ApiaryActiveLive
Try: pause · settings · learn · wipe
← Community / Reading Room
CT
craft · 24 min read

Caption the Hand-Off: How to Preserve Who Said What in a Web Video

Two people are demonstrating a folding workbench. The customer asks, “Does it lock on this side?” The maker, standing outside the shot, says, “Only after you…

AI disclosure: Drafted with AI assistance for Apiary. The footage, speakers, company, and timings in the examples are fictional. Device behavior described here is an acceptance test to run, not a report of a device test already performed.

Two people are demonstrating a folding workbench. The customer asks, “Does it lock on this side?” The maker, standing outside the shot, says, “Only after you lower the lever.” A click follows. Then the customer says, “There it is.” You can capture all three sentences accurately and still make an inaccessible clip. If a muted viewer sees the customer’s hands for the whole exchange, they can easily assign both question and answer to the same person. If the click is omitted, “there it is” suddenly has nothing to refer to.

This problem isn't solved by an accuracy score measured against a transcript. It lives between three assets: the edited video, the timed caption file, and the way the site presents both. The fix is to build captions around changes in attention and meaning, then check the exact footage on a muted, phone-sized player. This guide uses a tiny fictional scene to make that process concrete.

Start with what a viewer has to infer

Play the scene once with sound and once muted. Don't write a cue yet. On the muted pass, ask three questions: Who appears to be talking? What visual or audible event changes the meaning? What remains confusing when the picture is watched without sound?

At the start of the workbench demo, a hand in frame points to the hinge. A completely clear shot of the customer asking a question may not need the words “Customer:” on screen. By contrast, an answer from someone out of frame needs a label if the viewer has no other reliable way to identify that voice. The click does not need its own line every time a tool does something. In this particular cut, it proves that the motion reached an audible state that the final sentence refers to. That makes it worth retaining.

A selective sound cue is better than cataloging every noise in a room. A steady background fan can distract in captions if it has no narrative effect. A warning tone that makes the person stop is information. An unexpected crash offscreen is information if it changes the scene or triggers a reaction. Use plain descriptions rather than supplying invented source, emotion, or identity. “[metal clicks]” describes what can be heard; “[dangerous latch breaks]” claims more than the track shows.

The W3C explanation of prerecorded captions discusses presenting speech and relevant non-speech information. The site owner still has to judge which details matter in this content. Merely providing a file called “Captions” doesn't show that it succeeded.

Create a small cue plan before you write the file

A useful cue planning sheet is about time as much as text. It can have one row per event with the start and end in the final edit, the speaker or sound, the picture at that time, and a note describing any ambiguity. Keep the first pass rough; the draft file is a tool for review rather than an infallible transcription result.

Time in final cutVoice or soundWhat the picture showsCaption decision
00:01.000–00:02.600CustomerHand pointing, face offscreenLabel speaker because voice cannot be inferred
00:02.850–00:05.150MakerSame close-up, maker still offscreenNew cue with a different visible label
00:05.350–00:06.200Latch clickHand moves; lever partly out of frameKeep brief sound cue
00:06.350–00:07.500CustomerShot unchangedShow who responded

If the actual cut changes to a clear two-shot before the follow-up, you can reconsider the final label. Don't decide once based on an interview transcript and copy the result to every derived edit. A fifteen-second vertical version may hide the original speaker at exactly the point where a horizontal composition made the speaker obvious. A still cutaway can create similar confusion.

Choose meaningful cue boundaries. Ending a cue after “Only after you lower” forces the next one to carry the object by itself. Breaking around a clause or natural pause usually preserves meaning better. Don't set a universal word-per-line rule solely for your editor's monitor; rendered line length varies with type size, actual phone width, player controls, and viewer preferences. Test the cues on the real canvas. For reading ease, make editorial changes to avoid unnecessarily fast cuts rather than trying to compensate for every edit with tiny type.

A WebVTT fixture that is explicit about speaker labels

The WebVTT specification defines the timed-text format and cue text syntax. This fictional clip can use a small sidecar file. The labels below are ordinary visible text. A WebVTT voice span can carry semantic voice annotation, but don't assume its annotation name will be printed as a speaker label by a given browser. Use visible text where identification depends on it, and test whichever markup and styling you choose.

