Build tracks · build-plan/02-collateral-publishing-teleprompter.md

Build Plan §02 — Collateral Vault: Take it · We post it · Record it

Scope: PRD §9.3 (collateral vault, three delivery modes) and the admin collateral screen in §10. Date: October 3, 2026. Platform facts below were checked against primary docs on this date; anything not confirmable is marked. Reads with: PRD.md (authoritative), PRODUCT.md (authoritative: affiliate is on a phone, in a mobile browser, no native app in v1). PRD-Activation-Platform.md §7.3 is used for ideas only.

0. The short version

• Mode 1 (Take it) is pure product work with no external dependency. It ships first and alone, and it is where most affiliate activity will land for the first year regardless of what else exists. Build it to be excellent.
• Mode 2 (We post it) is gated by platform approvals, not code. Every network worth posting to requires an app review (Meta, TikTok) or has structural limits (LinkedIn self-serve is fine; X is pay-per-use and marginal; YouTube is now cheap). Recommendation: go through an aggregator (Ayrshare Business plan) for v1, start Meta and TikTok reviews in parallel on our own app anyway, and migrate network-by-network only if volume or cost justifies it.
• Mode 3 (Record it) is buildable as pure web on today's iOS Safari and Android Chrome, with one hard trap (in-app browsers block the camera) and one hard requirement (server-side remux/transcode of every recording). It is v1.5: it depends on Mode 2's publish step for the hand-off and on the upload/transcode pipeline, and it needs real-device QA that Mode 1 does not. A Capacitor wrapper is not needed to ship it; it becomes worth considering only if recording quality or in-app-browser arrivals turn out to matter more than expected.

1. Asset model

1.1 Entities

collateral_assets            the logical thing ("Launch carousel", "Why I use HealthyApp script")
  └── asset_versions         immutable; a new upload or edit = new version
        └── asset_renditions per-platform crops/encodes, generated from a version
scripts                      a kind of asset_version body (text), with read-time
asset_disclosures            which disclosure block attaches, and how (caption / spoken / on-screen)
asset_restrictions           program / tier gating
affiliate_asset_marks        per-affiliate: used / skipped / posted / seen, plus per-rendition download+copy events

1.2 Tables

collateral_assets

asset_versions (immutable rows)

A version that is referenced by any publish_jobs or recordings row is never deleted. Retiring an asset marks it retired; the versions remain so old posts and recordings resolve.

asset_renditions

scripts — not a separate table; a script is an asset_versions row whose asset category = 'script'. It carries three extra columns on asset_versions that are null for other kinds:

A script version always has a caption_text (the caption that goes out with the recorded video) and disclosure_placement must include spoken_in_script and on_screen_overlay (see §7, FTC video disclosure).

disclosure_blocks

asset_restrictions

affiliate_asset_marks

asset_events (append-only, high-write)

asset_events is what the "collateral usage vs. revenue" report in PRD §10 is built from: join asset_events.campaign_label to clicks.campaign (every Mode 1 copy and every Mode 2 post stamps a campaign label of the form col-<asset_short_id>-<platform>), so revenue can be traced to an asset without the affiliate doing anything.

1.3 Token substitution rules

Tokens are resolved at read time, per viewing affiliate, never stored resolved. Supported set (PRD §9.3 plus two needed by Mode 2):

Rules:

1. Unknown tokens fail validation at admin save time; a version with an unknown token cannot be published.
2. Resolution is a pure function (body, affiliate, product, campaign, disclosure) → string in packages/core, unit-tested, shared by portal, publish worker, and teleprompter.
3. For spoken scripts, {{short_link}} is rendered as a speakable form: go dot affiliate dot coach slash J-WHIST with the code letter-grouped, and the printed caption carries the real URL. (Novices reading "https://" aloud is a real failure mode.)
4. Per-platform link placement overrides apply after substitution (§3.5): on Instagram the caption's {{ref_link}} is replaced with "link in my bio" text and the link is pushed to the bio-link step.

1.4 Program/tier restriction evaluation

Evaluated in one RLS policy so the vault list query is a single select. The affiliate never sees an asset they are not eligible for; there is no "locked" teaser in v1 (a locked teaser is a decision for later, see §8).

2. Mode 1 — Take it

2.1 Behaviours (exact)

Everything above works with zero third-party dependencies and zero platform approvals.

