from files that were recently updated and show a small timestamp gap due to production workflow timing). For safety content and regulatory training, consider using a shorter threshold — 30 or 45 days — in the annual scan to ensure that the highest-audit-exposure content is reviewed more frequently. The annual scan results should be documented in the programme annual review report, including the total items scanned, the number flagged for manual review, the number found to be stale, and the remediation actions initiated. This documentation is directly responsive to the OCR document request item asking for records of the organisation’s process for maintaining caption accuracy over time.

Integration point 3: Governance policy clause

The caption governance policy should include a specific clause on caption file lifecycle management that defines the four trigger types that require caption review, the decision criteria for remediation path selection (by reference to the decision matrix in this post, adapted to the organisation’s specific content types), the LMS propagation procedure for the specific platforms in use, and the archive retention requirement. The governance policy clause is the document that makes lifecycle management a defined institutional obligation rather than an informal practice that depends on individual knowledge and initiative. When the accessibility coordinator who set up the lifecycle management workflow leaves the organisation, the governance policy clause ensures that their successor knows the standard and can continue applying it without needing to reconstruct it from memory.

The caption programme governance policy template includes a Section 7 on caption file lifecycle management that covers all four trigger types, the 10% threshold standard, the remediation path decision framework, the LMS propagation requirement, and the archive retention period. The template section can be adopted as written for organisations that are establishing their lifecycle management standard for the first time, or adapted for organisations that have existing lifecycle management practices that need to be formalised. The caption compliance self-assessment checklist includes lifecycle management as one of its eight assessment domains, with specific questions about whether each governance integration point is in place and functioning. The caption programme maturity model maps lifecycle management governance maturity from Reactive (Level 1: no systematic staleness detection, no archive) through Strategic (Level 5: automated staleness detection integrated with content management system, complete version-controlled archive with retention management).

Eight failure modes in caption file lifecycle management

Most caption lifecycle management failures are predictable. They occur in response to the same structural pressures — time constraints, workflow fragmentation, unclear ownership, audit archive complexity — and they produce the same audit trail gaps. Understanding the failure modes before they occur is the most efficient way to design them out of the workflow rather than discovering them in an OCR document request.

1. Replacing the video without triggering a caption review

This is the most common failure mode in content libraries with active update cycles. A content producer updates the video in the LMS — re-recording a segment, splicing in updated screen recordings, refreshing the slide deck narration — without notifying the caption programme. The outdated caption file remains attached to the new video. Learners watching the updated video read captions describing the old version. The failure is invisible until a caption-dependent learner reports the discrepancy or a periodic audit surfaces it.

Prevention: the content update intake form with a mandatory caption review checkbox is the primary control for this failure mode. The intake form makes it structurally impossible to initiate a video update without triggering a caption assessment. Secondary controls include making the caption programme manager a required reviewer on all LMS content update change requests and including a caption currency check in the course publication approval workflow.

2. Overwriting the superseded caption file at replacement time

When the updated caption file is uploaded to the archive or to the LMS, the uploader replaces the original file at the storage layer rather than creating a new versioned file. The superseded file is permanently deleted. The version history is gone. If OCR subsequently asks for the caption file that was in use during a specific period, the organisation cannot produce it.

Prevention: the archive naming convention that includes version number and date makes this failure mode structurally unlikely — a file named COMP-001_vid-003_srt_v2_2025-11-02.srt is not accidentally overwritten when uploading COMP-001_vid-003_srt_v3_2026-07-11.srt because the filenames are different. Write access restricted to the caption programme manager reduces the risk further by ensuring that archive uploads are performed by someone who knows the naming convention and the intent of the archive. The archive step should be completed before the updated file is uploaded to the LMS — archive first, then update production.

3. Treating a glossary reprocess as equivalent to a re-caption for audit purposes

A glossary reprocess corrects vocabulary errors in the caption file without verifying the accuracy of the full transcript against the current audio. If the reprocessed file is logged in the modification record as a “full accuracy review and correction,” it misrepresents the scope of the remediation. If OCR later reviews the modification log and asks about the scope of the review, the description “full accuracy review” for a file that received only a glossary reprocess is inaccurate, which undermines the credibility of the modification record as a whole.