WEBVTT

question
00:00:01.000 --> 00:00:02.600
Customer: Does it lock on this side?

answer
00:00:02.850 --> 00:00:05.150
Maker: Only after you lower the lever.

latch
00:00:05.350 --> 00:00:06.200
[metal clicks]

response
00:00:06.350 --> 00:00:07.500
Customer: There it is.

This is not a claim that every two-speaker clip needs those same labels. In an interview, a speaker's real, verified name might be useful once introduced. In a public demo where names are irrelevant, their role is enough. In a group conversation where voices interrupt each other over a cutaway, an extra cue or a deliberate edit may prevent an attribution error. Avoid relying only on color to distinguish people. In native caption controls, the viewer may choose styles different from yours, and the same color can look different on a different display.

The asset should match the clip it accompanies. Put its revision next to the exact video revision, not just in a generic folder named “latest.” An example says more than a naming policy: if a publish trims half a second of silence from the opening, the original cue start of 01.000 now appears half a second after the question begins. You can discover that in a muted playback test even though the file validates syntactically and the text itself is unchanged.

Wire it to a real player

HTML's track element lets a media element associate a timed text resource with the clip. A small example, assuming all files are on the same origin:

<video controls playsinline preload="metadata"
       poster="/assets/workbench-demo-r2.jpg">
  <source src="/assets/workbench-demo-r2.mp4" type="video/mp4">
  <track kind="captions" src="/assets/workbench-demo-r2.en.vtt"
         srclang="en" label="English captions">
</video>

The clip and track have the same revision in their filenames. This reduces the chance a deployment updates only one when the other has meaningfully changed. Verify the correct resource returned, its content type, the visibility of the caption control, and the ability to turn it on and off. If the video and track live on different origins, consult the relevant loading and cross-origin rules for the media setup. A network request for the MP4 that succeeds does not establish that the browser could fetch the VTT.

Keep the distinction between text tracks intentional. A subtitle track can present translated speech. A caption track covers audible information viewers need, which can include cues for sound and identity. A transcript can provide a full readable record separately. These resources can share source material, but one cannot silently stand in for the others. For example, a polished transcript without timing won't tell a viewer when to watch a partially obscured lever.

If there is a separate no-audio background preview, don't advertise its decorative open captions as proof that the main player has a working track. Conversely, a functional caption track in the full player doesn't make a social export automatically captioned. A player, an embedded copy, and a video uploaded elsewhere may have independent distribution and settings. Mark each target in the release record rather than assuming the accessibility affordance traveled with the title.

Review the difficult section, not only the first five seconds

A muted review is for testing meaning. Ask a reviewer who has not heard the original to watch the complete exchange muted with captions enabled. Then ask: what caused the sound, who answered, and did the action occur before or after the explanation? An answer that comes only from your own memory of the original clip does not validate the presentation.

Next, check playback with audio enabled. A cue that arrives a sentence early can disclose a response before the speaker says it. A cue that arrives late forces readers to connect a line with the wrong image. Watch again after seeking into the maker's answer because some custom players mishandle currently active captions when playback starts in the middle. If playback speed is a feature, try normal and one different speed; the cue should track media time, not a separate countdown based on elapsed wall time.

Make the sample destinations reflect how audiences actually see the clip. A large desktop player may have ample room for two lines. A phone in portrait or an in-app embed might place playback controls right over the lever. In the fixture, “Customer: Does it lock on this side?” is short, but the player's increased-caption-size setting can still make it wrap. Check whether the lines obscure the only important detail in picture and whether the control remains reachable using the interaction methods your site supports.

Here is a small handoff checklist with specific failure examples:

CheckWhat to verifyExample failure
Playback mutedScene still explains speaker turn and latch clickBoth lines look like one offscreen person
Playback audibleCues align with voices and meaningful soundFinal cue arrives during the maker's answer
Seek into cueCurrent caption appears and clears predictablyCaption only comes back after restarting
Portrait viewportCue and controls remain legible and do not hide essential contentTwo-line cue overlays the workbench lever
Track toggleCaptions can be enabled and disabled in actual embedTest works on direct page, embed hides control
Network proofMP4 and VTT describe the same cutCorrect filename points at superseded timecodes

A syntax checker may be useful for locating malformed timing or broken cue IDs, but it can't hear the latch. An automated accessibility check can't confirm the offscreen speaker's name unless you've given it a trustworthy reference. Use tools to eliminate mechanical defects, then use editorial review to verify meaning and ambiguity.