2.2 Rendition matrix

Generated automatically on admin upload (§5). Admin can override any cell by uploading a hand-cropped file.

Rule: every rendition's video is MP4, H.264 High, AAC, +faststart, ≤ 30 fps, because that intersection is accepted everywhere. We never deliver WebM to an affiliate.

2.3 Where the tracked link lands, per platform

This is the part Mode 1 has to be honest about, because the affiliate will blame the link when the real problem is platform mechanics.

2.4 Disclosure baked in

{{disclosure}} is forced into every caption at caption_start for Instagram/TikTok/X (short formats where a reader may not expand) and caption_end for Facebook/LinkedIn/email unless the admin pins otherwise. Images and videos destined for Instagram/TikTok/Reels carry an on-screen disclosure burn-in as a rendition option (admin toggle per asset, default on for anything that is itself an endorsement). See §7.

3. Mode 2 — We post it

3.1 Per-network findings (as of October 3, 2026)

Two cross-cutting facts: (1) every Meta and TikTok approval requires a working UX to record a demo video, so Mode 2 cannot be approved before it is built; (2) all of them require the media to be fetchable at a public HTTPS URL at publish time — renditions are served from a short-lived signed URL (Supabase Storage) generated inside the publish job.

3.2 Build vs. aggregator

Options compared

Cost model (Ayrshare Business), affiliate connected-count scenarios

Versus direct: zero marginal cost, but ~6–8 weeks of engineering across six OAuth flows + token refresh + six media upload protocols + per-network error taxonomies, plus 4–8 weeks of calendar time for Meta and TikTok approvals that cannot start until the UX exists, plus the ongoing cost of chasing API deprecations (Meta deprecated graph-video.facebook.com this year; X repriced twice).

Recommendation

1. v1: Ayrshare Business. It is the only option whose core abstraction (a user profile that our customer connects through a white-labelled page we open with a JWT) matches our shape without adaptation. $599–1,200/month is less than one engineer-week, and it unblocks IG + FB + TikTok + LinkedIn + YouTube on day one with approvals already held.
2. Only connected affiliates cost money. Profiles are created lazily at the moment an affiliate first taps "Connect an account", and deleted after 90 days with no connected networks, so the bill tracks real Mode 2 users, not sign-ups.
3. In parallel, register our own Meta app and TikTok app and start Business Verification now (it is a prerequisite for everything Meta and takes weeks of calendar time, little engineering time). We don't build against them yet.
4. Migration trigger: when monthly aggregator spend exceeds ~$2,500 for three consecutive months or a network we depend on breaks through the aggregator twice in a quarter, migrate that one network to direct behind the same publish_adapter interface. The adapter interface (§3.4) is designed so each network is independently swappable.
5. X ships only as aggregator pass-through, off by default, with the per-post cost surfaced to the admin.

3.3 OAuth connection manager

Affiliate-facing surface: Settings → Connected accounts.

3.4 Publish queue

publish_jobs

States

draft → confirmed → queued → sending → posted
                      │          │
                      │          ├→ retrying (transient: 5xx, rate limit, token refresh)  → sending
                      │          └→ failed   (permanent: media rejected, policy, revoked token, review-gated)
                      └→ cancelled (affiliate, or connection removed, or asset retired with revoke_now)

Adapter interface (packages/core/publish/adapter.ts): preflight(job) → PreflightResult, send(job, signedMediaUrl) → {externalId, url}, status(externalId), mapError(e) → {code, retryable, userMessage}. One implementation per network over the aggregator; a direct implementation can replace any one later.

3.5 The "exactly what we will post" preview

PRD §9.3 calls this the thing that makes the feature usable. Requirements:

1. Rendered per network, in that network's frame. A faithful (not pixel-perfect) mock of the IG feed card / Reel / TikTok / LinkedIn / Facebook post, built from our own React components, showing: the exact rendition image or the first frame of the exact video, the handle and avatar from social_connections, and caption_resolved with the disclosure visible and the link shown where it will actually be (for Instagram: "Link in my bio" text in the caption plus a second card: "We'll also copy your tracked link so you can paste it in your bio — Instagram doesn't allow clickable caption links").
2. Caption is editable in place, except the disclosure span, which is rendered as a locked chip. Edits change caption_resolved on the job; they do not touch the asset.
3. Shows the literal API payload truthfully in plain words: "Posting as @handle · Instagram Reel · 0:38 · caption 412 characters · scheduled Tue 9:00 AM your time."
4. Scroll-to-confirm. The Confirm button sits below the full preview; for N networks, the affiliate swipes through N previews and confirms once. Confirmation freezes caption_resolved and writes the consent snapshot.
5. What we show is what we send. The preview and the adapter consume the same caption_resolved and rendition_id. There is no second rendering path.