Prevention: distinguish clearly in the modification log between three types of caption update events: glossary reprocess (vocabulary corrections applied via glossary update, no full accuracy verification); manual corrections (specific identified errors corrected with logged change descriptions); and full re-caption (new caption file generated from current audio, replaces all prior transcript content). Each event type has a different scope and a different audit trail character, and the modification log should describe which type occurred for each update event.

4. Propagating a caption file update via SCORM package replacement

When a caption file is embedded in a SCORM package and the content team updates the SCORM package to update the caption, the full SCORM package replacement triggers a version event in most LMS platforms. This version event may invalidate learner completion records for all learners who completed the prior SCORM version, depending on the platform configuration and the SCORM package manifest version settings. For compliance training where completion records demonstrate that employees received required training, a mass-completion-record invalidation is an operational emergency — and it is created by a caption file update that should have been a non-event.

Prevention: always attempt to update caption files at the file level (replacing the subtitle track independently of the SCORM package) rather than at the package level. When file-level update is not possible because the caption is embedded in the SCORM output, use the LMS’s native SCORM versioning workflow rather than replacing the package directly. Test the update in a staging environment with a test learner completion record before updating production, and confirm with the platform administrator that the method being used preserves existing completion records.

5. Applying manual corrections without logging the change

Manual edits to SRT or VTT files in a text editor leave no audit trail by default. The saved file has a new modification timestamp, but there is no record of what changed, who changed it, why, or what the previous state was. If the modified file is later reviewed in an OCR document request or produced in civil litigation discovery, the absence of a modification log means the organisation cannot explain what the file contained before the edit or why the edit was made. In the context of a disability discrimination claim where the accuracy of a specific caption file during a specific period is at issue, “we edited it but don’t have a record of what changed” is a damaging response.

Prevention: log all manual corrections in the modification log before uploading the corrected file to the LMS. The log entry takes less than two minutes and creates a record that can be produced on demand. The archived pre-correction version of the file (preserved under the prior version filename) provides the “before” state that the modification log describes changing. Together, the archived file and the modification log entry constitute a complete audit record for the correction event.

6. Assuming the LMS “last updated” timestamp is authoritative for staleness detection

Some LMS platforms display a “content last updated” or “course last modified” timestamp in the course management interface. This timestamp is frequently used by L&D teams as a proxy for content freshness during staleness detection sweeps. The problem: the LMS course record timestamp typically reflects the last modification to the course record — including changes to course description, enrollment settings, notifications, or any other course metadata field — not specifically the last modification to the video file or the caption file. A course that had its enrollment settings changed last month may show a “last updated” date of last month even if the video and caption file were last touched two years ago.

Prevention: compare the caption file’s generated-at or last-updated timestamp — available from the captioning vendor metadata or the GlossCap project dashboard — against the video file’s last-modified timestamp at the file system or CDN layer. These timestamps reflect actual file changes, not course record metadata changes. The LMS course record timestamp is a useful starting point for identifying courses that have been recently touched, but it should be validated against the actual content file timestamps before concluding that a course’s content is current.

7. Delaying archive during a rapid-turnaround caption update

When a safety content update or regulatory change requires an urgent caption update — a revised procedure needs to be live in the LMS today, a regulatory citation needs to be corrected before the next training cohort completes the module — the archive step is frequently skipped because the team is focused on getting the correct file live as quickly as possible. The superseded file is overwritten or deleted. The urgency of the update is real, but the archive step is not what creates the delay: saving a copy of the existing file to the archive folder with the correct filename takes under 60 seconds. The decision to skip the archive step in an urgent update sacrifices permanent compliance documentation for less than a minute of saved time.

Prevention: build the archive step into the urgent-update checklist as Step 1, before the updated file is uploaded to the LMS. The checklist sequence for an urgent update should be: (1) save a copy of the current (about to be superseded) caption file to the archive folder with the versioned filename; (2) upload the corrected caption file to the LMS; (3) verify in the LMS that the correct file is now live; (4) log the modification event in the modification record. The archive step is first because it is the step most likely to be forgotten if it is last.