Fix ambiguity at the editing layer when text won't solve it

Imagine that a third voice interrupts the maker with “wait.” There is no identifying information in the shot. You could publish “[voice offscreen]: Wait,” which is accurate but may leave an essential safety demonstration unclear. If the third voice changes the instruction, consider a different edit. Let the demo show who interrupted or retain enough footage to explain what changed. Captions are not a good place to invent context the audio and image never established.

A similar problem comes up when the click is a graphic effect laid over footage. If the viewer needs to distinguish a real mechanical latch sound from a produced transition sound, adding “[metal clicks]” alone can mislead. Use a more accurate description or rethink whether the edit implies a mechanism really engaged. Source accuracy remains an editorial responsibility whether the viewer listens or reads.

Keep separate records for copy approval and caption review. Someone can approve the wording “Only after you lower the lever” without checking which revision of the shot the caption overlays. The video approver can approve the fit of the frame without watching it muted. A release owner can then see which checks they've actually earned, instead of assuming one review includes all of them.

The fastest correction is often targeted. If the media cut hasn't changed and a speaker label is wrong, edit the corresponding cue and rerun a small affected-window check plus a quick seek test. If a video has been recut, run the timing review against the whole new master. Don't move every cue by a fixed offset if edits were made in multiple places. The correction should identify the affected asset and revision so viewers and support can tell whether an older embedded copy needs its own update.

Make it part of the ordinary release

Use the same acceptance structure for the next real demo. Capture speaker changes in the review notes. Assign one person to produce the cue track against the final cut, another to review disputed words or sounds, and a release owner to verify the delivered file pair. A small team may give one person all three jobs, but the checks remain different. A review can contain exact timestamps and should say if any line remains uncertain.

Keep the reviewed master, timed file, and associated version together. If a short clip moves to another distribution service, record that new delivery target and repeat the checks it can actually support. Don't label the source footages as checked just because a copy was approved elsewhere. The right report is specific: “English captions checked on the mobile embed for r2, including the offscreen answer and latch cue; larger-text setting and seek passed.” Add other clients when actually checked, not as an unearned claim of universal support.

One accurate speaker label, one useful sound cue, and a deliberate muted review do more for this particular scene than ten decorative changes to the video player. The real question isn't how many words the captioner transcribed. It's whether a viewer who can't hear the track can follow the same exchange and arrive at the same conclusion.

Include language changes in the handoff notes

A brief exchange in another language needs a deliberate editorial choice. If both viewers who hear the source and those who read it should understand the meaning, decide whether the original edit already supplies a translation on screen. A separate caption overlay that repeats a hard-coded translation can create two competing lines. A generic cue such as “[speaks Spanish]” may faithfully identify the language but omits meaning if the clip expects the audience to understand the instruction. For a release with multiple tracks, verify translations against the final edit and name each language accurately rather than offering an unlabeled secondary track.

The same principle applies to names. Automatic captioners can produce plausible spellings for unfamiliar brands, places and trades. Verify the name against an approved script, product label, or another authoritative source when available. A review note can flag the exact uncertain cue without replacing the whole script. If the word affects safety or the proper operation of the product, defer publication of the affected instruction pending clarification rather than accepting a plausible guess.

Finally, distinguish what is being captioned from what appears as ordinary text in the shot. A tutorial can show a hidden timer countdown and then refer to it without speaking the number. Captions may already convey the sound and speech perfectly while a viewer who cannot see the timer still needs another access path for the visual information. That is a reason to evaluate description or transcript needs separately, not to claim that one caption track covers every form of access.

Make a speaker map before assigning names to cues

On a small shoot, the first caption draft is often written by somebody who was standing next to the camera. That person knows the lead and the handoff, recognizes a voice from the hallway, and understands which of several similar items the speaker is pointing at. A person receiving the published clip has none of that context. A caption review should start on the listener's side of the gap. Give a reviewer the finished picture with muted playback and ask them to follow one real exchange. The goal is to discover which identity and timing facts the current file fails to provide, not whether the editor can recite everything that was said.