3.6 Consent

Shown once at first connection and re-shown whenever the text version changes; the specific version accepted is stored in publish_consents and referenced by every job:

• We will post to the accounts you connect only when you tap Confirm on a specific post; we never post automatically.
• Posts will carry the required disclosure, and you may not remove it.
• You are the publisher: you are responsible for the content on your account; you can delete any post from the network at any time.
• We store the text and media we post, and the post's URL, for your records and ours.
• Disconnecting removes our access immediately and cancels anything scheduled.
• Per-network consent is additionally captured by the network's own OAuth screen; for TikTok, the affiliate chooses the privacy level and commercial-content disclosure on each post (TikTok requires this in-UX).

4. Mode 3 — Record it

4.1 Platform evidence

4.2 Architecture decision

Pure web, with an in-app-browser fallback. No native wrapper for v1.5.

Reasons:

• Every required primitive — camera preview, MP4/WebM recording, overlay text, wake lock, resumable upload — works in iOS Safari and Android Chrome today. The things that do not work (orientation lock on iOS, camera in in-app WebViews) are not fixed by a PWA either; only a native app fixes them, and PRODUCT.md rules out a native app in v1.
• Quality: the browser gets the same camera and hardware encoder a native app does. The gaps (no manual exposure/focus, no 4K, slightly less control over bitrate) do not matter for a 45-second selfie clip that Instagram will re-encode.
• Capacitor is a real, already-exercised option for this owner. When to revisit: (a) analytics show > 25% of record-flow arrivals are in in-app browsers and the "Open in Safari" fallback converts poorly; (b) iOS codec drift produces a release where MP4 recording is broken for weeks (as the 18.4 beta briefly did) and we need a stable path; (c) affiliates ask for a camera-roll import + trim flow that a native picker does much better. Capgo's camera-preview plugin provides a native preview with HTML overlays, so the teleprompter UI would port unchanged. Until one of those triggers fires, a wrapper adds App Store review, update latency, and a second deploy target for no measurable gain.

In-app-browser fallback (required): detect the WebView by user agent (Instagram, FBAN|FBAV, musical_ly|BytedanceWebview, LinkedInApp) and by probing navigator.mediaDevices?.getUserMedia presence; show a full-screen card: "Your camera isn't available inside Instagram's browser. Tap ⋯ → Open in Safari/Chrome." with a Copy link button and, on Android, an intent:// link that opens Chrome directly. Do not try to start the camera in that state.

4.3 Teleprompter — PRD requirements → technical requirements

Additional requirements not in the PRD but forced by the platform:

• Permission pre-prompt. A friendly card before the real getUserMedia call ("We'll ask for camera and microphone. Nothing is saved until you tap Keep."). iOS asks per site, remembers for the session; denied permission is detectable via NotAllowedError and gets a "how to re-enable in Settings" card.
• Audio. echoCancellation: true, noiseSuppression: true; mono is fine. Bluetooth headsets are honoured automatically.
• Mimetype selection. Prefer in order: video/mp4;codecs=avc1.42E01E,mp4a.40.2, video/mp4, video/webm;codecs=vp9,opus, video/webm;codecs=vp8,opus, video/webm. Record with videoBitsPerSecond: 6_000_000, no timeslice (single blob on stop) to avoid the Safari empty-chunk bug.
• Max take length 3 minutes hard, with the script's read_time_s × 1.5 shown as a soft target bar.