8. Treating caption file lifecycle as a post-project cleanup task

This is the structural failure mode that produces all the others. When caption lifecycle management is positioned as a downstream cleanup activity — something that happens after content updates are complete and the project team has moved on — it accumulates as a backlog of deferred work that grows faster than it is cleared. Stale caption files accumulate in the LMS for months between cleanup cycles. For safety and regulatory training, the accumulation period is a period of active compliance exposure: Deaf and hard-of-hearing learners are accessing content through captions that do not match the current audio, and the organisation has no systematic awareness of the gap.

Prevention: embed caption review in the content update workflow as a concurrent task, initiated at the same time as the content update and resourced in the same project plan. The intake form integration (Integration Point 1 above) is the structural mechanism for achieving this. When caption review is part of the same project as the content update, it is tracked, resourced, and completed within the same project timeline rather than deferred until a quarterly or annual cleanup cycle. The caption compliance self-assessment checklist includes a question specifically about whether caption lifecycle management is embedded in the content update workflow or managed as a periodic cleanup, and the caption programme maturity model identifies the shift from cleanup-cycle management to concurrent workflow integration as one of the defining transitions between maturity Level 2 (Developing) and Level 3 (Established).

FAQ

If we re-record only one segment of a 30-minute training video, do we need to update the full caption file or just the segment?

If the re-recorded segment is spliced into the existing video at the production layer — the edited video is exported as a single new file that replaces the original — the resulting video file is new, and the existing caption file’s timestamps will be wrong for every segment that comes after the splice point because the total duration has changed. Inserting a re-recorded segment that is longer or shorter than the original segment shifts every subsequent timestamp forward or backward, making the caption file progressively less synchronised for everything that follows the splice. In this scenario, you need to update the full caption file for the full video, not just the re-recorded segment. The practical approach is to use Path 1 (re-caption from scratch) for the full updated video, which produces a correctly timed caption file from the final video output rather than attempting to adjust the existing file’s timestamps around the splice point — a manual process that is error-prone and produces an audit trail that is harder to explain than a clean re-caption.

If the video is being published as a new separate module — the revised segment as a standalone resource with its own LMS listing, rather than as a replacement for the full original video — caption that segment as a new file with a new version history and archive it independently. The original video and its caption file remain archived under their existing version history. The new standalone module starts its own version history at v1.

Our vendor’s SLA guarantees accuracy for the original captioning job but not for subsequent updates. Are we responsible for the accuracy of the updated caption file?

Yes, you are responsible. The caption file that is live in your LMS is your accessible-format delivery artifact for learners who depend on captions. If that file is inaccurate because the underlying audio changed after the initial captioning job and the file was not updated, you bear the compliance risk for that inaccuracy regardless of whether your vendor’s SLA covers post-delivery updates. The vendor SLA limits the vendor’s liability; it does not shift your organisation’s compliance obligation to the vendor.

Review your vendor contract to determine whether caption updates — for re-captioning after content changes, not for correcting errors in the original job — are covered at the original per-minute rate, at a different rate, or require a new statement of work. Most captioning vendors treat content updates as separate orders from the original captioning job, billed at the standard per-minute rate for the updated segments. The caption vendor SLA review checklist covers the specific contract provisions to review and negotiate when the captioning relationship needs to cover content update scenarios, including rate structures for update work and the vendor’s obligation to apply the current glossary state to update jobs.

How do we handle caption files for video content we purchased from a third-party compliance training vendor?

If the vendor provides caption files as part of the content package, you are dependent on the vendor to update those files when the content is updated. The critical point is that your compliance obligation is not satisfied by the existence of a vendor-provided caption file — it is satisfied by the existence of an accurate vendor-provided caption file that reflects the current audio content of the video being served to your learners. If the vendor updates the training video to reflect a regulation change but does not provide an updated caption file, you are serving inaccurate captions through a vendor-supplied file, and the compliance risk lands with you.