Create a minimal speaker map from the permitted source and final edit. Use the role the viewer needs, not every internal name the production team knows. In a fictional library workshop with a host on screen and an instructor answering from a stockroom, HOST and INSTRUCTOR may be enough when their names are never established for viewers. If someone introduces themself on camera, a visible first name can be appropriate afterward. In archival work or a scene with no established identity, a neutral label such as VOICE FROM THE HALL may better reflect the evidence than resolving the identity from unrelated employee records. Ensure labels aren't changing merely because a person walks into frame.

For a synthetic receipt, write all exchanges that could be confused if the visible mouth happens not to move. At the first turn, the host asks where a spare handle goes. At the second, a quiet off-screen voice says to hold it. Before the viewer sees the instructor, the host repeats hold as a question. That repetition makes the speaker assignment consequential. A combined caption reading “Hold it. Hold it?” stops short of telling the reader who asked the question. The exact words can be right while the exchange is functionally wrong. The map should record that scenario as a test case requiring distinct speaker turns or unmistakable visible context.

Give labels a consistent spelling and position when they recur. If the cue begins with INSTRUCTOR, don't switch to TEACHER for the same person merely to save characters in the next caption. Keep capitalization and punctuation aligned with the style guide. When using optional position or voice features, include a rendered check rather than assuming the player exposes every intended cue annotation. Make any necessary information visible in a form the target viewer can actually read. Labels should clarify context, not look like dialog or require color perception to distinguish two speakers. A semantic marker is valuable when a tool uses it, but its presence is not proof that a label will appear in the video player.

An editor may intentionally withhold identity for a genuine narrative reveal. A caption shouldn't spoil the reveal by naming a speaker before a hearing viewer could reasonably tell who it is. Use a description that preserves the shared experience and update the label at the corresponding moment. The approach differs if the introduction is already audible but outside the visual frame; in that case masking it can create an information gap. Test the scene from the final released cut, not an assembly in which the introductions appear in an earlier order. Keeping the speaker map next to the final video receipt makes this distinction much easier to review after an edit.

Run a timed review around the contested exchange

Build a tiny cue-by-cue worksheet rather than trying to judge a whole ten-minute clip in one sitting. Choose the last clear exchange before the contested handoff, the handoff itself, and the first meaningful response after it. Mark which cue should begin when an utterance or relevant sound starts and what should be visible near its end. Then annotate only the differences a viewer would use to follow the subject: who is talking, what they are answering, whether a non-speech sound changes the action, and whether the cue can be read while the corresponding visual event is still accessible. A cue timeline is not an exhaustive transcript of every mouth movement.

Try a deliberately poor draft in a safe synthetic test. Let speech from the off-screen instructor begin just as the host's hand lifts an object on screen. Show the words before the instructor starts, without a label. Ask a muted reviewer to describe who requested the next step. In a second draft place the cue at the audible turn with a visible role label, holding the same text and video otherwise. Weaknesses become easier to identify when the test changes one relevant thing. Do not conclude that every slight timing variation is an error. Look for cases where anticipation, lag, or grouping changes who a viewer thinks did or said something.

Next look at reading time under realistic conditions. A long speaker label and a quotation can consume more screen width than the printed worksheet suggests. You might remove an unnecessary repeated label when there is no ambiguity, but not the information needed to distinguish the turn. Break a long exchange at a natural boundary rather than using a hard line-length rule regardless of wording. Some cues need enough time for a viewer to read two short lines while they also watch what the instructor is demonstrating. A good edit can shorten the spoken source or swap an obstructive shot; it doesn't have to compress critical cues until the audience gives up.

The moment when the caption disappears can be as useful as the start. In the made-up clip, the phrase hold the handle is immediately followed by a visual close-up with a small label near the bottom. If the cue trails far beyond the utterance it can obscure the label after readers already received the instruction. Conversely, forcing it to vanish exactly on the final consonant may leave too little time to read it. Try appropriate layout or video decisions and verify the results on the actual video. The worksheet should note which fact matters in the scene and what changed. No one should have to trust a numeric timing threshold detached from the content.

Review at ordinary speed before slowing down to diagnose. Slower playback can help an editor check individual sounds and hard words, but it changes the viewer's reading budget. Conversely muted ordinary playback gives the reviewer the intended timing with no rescue from audio. When instructions are repeated, test only one pass and then repeat it knowingly. Someone seeing the clip for the fifth time may already know what the author meant and fill in missing transitions. A practical report can explicitly note which observations came from a first viewing and which from stepping through the clip.