4.4 Review / retake / trim

4.5 Upload pipeline

browser ── TUS resumable (6 MB chunks) ──► Supabase Storage  bucket: recordings-raw/{affiliate}/{recording_id}.{ext}
                                                │
                                                ▼  storage webhook → enqueue
                                   transcode job (Vercel fn, maxDuration 300, ffmpeg-static)
                                                │  ffprobe → -ss/-to trim → scale/pad to 1080x1920 →
                                                │  libx264 High, crf 20, 30 fps, aac 128k, +faststart,
                                                │  optional disclosure burn-in (drawtext, 0–3 s), loudnorm
                                                ▼
                                   recordings-ready/{recording_id}.mp4  + poster.jpg
                                                │
                                                ▼
                                   recordings.state = ready → portal polls / Realtime → "Publish" enabled

recordings

Rules: signed TUS upload URL minted by our API per recording (24 h validity); raw files auto-deleted 7 days after ready; ready files retained 90 days or until the affiliate deletes; transcode failure returns the affiliate to Review with "We couldn't process that take — try recording again" and keeps the local blob so nothing is lost.

4.6 Disclosure burn-in

Default on for every Mode 3 recording: drawtext of disclosure_blocks.short_text in a high-contrast band, top-left (outside the platform UI safe zones), seconds 0–3, plus the spoken line already in the script. This satisfies the FTC's position that a disclosure in a Reel/TikTok text description alone is not clear and conspicuous (§7). Admin can disable per asset for non-endorsement scripts (e.g. a pure "how to" clip) — recorded in the version's disclosure_placement.

4.7 Hand-off to Mode 2

When recordings.state = ready, the Review screen's primary button becomes Publish, which opens the Mode 2 composer pre-filled: recording_id, the script version's caption_text resolved, networks limited to those that accept 9:16 video (IG Reel, FB Reel, TikTok, LinkedIn, YouTube Short, X). The preview uses poster.jpg plus the ready MP4. Secondary button: Download (Mode 1 behaviour on the ready file; on iOS this saves to Photos via the share sheet). If the affiliate has no connected accounts, Publish shows the connect flow inline and then returns to the composer with the recording intact.

5. Admin authoring

Scripts

• Editor: plain textarea with a token palette (tap to insert), live word count and read time at 150 WPM in the header, a target-seconds field, and a warning banner when read time > target by more than 20%. Read time recomputes on every keystroke using a sample affiliate ("Jamie Rivera", code JRIVERA) so token lengths are realistic.
• A "Preview on phone" QR opens the actual teleprompter screen in draft mode on the admin's phone, so they can read it at speed before publishing.
• Required before publish: {{spoken_disclosure}} present, caption text present with {{disclosure}}, claims checkbox ("I checked this against the claims page") ticked; the claims page link sits beside the checkbox.
• Versioning: Save creates a draft version; Publish makes it current; prior versions are listed with diffs (word-level) so a compliance edit is auditable.

Assets (image / video)

• Upload once (any size ≤ 500 MB, direct-to-storage TUS from the admin browser). On arrival, the rendition job runs ffprobe/sharp and generates every row in the §2.2 matrix that the asset's category calls for, using smart crop (sharp attention strategy for images; centre-crop with letterbox-avoidance for video, with an admin-adjustable focal point overlay if the auto crop is wrong).
• Each rendition cell shows a thumbnail, a "Replace" control to upload a hand-made crop, and the derived file size. A video asset additionally gets a poster frame picker.
• Disclosure: choose block (default: product's current), placement checkboxes, "burn into video" toggle.
• Restrictions: program multi-select, min tier.
• Publish: the asset becomes visible; "New" badges appear for affiliates. Retire: choose revoke_now or grace_until, see count of scheduled jobs affected.
• Usage panel: views, copies, downloads, posts, and attributed clicks/conversions/revenue via the col- campaign stamp — the "collateral usage vs. revenue" report, one asset at a time.

6. Build sequence

Sizes: S ≤ 2 days, M ≤ 1 week, L 2–3 weeks. Dependencies in brackets.

Phase 6 — Mode 1 (ships first, must be excellent alone)

Exit: an affiliate on a phone copies a disclosed, link-substituted caption and saves a correctly cropped JPEG for Instagram in under 30 seconds.

Phase 6b — Mode 2

v1 Mode 2 networks: LinkedIn, Instagram, Facebook Page, TikTok — all available on day one through the aggregator, whose approvals we inherit. Wait on our own approval: nothing blocks v1, but our own Meta and TikTok apps (6b.0) are the migration path and should be in review during 6b so the option exists. YouTube and X ship in v1 as low-priority adapters, X disabled by default.