Your LMS deployment documentation and your contract with the vendor should both address this. The contract should specify that caption accuracy responsibility for vendor-supplied content rests with the vendor and requires the vendor to deliver updated caption files whenever content is revised — not as an optional service, but as part of the standard content update delivery. If the vendor’s current contract does not include this provision, a contract amendment requiring it is the appropriate next step. If the vendor refuses the amendment, your options are to retain the right to generate your own replacement caption files for vendor content deployed in your LMS (ensuring you have the licence terms to do so), or to factor the vendor’s caption update practice into your next vendor renewal decision. Additional guidance on purchased compliance content captioning requirements is covered in the dedicated post on purchased compliance content captioning.

Can we use YouTube’s auto-captions as the production caption file in our LMS after updating a training video?

No. YouTube auto-captions do not meet WCAG 2.1 AA accuracy requirements for three independent reasons: accuracy, portability, and governance. On accuracy, YouTube’s auto-captions typically produce 80 to 90 percent accuracy on general content and lower than that on domain-specific training content with technical vocabulary, regulatory terminology, or speaker accents outside the training distribution — well below the 99 percent threshold required by WCAG 2.1 AA. On portability, YouTube auto-captions are not available as a downloadable caption file in a format suitable for LMS upload without a manual extraction step; the auto-generated captions are produced by YouTube’s system and are not independently usable as a production caption artefact. On governance, YouTube auto-captions are not subject to your caption governance policy, your accuracy QA process, your modification logging requirements, or your archive retention requirements — they exist in YouTube’s system and are not part of your accessible-format delivery infrastructure.

If you are hosting training video on YouTube as a distribution channel (for external-audience content or supplemental learner resources), you need to generate accurate caption files independently and upload them as closed captions on the YouTube video in addition to making them available in your LMS. The auto-captions WCAG and ADA compliance status post covers the accuracy data and compliance reasoning in detail for all major auto-caption platforms, including YouTube, the major LMS built-in auto-caption features, and Zoom’s live auto-caption accuracy.

Our LMS shows a “last updated” date on the course, but we’re not sure if this means the video was updated or just the course settings. How do we determine actual video content freshness?

The LMS “last updated” date on the course record is a course metadata timestamp, not a content file timestamp. It reflects the last time any field in the course record was changed — including non-content changes like enrollment setting updates, completion rule modifications, notification configuration changes, and description edits. To determine actual video content freshness, you need to access the file system rather than the course metadata: the video file’s last-modified timestamp at the storage layer tells you when the video was last uploaded or replaced, which is the relevant indicator of content change.

In practice, accessing the file-level timestamp depends on how your LMS stores and delivers video content. If your LMS delivers video via a CDN (which is standard for all major enterprise platforms), the CDN management console exposes file-level metadata including the last-modified timestamp for each asset. If the LMS stores video files in its own internal file system, the LMS administration interface typically exposes file-level metadata for content management staff, separate from the course-level metadata visible in the instructor or learner interface. Compare this video file last-modified timestamp against the caption file’s generated-at or last-updated timestamp from your captioning vendor or SaaS platform. If the video was modified more recently than the caption was generated or last reprocessed, the caption is a staleness candidate for manual review.

Is there a difference between updating a caption file for accuracy and updating it for compliance? Do both require archiving the superseded version?

Both require archiving the superseded version, but the audit trail purpose is different for each update type. An accuracy update — correcting errors in the original transcript that resulted from the captioning production process, not from audio change — generates a document that OCR may request to evaluate the quality of your caption programme’s self-correction process. Specifically, OCR investigators may want to see the original file (showing the errors), the correction event (documented in the modification log), and the corrected file (showing the resolution), as evidence that your programme has a functioning QA and correction workflow. The superseded version with the original errors is part of that story; without it, you have only the corrected version, which alone does not demonstrate that you identified and resolved the problem.