Verify which sounds carry information in this cut

A transcript might include ten seconds of footsteps and room noise. Captions need not plaster an environment with a sequence of repetitive atmosphere labels. Instead identify sounds carrying information that a hearing viewer gets at that point. In a fictional exhibit workshop, a sharp latch click tells the worker the object is closed, and the host immediately says now. The click explains the change in action and should be considered for a cue. A quiet background vent is not consequential unless the next line refers to it. If the source audio lacks the latch sound or it's indistinct and the action is fully shown, evaluate the phrasing accordingly. Captioning is not a license to invent confident sound events.

Try a second fictional example in which a door shuts out of frame. The participants pause, somebody looks back, and then the host asks if the room is clear. A label such as [door shuts] can connect those beats. A label such as [angry interruption] goes beyond the evidence unless the scene makes that interpretation explicit. Describe relevant sounds in ordinary words and tie them to the correct beat. If a sound source is unclear, be as specific as useful without guessing. A reviewer should be able to point to a sound that was present and explain what it contributes, not merely insist that the published file contains a certain number of bracketed phrases.

Music can perform several roles. Quiet bed music that never influences dialogue or sets up an obvious change is different from a melody whose start cues a slide or whose abrupt stop makes the next sentence intelligible as an interruption. If a clear change matters, identify it succinctly. Do not write out arbitrary lyric text from a clip without appropriate rights and need. If a sound cue competes with the only usable speaker identification at a dense cut, prefer a meaningful grouping that does not imply the wrong person said the words. The right question is what the viewer needs to participate, not whether every noise can be assigned a label.

For a synthetic quality test, prepare one scene where removing a sound label causes reviewers to misread the host's reaction and another scene where labels for irrelevant movement make it harder to read important instructions. This guards against two simplistic rules: label no sounds, or label all sounds equally. A well-scoped review can decide based on ordinary playback context. Keep a disagreement list for editorial decisions rather than hiding the judgment under a supposedly automated accuracy score. The specification and accessibility guidance inform your objectives; they do not assign editorial importance to sounds in footage they have never seen.

Match track behavior to the real publishing surface

A captioning problem can be introduced after the file is right. If a web player loads a different language track by default, serves an older cue file, or never exposes buttons on a narrow screen, repairing punctuation in a local WebVTT file will not fix the experience. Write down how the captions are expected to appear: available in the player picker, not automatically selected unless the product intends that behavior, and associated with the published clip. Check the effective video and track version by a stable release identifier or equivalent asset mapping. A player option labeled English only proves that a label exists. It doesn't show that the right English track is actually shown.

Imagine an invented release with English captions and an English transcript link. The transcript is correct, but the player fetches an old track built against an earlier video cut. In the first scene the words align because the cuts share an intro. At the first edit, the track falls behind. This is why the handoff fixture should include one timed editorial marker after a known edit, rather than only checking the initial loading state. If the track has an intentionally different revision from the video because captions were corrected, the mapping should name both associated parts. An unsigned pair of files with matching names do not prove they belong to the same version.

Automation can parse the track and flag structural errors, missing expected metadata, or cues outside the known clip duration. A preview check can establish that a particular browser displays an actual cue when the track is selected. A muted human review can assess intelligibility and whether the visual content remains visible. These are different signals; keep those three rows separate in the release receipt. A report that says validated syntax but untested muted playback helps prioritize the right next action. It should not be converted into checked accessibility with no evidence about real content or device controls.

Include a case that detects if former conditional behavior changed. An ordinary embedded player may look different when opened directly, in fullscreen, or inside a mobile layout with sticky controls. Test the mode the audience normally uses and at least one important alternative. On a narrow view, check whether settings cover the bottom edge and whether captions remain readable once settings close. Fullscreen may use different native UI. Record the device and browser if the behavior is specific to them. Do not turn a single preview into a claim about every browser combination. Viewer settings can present legitimate differences; a usable test explains which of them matter for the release.

A caption can be selectable, while operation of the player still fails for a reader who navigates without a pointer. Where the intended controls are keyboard operable, give a reviewer a clear sequence: start from the neighboring content, reach the player, select captions, return to the video, and reach the next item in the page. Take note of the actual focus behavior and observed UI. Another reviewer can verify spoken feedback using the available assistive technology, with results labeled for that configuration. Matching one role name or one keyboard test alone doesn't settle the broader experience. Save what was or wasn't exercised rather than making an unsupported blanket compliance claim.