Phase 6c — Mode 3 (v1.5)

Why v1.5, not v1: (1) its publish hand-off needs Mode 2 to exist; (2) its upload/transcode pipeline is the only part of the whole product that needs ffmpeg at runtime on user content, which deserves its own hardening pass; (3) it is the one surface that needs a physical-device QA matrix rather than browser tests; (4) PRD §14's success criterion — a first-time video maker gets from script to posted without leaving the portal — is only meaningful once Modes 1 and 2 are proven and there is a real affiliate population to test it on. Shipping it two to four weeks after v1 costs nothing in the plan; shipping a half-tested camera surface in v1 costs trust.

7. Risks

8. Open decisions (recommended defaults)

Resolved Oct 3, 2026 — see BUILD-PLAN.md §5. Q1 Ayrshare. Q2 X is enabled but gated: only affiliates with an open balance (pending + approved > $0) may post to X, and the per-post link fee is deducted from their payout as a fee ledger row referencing the publish_job; the composer states the cost before posting. Q3 affiliates may edit anything in the composer, including the disclosure — no locked span; warn once, log the removal, post. Q4 no review queue in v1. Q5 burn-in OFF by default, affiliate opt-in per take; the spoken disclosure in scripts stays mandatory. Q9 v1.5 stands.

1. Aggregator choice — Ayrshare Business at $599/mo from day one, or Zernio's per-account pricing to start cheaper? Default: Ayrshare. Its user-profile + white-label JWT linking model is the exact shape we need and it is the incumbent; Zernio is cheaper at low counts but just rebranded and its pricing changed with it. Revisit if Ayrshare spend passes ~$2,500/mo.
2. Enable X at all in v1? Default: adapter built, disabled; admin can enable per program and sees $0.20/post.
3. Can affiliates edit captions in the composer? Default: yes, except the locked disclosure span; log the diff. Alternative: locked captions in v1 for HealthyApp because of health-claims exposure. If Jonathan wants zero claim risk, lock them for HealthyApp only.
4. Admin review queue for edited captions / recorded videos before publish? Default: no in v1 (it would reintroduce the daily admin login the product is designed to eliminate); build as a v1.5 option gated per program.
5. Disclosure burn-in default on for all Mode 3 recordings? Default: on. Admin can turn it off per script.
6. Show locked/ineligible assets as teasers to motivate tier progress? Default: hidden entirely in v1. Teasers imply earnings-based incentives that start to look like rank mechanics; keep it plain.
7. Default scroll speed 150 WPM? Default: yes, persisted per affiliate after first adjustment.
8. Recording retention? Default: raw 7 days after ready; ready MP4 90 days or until deleted; posted videos' URLs kept indefinitely for audit.
9. Mode 3 timing — ship with v1 or 2–4 weeks after? Default: v1.5, as argued in §6. If the first HealthyApp cohort is heavily novice and has literally nothing to post, pull 6c.1–6c.4 forward and ship the Download hand-off before the Publish hand-off.
10. Capacitor wrapper? Default: no. Revisit only when 6c.7 shows > 25% in-app-browser arrivals with poor fallback conversion, or an iOS release breaks MP4 recording for more than one point release.
11. Instagram login path — Instagram Login (no Facebook Page) vs Facebook Login? Default: Instagram Login for our own app registration (6b.0); it removes the Page requirement that would exclude many Creator-account affiliates. Through the aggregator this is handled for us.

Sources

Instagram / Facebook (Meta)

• Meta — Instagram Platform: Content Publishing (account types, media types, JPEG-only, 100 posts/24 h, permissions for Instagram Login vs Facebook Login): https://developers.facebook.com/docs/instagram-platform/content-publishing
• Meta — IG User Media reference (carousel limits, Stories constraints): https://developers.facebook.com/docs/instagram-platform/instagram-graph-api/reference/ig-user/media/
• Meta — Instagram Platform changelog: https://developers.facebook.com/documentation/instagram-platform/changelog
• Meta — Pages API: Posts (permissions incl. pages_manage_posts, publish_video): https://developers.facebook.com/documentation/pages-api/posts
• Meta — Video API overview (graph-video host deprecation): https://developers.facebook.com/docs/video-api/overview
• Phyllo — Instagram API guide 2026 (Instagram Login path, no Page required): https://www.getphyllo.com/post/instagram-api-guide
• postproxy.dev — Instagram Reels API publishing guide (Reels specs; vendor claim that Creator accounts can't publish Reels — unconfirmed against Meta docs): https://postproxy.dev/blog/instagram-reels-api-publishing-guide/
• Aurora Inbox — Meta Business Verification timelines (2026): https://www.aurorainbox.com/en/2026/05/14/what-is-meta-business-verification/
• Meta community forum — Business Verification blocking App Review: https://communityforums.atmeta.com/discussions/Questions_Discussions/business-verification-in-review-10-days-%E2%80%94-blocking-app-review-submission/1372323

