Manual export for Threads, Instagram, Facebook Pages, YouTube and TikTok

Issues #19-#23, offline slice. Verified 2026-10-08.

These five platforms need Meta app review, Google API compliance audits or a TikTok Content Posting API audit before an app can post for a user. The owner has not completed any of them, so ReachWeaver has no live adapter for any of the five. manual-export is the supported fallback. It turns one variant (and an optional media manifest) into a creator-ready package. The account holder then posts it themselves in the platform’s own app or website.

The package never sends, uploads, schedules or shares anything. Its status is always manual_export_not_published, and its wording never claims that a post exists. The module has no send path (SEND_PATH = None) and imports no network or publisher code.

Command

reachweaver manual-export --platform youtube --variant variant.json \
  [--media media/manifest.json] [--account-meta account.json] --out exports/youtube-tour [--json]

--platform: threads, instagram, facebook, youtube or tiktok for the sourced packages below, or any other manual_only id in the distribution registry for a generic package.
--out must be a new path. An existing file or directory, even an empty one, is refused. The directory is created with mode 0700, and each file with mode 0600 (O_EXCL, O_NOFOLLOW).
If preflight finds a blocking error, the command prints BLOCKED, writes nothing and exits 2.
Output files: manual-export.json (the full package), CHECKLIST.md, and the exact text. For YouTube that is title.txt and description.txt; for the others it is caption.txt. Text files contain the exact text, with no added newline.

From Python: platform_preflight(platform, variant, account_meta=None) -> (errors, warnings), build_export(platform, variant, media_manifest=None, account_meta=None) -> dict, render_markdown(package) and write_export(package, out_dir).

Generic package for every other manual-only destination

Private #44, item C2. Offline only. Every other destination that distribution_registry.json marks reachweaver_delivery: manual_only gets a generic package from the same command, for example --platform dev, x, product_hunt, reddit or hacker_news. The five platforms above keep their own sourced packages, byte for byte (tests/test_manual_package.py::test_cli_output_matches_golden_bytes). facebook_pages maps to --platform facebook. A destination that is not manual_only (Bluesky, Mastodon, Telegram, LinkedIn, GitHub, Snapcraft, Gitee) is refused. The code is reachweaver/manual_package.py: package_preflight(platform, variant, account_meta=None), build_package(...), render_markdown, render_notes and write_package(package, out_dir). It adds no delivery mode, no browser automation, no network access and no registry change.

The generic package knows only what the registries already record:

Label and help page. The label comes from the distribution registry. The official page is the registry’s scheduling.official_source, with its checked_on date, and only when the registry has one. That page may describe scheduling rather than the composer. Otherwise the package says that the registry has no official page.
Limits. No registry field records a text limit for these destinations, so the text limit is limit unknown — check the platform, except for the Chinese platforms below. Media rows come only from the media.py CONSTRAINTS table for the destination (DEV uses the devto table). A row without a value, and a destination without a table, also print limit unknown — check the platform (tests/test_manual_package.py::test_known_media_rows_come_only_from_media_constraints). The capability registry’s compliance_constraints for the destination are listed as registry notes with their source and date, for example X’s “Up to 4 photos, 1 GIF or 1 video per post”. ReachWeaver measures the text (characters, UTF-16 units, UTF-8 bytes) but checks it against no limit.
Media. With --media, the manifest is validated through media.validate_media_manifest, exactly as above. The package lists each asset whose targets include the destination id (DEV also accepts devto): path, alt text, caption files, detected size and the rights record from the manifest (rights_confirmed, license, attribution, provenance).
Checklist. Verify claims; sign in to the intended account; open the platform yourself; paste post.txt (and title.txt if the variant has platform_options.title) exactly; the text-limit line; attach each asset with its alt text and rights; confirm rights, consent and disclosure; and keep the native post URL as first-party evidence. ReachWeaver has no command that records a hand-made post (reconcile covers only queued sends), so the package records nothing. DEV packages add DEV’s rules: disclose AI assistance in the post, and no AI-generated comments (DEV AI guidelines, checked 2026-10-15). Hacker News adds “do not schedule or automate” (registry scheduling prohibited); Reddit adds “post by hand” (moderator_only).
Output. manual-export.json (kind reachweaver.manual-package.v1, status manual_export_not_published), CHECKLIST.md, and post.txt (plus title.txt, and topics.txt for a Chinese platform with topics). The same new-directory, 0700/0600 and no-overwrite rules apply, and a blocked package writes nothing.

Chinese platforms

Weibo, Xiaohongshu, Douyin, Bilibili, WeChat Official Accounts, Zhihu, WeChat Channels, Kuaishou, Toutiao, CSDN, Juejin and OSCHINA get a China-aware generic package (reachweaver/china.py). It adds a china block to manual-export.json and a “China” section to CHECKLIST.md:

Sourced limits. Each fact has a status: verified (an official help, rule or legal page; only these block), verified_client (the creator page’s own code, warning only), verified_api (a developer API reference, warning only) or unknown. Weibo’s 5,000-character post and 20 MB image limits block; Xiaohongshu’s 20-character title is measured the way its creator page counts it (ASCII is half a character).
Topics. platform_options.topics (topic words without #) is written to topics.txt in the platform’s syntax (Weibo #话题#, Xiaohongshu #话题, Bilibili comma-separated tags). The post text is never changed. On Weibo a #tag without its closing #, or a topic outside 4 to 32 characters, warns.
Declarations. ai_generated_label (CAC labelling measures Art. 10, in force 2025-09-01), advertising_label (SAMR Internet Advertising Measures Art. 9) and, on WeChat and Bilibili, originality. Each has selected: null; the CLI prints DECIDE YOURSELF for each.
Localization. Text without Chinese characters warns, and the checklist asks for a fluent reviewer and the Asia/Shanghai time zone.