A compliance update — updating the file because the underlying audio changed — generates a document that establishes the version of the content that was accessible to learners during a specific historical period. OCR complaints and civil litigation discovery may ask for the caption file that was in use during a specific date range, which is exactly the superseded version that your archive preserves. Both categories of superseded file have distinct audit trail value: accuracy-correction archives demonstrate programme quality management; compliance-update archives demonstrate temporal content accuracy. Archive both, retain both for the full retention period, and log both in the modification record with the appropriate event type description.

We have 3,000 captioned videos in our LMS. How do we prioritise the staleness audit?

Prioritise by risk profile, not by volume or chronological content age. Risk profile in this context means the intersection of the trigger type most likely to affect the content and the audit exposure category of the content type. A framework that works for libraries of this size is a three-tier prioritisation: Tier 1, review first; Tier 2, review second; Tier 3, review last.

Tier 1 content — the highest audit exposure and most likely to be affected by active trigger types — includes safety procedures and safety-adjacent content (any video that describes a process, procedure, or equipment operation), regulatory and legal compliance training (HIPAA, OSHA, FINRA, ADA, code of conduct, data privacy), and any content that explicitly references named products, product features, or specific software UI elements that have changed in the past 18 months. For a 3,000-video library, Tier 1 typically represents 15 to 25 percent of the library — roughly 450 to 750 items — based on typical enterprise L&D content distribution patterns.

Tier 2 content includes onboarding and role-specific training content, content that references internal processes or policies that may have changed, and any content produced more than two years ago in a domain where external conditions (regulations, technology platforms, product landscape) have changed since production. Tier 3 content is evergreen professional development and skills training — communication, leadership, productivity, general professional skills — where the content is unlikely to have been affected by any of the four trigger types and the audit exposure is low. Within each tier, start with the content that has the highest learner-hour exposure: the courses with the largest learner populations and the highest completion rates over the past 12 months. These are the courses most likely to have affected the largest number of caption-dependent learners if they are stale, and they are also the courses that an OCR investigator is most likely to request in a document request that asks for a representative sample of your captioned training content.

For the specific audit execution methodology — how to pull the course population data from the LMS, how to structure the timestamp comparison, how to document the audit findings, and how to build the resulting remediation queue into a phased project plan — see the enterprise LMS caption audit methodology and the large-scale caption backlog remediation playbook, which together cover the full audit-to-remediation workflow for libraries in the 500-to-5,000 video range.

Closing: the trigger-to-archive workflow in practice

Caption file lifecycle management is not a new problem — every large L&D content library has outdated caption files attached to updated videos, and most teams have no systematic process for identifying or remediating them. The compliance exposure from stale caption files is real and growing: OCR document requests explicitly ask for caption files that were in use during the complaint period, VRAs require accuracy maintenance and reporting over multi-year monitoring periods, and civil litigation discovery can surface the modification history — or the absence of one — for specific caption files at specific points in time. A caption file that does not match its current audio is not a minor content quality issue; it is an accessible-format delivery failure for every learner who depends on that caption for access to the training content.

The operational solution is simpler than the compliance risk might suggest. A trigger-to-archive workflow that catches caption updates at the point of content change, applies the right remediation path based on the scope and type of the audio change, propagates the update to the LMS without losing learner records, and archives the superseded file before the replacement is uploaded — this workflow, embedded in the content development process as a concurrent task rather than a downstream cleanup, eliminates the accumulation of caption staleness without creating a separate administrative workload.

The eight failure modes in Section 9 all share a common root: they occur when caption lifecycle management is treated as an afterthought rather than as a built-in component of the content update process. The governance integration points in Section 8 — the intake form, the annual audit, the policy clause — are the structural interventions that make lifecycle management automatic rather than discretionary. With those governance structures in place, the operational work of lifecycle management flows from the same triggers, the same projects, and the same review processes that drive all content development, and no separate caption lifecycle management effort is required.

The caption programme annual review process provides the full annual audit framework including staleness detection as a standard component. The caption programme governance policy template provides the policy clause on lifecycle management ready to adapt. The caption compliance programme build guide covers the full programme structure, of which lifecycle management is one operational domain. And the caption compliance self-assessment checklist includes a scored evaluation of your current lifecycle management practices against the framework in this post, so you can identify the specific gaps to close and measure progress over time.