Use an editorial handoff that can survive a recut

Caption review tends to bounce among production, copy editing, and engineering. A compact handoff can keep the first reviewer from making guesses that the last reviewer later treats as facts. It names the current media, the cue file, which checkpoints were reviewed, and any uncertain attribution. It gives the expected meaning of the crucial exchange in plain prose rather than only fixture times. It also names one blocker with the specific step needed, such as confirming a blurred spoken name from the clear source recording. Do not resolve ambiguity by guessing from a web search for a brand or person's name that happens to sound similar.

Consider a synthetic correction. A second edit removes two seconds of silence before the instructor answers, and the response now arrives over a new insert shot. It would be easy to shift every cue earlier by two seconds and assume the file still works. But the insert may cover a speaking face that helped viewers identify the voice. The revised track needs both clock alignment and a fresh check of whether the speaker is apparent. Conversely an unrelated correction to the visible label might leave cues unchanged if the interaction timing is identical. Define review checkpoints for the changed part, then spot-check an earlier and later part to see whether the version mapping was accidentally replaced.

Ask the final reviewer to report what they observed rather than only approving the file. For example: “At normal speed with sound muted, the pause instruction was attributed to the off-screen instructor before the host's question, and the object label was still visible afterward. Mobile fullscreen output not checked.” That statement is specific enough to be compared with a later report without naming private footage or implying a global pass. If the chosen copy has an intentional unresolved limitation, state what the team decided and who owns the next review.

Keep this workflow small on an ordinary release. One clear speaker transition, one meaningful sound and one protected visual marker make an effective focused check. When incidents require more detail, extend the cue map rather than changing the definitions of passing. Good captions are valuable even as player styles and implementation stack change. The durable work is identifying the distinctions the viewer needs, making them legible in time, and proving the intended file actually reaches the audience.

Compare the transcript when the artifact allows it

A linked transcript can help someone search or revisit a detail, but it serves a different timing purpose from an overlay. In the fictional workshop, a transcript reader may want one accessible description of the off-screen exchange on its own. A caption viewer instead needs the turn as it happens while watching the object move. Compare the two for stable names, factual wording, and any actual editorial corrections. A discrepancy may reveal that one artifact still points to an earlier cut. The appropriate fix isn't to remove timing information from the caption or to attach a word-perfect but unreadable block of text to the player.

If both artifacts are corrected after release, keep the change scoped. Record that a role label or sentence was corrected, associate the update with the correct source cut, and test whatever locations link to each asset. A sidebar can still point at an older transcript while the player's caption picker has the latest version. One quick before-and-after check on the contested turn and a destination check on the link prevent an edit in one surface from being mistaken for a fix in both.

Primary references

Frequently asked
What is Caption the Hand-Off: How to Preserve Who Said What in a Web Video about?
Two people are demonstrating a folding workbench. The customer asks, “Does it lock on this side?” The maker, standing outside the shot, says, “Only after you…
What should you know about start with what a viewer has to infer?
Play the scene once with sound and once muted. Don't write a cue yet. On the muted pass, ask three questions: Who appears to be talking? What visual or audible event changes the meaning? What remains confusing when the picture is watched without sound?
What should you know about create a small cue plan before you write the file?
A useful cue planning sheet is about time as much as text. It can have one row per event with the start and end in the final edit, the speaker or sound, the picture at that time, and a note describing any ambiguity. Keep the first pass rough; the draft file is a tool for review rather than an infallible transcription…
What should you know about a WebVTT fixture that is explicit about speaker labels?
The WebVTT specification defines the timed-text format and cue text syntax. This fictional clip can use a small sidecar file. The labels below are ordinary visible text. A WebVTT voice span can carry semantic voice annotation, but don't assume its annotation name will be printed as a speaker label by a given browser.…
What should you know about wire it to a real player?
HTML's track element lets a media element associate a timed text resource with the clip. A small example, assuming all files are on the same origin:
References & sources
  1. Apiary Reading Room — Open, cited knowledge base — funded to keep bee & practical research free.
From the Apiary Reading Room. Opinion & editorial — not financial advice. We don't overclaim.
More from the Reading Room