TikTok

• TikTok for Developers — Content Posting API: Get Started (unaudited = private only; audit; scopes): https://developers.tiktok.com/doc/content-posting-api-get-started
• TikTok for Developers — Direct Post reference (6 req/min, privacy_level must match creator_info, brand_content_toggle, brand_organic_toggle, 2,200-char title): https://developers.tiktok.com/doc/content-posting-api-reference-direct-post
• Vorp Labs — TikTok Content Posting API summary (SELF_ONLY cap, test-user limits): https://vorplabs.com/agent-tools/tiktok-content-posting-api
• Outstand — TikTok Content Posting API: Direct Post, Audit, Limits: https://www.outstand.so/blog/tiktok-content-posting-api
• Bulkpublish — TikTok audit process (2–4 weeks, multiple rounds): https://www.bulkpublish.com/blog/tiktok-content-posting-api/

LinkedIn

• Microsoft Learn — Share on LinkedIn (self-serve, w_member_social, ugcPosts, 150 req/member/day): https://learn.microsoft.com/en-us/linkedin/consumer/integrations/self-serve/share-on-linkedin
• Microsoft Learn — Posts API (Community Management, 2026-05): https://learn.microsoft.com/en-us/linkedin/marketing/community-management/shares/posts-api?view=li-lms-2026-05
• Upload-Post — LinkedIn API access tiers 2026: https://www.upload-post.com/linkedin-api/

X

• Outstand — X API Pricing 2026 (pay-per-use; $0.015/post, $0.20 with link): https://www.outstand.so/blog/x-api-pricing
• postproxy.dev — X API pricing 2026 (tiers retired, Enterprise ≈ $42k): https://postproxy.dev/blog/x-api-pricing-2026/
• SocialNexis — X API tiers in 2026: https://socialnexis.com/guides/x-api-basic-enterprise-automation-rules

YouTube

• Google — YouTube Data API quota calculator: https://developers.google.com/youtube/v3/determine_quota_cost
• Google — YouTube Data API getting started (10,000 units/day default): https://developers.google.com/youtube/v3/getting-started
• Blotato — YouTube API pricing / videos.insert cost reduction Dec 2025: https://www.blotato.com/blog/youtube-api-pricing
• Phyllo — Is the YouTube API free in 2026: https://www.getphyllo.com/post/is-the-youtube-api-free-in-2026-quota-limits-costs-when-to-pay

Aggregators

• Ayrshare — Pricing (Premium $149, Launch $299, Business $599+ with per-profile tiers): https://www.ayrshare.com/pricing/
• Ayrshare — Business Plan overview (user profiles, white-label JWT linking page): https://www.ayrshare.com/docs/multiple-users/business-plan-overview
• Ayrshare — API integration for Business plan: https://www.ayrshare.com/docs/multiple-users/api-integration-business
• Zernio (formerly Late / getlate.dev) — Pricing (2 free, $6/$3/$1 per account tiers, X pass-through): https://zernio.com/pricing
• Post Bridge — Pricing ($39 Marketer / $99 Operator): https://www.post-bridge.com/pricing
• Buffer — Social media APIs that support multi-platform posting: https://buffer.com/resources/social-media-api-multi-platform-posting/
• postproxy.dev — Best social media scheduling APIs compared (2026): https://postproxy.dev/blog/best-social-media-scheduling-apis-compared/

Instagram link mechanics

• Inro — Instagram affiliate links guide 2026 (caption links not clickable; bio up to 5 links; Story sticker): https://www.inro.social/guides/instagram-affiliate-links
• Unilink — Why is my Instagram link not clickable (2026): https://www.unilink.us/blog/why-instagram-link-not-clickable
• InstantDM — Instagram comment links 2026: https://instantdm.com/blog/instagram-comment-links-2026-guide

FTC disclosure in video

• FTC — Warning letter (Grasso) stating disclosures must be in the video itself, superimposed text, both visual and audible: https://www.ftc.gov/system/files/ftc_gov/pdf/warning-letter-grasso.pdf
• Finnegan — Consequences for trade associations and influencers under the Endorsement Guides (Nov 2023): https://www.finnegan.com/en/insights/blogs/incontestable/not-so-sweet-consequences-for-trade-associations-and-influencers-who-didnt-follow-ftcs-endorsement-guides.html
• Track360 — FTC affiliate disclosure rules: operator compliance guide (2026): https://track360.io/blog/ftc-affiliate-disclosure-rules-operator-compliance-guide-2026

Mobile browser capture

• WebKit blog — MediaRecorder API (Safari records MP4/H.264/AAC; feature detection): https://webkit.org/blog/11353/mediarecorder-api/
• caniuse — MediaRecorder (iOS Safari 14.5 → 27.x supported; Chrome Android supported): https://caniuse.com/mediarecorder
• BuildWithMatija — iPhone Safari MediaRecorder support incl. WebM added in Safari 18.4: https://www.buildwithmatija.com/blog/iphone-safari-mediarecorder-audio-recording-transcription
• Chromium blink-dev — Intent to Ship: H26x codec support for MediaRecorder (MP4 muxing on Windows/macOS/Android where platform encoders exist): https://groups.google.com/a/chromium.org/g/blink-dev/c/YJ1QijNiHeM/m/3EhMjMVwAgAJ
• WebKit Bugzilla — Safari 26 MediaRecorder empty blobs with small timeslice: https://bugs.webkit.org/show_bug.cgi?id=301507
• WebKit Bugzilla — iOS 18.4 beta MediaRecorder video/mp4 regression: https://bugs.webkit.org/show_bug.cgi?id=290002
• addpipe — Duration in MP4 files produced by Chrome/Safari MediaRecorder (bad metadata): https://blog.addpipe.com/duration-in-mp4-files-produced-by-chrome-safari/
• addpipe — Dealing with huge MediaRecorder slices (Safari pause/resume quirks): https://blog.addpipe.com/dealing-with-huge-mediarecorder-slices/
• MDN — MediaRecorder.isTypeSupported(): https://developer.mozilla.org/en-US/docs/Web/API/MediaRecorder/isTypeSupported_static
• Argos Identity — WebView camera troubleshooting (in-app browsers block getUserMedia; recommend opening in a standard browser): https://developers.argosidentity.com/en/idcheck/getting-started/webview-camera-troubleshooting
• VIDA — WebView camera prerequisites: https://docs.vida.id/identity-stack/verify/liveness/integration/liveness-web-sdk/webview-camera-prerequisites
• Apple Developer Forums — camera in WKWebView / SFSafariViewController behaviour: https://developer.apple.com/forums/thread/670322
• caniuse — Screen Wake Lock API (iOS Safari 16.4+; PWA bug fixed in iOS 18.4): https://caniuse.com/wake-lock
• Progressier — Screen Wake Lock PWA capability notes: https://progressier.com/pwa-capabilities/screen-wake-lock
• caniuse — ScreenOrientation.lock (not supported in iOS Safari through 26.x): https://caniuse.com/mdn-api_screenorientation_lock
• Capgo — Capacitor camera-preview plugin (native preview with HTML overlays, video recording): https://capgo.app/plugins/camera-preview

Upload and transcode infrastructure

• Vercel — Functions limits (4.5 MB body, 300 s default / 800 s max / 1800 s beta, 4 GB memory, large functions): https://vercel.com/docs/functions/limitations
• Vercel — User-uploaded files with Sandbox and Blob (ffmpeg in Sandbox pattern): https://vercel.com/kb/guide/user-uploaded-files-vercel-sandbox-and-blob
• Supabase — Storage resumable uploads (TUS, 6 MB chunks, > 6 MB must use TUS, signed upload URLs): https://supabase.com/docs/guides/storage/uploads/resumable-uploads

Rendered from build-plan/02-collateral-publishing-teleprompter.md. Edit the source, not this page.