Human-only Chinese spaces (Douban, Maimai, Qzone, Baidu Tieba) keep the notes form below. Full guide and sources: CHINA.md. Tests: tests/test_china.py.

Human-only destinations: write it yourself

Every destination the registry marks human_only (Reddit, Hacker News, Discord, WhatsApp communities, Facebook Groups, Indie Hackers, Lobsters, Stack Exchange and the rest) gets no agent-written post text. This follows the existing rules: Hacker News text is human-written only, ReachWeaver writes no Reddit text, and engagement drafts skip DEV and HN.

A variant text or platform_options.title for such a destination is a blocking error (tests/test_manual_package.py::test_agent_text_or_title_is_blocked_for_every_human_only_destination).
The variant may carry talking_points: at most 10 short notes of at most 200 characters each. That cap is ReachWeaver’s own, not a platform limit; it keeps notes from carrying a ready-made post.
The package has content.ready_to_paste: false, no text field and no post.txt. It writes NOTES.md instead: a private notes form that starts with “Write this yourself.” and has blanks for what you want to say, the community rules you checked, links, and the post URL after you post (tests/test_manual_package.py::test_written_package_has_notes_and_no_text_file).

reachweaver manual-export --platform hacker_news --variant notes.json --out exports/hn-notes
# notes.json: {"talking_points": ["Offline-first design", "No auto-posting"], "link_url": "https://example.org/docs"}

Batch of manual packages

Salvaged from draft PR #65, offline only. manual-export-batch prepares 1 to 25 packages in one run:

reachweaver manual-export-batch --file examples/manual-batch.example.json --out exports/regional-batch [--json]

{
  "schema_version": 1,
  "items": [
    {"platform": "xiaohongshu", "variant": {"platform": "xiaohongshu", "text": "Exact text"},
     "account_meta": {"handle": "@my-account"}, "at": "2026-11-02T19:00:00+08:00"},
    {"platform": "threads", "variant": {"platform": "threads", "text": "Exact text"}}
  ]
}

Same checks as one package. Each item goes through build_export (the five sourced platforms) or build_package (every other manual_only destination), so its preflight, limits, checklist and files are exactly what manual-export --platform produces. platform takes the same values as manual-export --platform. Media manifests are not taken in a batch; a media format only gets the “nothing validated” warning, so run manual-export --media for an item whose media you want checked.
Whole-batch preflight. Every item is checked before anything is written. One blocking error in any item prints BLOCKED: nothing written, exits 2 and creates no directory. Malformed batches (other top-level keys, 0 or more than 25 items, unknown item fields) and secret-like keys anywhere are refused.
Human-only destinations are refused. Reddit, Hacker News and every other human_only destination cannot be in a batch: prepare their notes form one at a time with manual-export, and write the post yourself. A repeated platform and text gets a warning.
SHA-256 of the exact text. --out must be a new path; it holds one folder per item (01-xiaohongshu, 02-threads, …) written by the single-package writer, plus batch-manifest.json. The manifest lists, for each item, its folder, platform, public account fields and the SHA-256 of every exact-text file (post.txt, title.txt, topics.txt, caption.txt, description.txt). After writing, each file is read back and compared with that hash; a mismatch stops the run. Folders are 0700 and files 0600, as above.
Advisory time. An item’s optional at must be ISO 8601 with an explicit offset. It is recorded as planned_local_time for your own planning; nothing is scheduled in ReachWeaver or on any platform.
Code: reachweaver/manual_batch.py (build_batch, write_batch, manifest), tests in tests/test_manual_batch.py. It has no send path (SEND_PATH = None), no network import and opens no browser. A batch is not permission for bulk posting.

To see which route a destination has before you add it to a batch, run reachweaver delivery-options --platform ID (reachweaver/delivery_options.py). It reuses the onboarding blocker audit, so its routes (official_api_adapter, manual_export, draft_export, native_manual, human_only, unsupported) match the troubleshooting table.

Variant file

{
  "platform": "youtube",
  "format": "video",
  "text": "Description or caption, exactly as it should appear.",
  "link_url": "https://example.org/docs",
  "platform_options": {"title": "ReachWeaver in 60 seconds", "tags": ["oss"], "shorts": false}
}

text is the caption. For YouTube it is the description.
format: Threads text|link|image|video|carousel (default text); Facebook text|link|image|video (default text); Instagram image|video|carousel (required); TikTok and YouTube video (required).
platform_options: YouTube title (required), tags, shorts. Facebook target_type (page).
Fields from the canonical PlatformVariant (id, account_id, campaign_id, locale, …) are accepted and ignored. Unknown fields produce a warning. Secret-like keys (token, password, session, …) block the export.
Choices the account holder must make are never taken from the variant. privacy_level, privacy_status, made_for_kids, contains_synthetic_media, is_aigc, disable_comment and the like are ignored with a warning, and the package leaves them unselected.

Account metadata (--account-meta) is public information only: account_id, handle, display_name, account_type (page, profile, group, business, creator, personal), and tiktok_api_audit (unaudited|audited|unknown, default unknown). Secret-like keys block the export.

Preflight rules

Errors block the export. Warnings need human judgment.

Platform	Blocking errors	Warnings
Threads	More than 500 characters, with emoji-like code points counted by their UTF-8 bytes; more than 5 links	More than one hashtag (only one topic tag per post); a tag longer than 50 characters
Instagram	Caption over 2,200 characters; more than 30 hashtags; more than 20 @ tags; text-only format	A URL in the caption (do not rely on it being tappable); 6 to 30 hashtags (the app is moving to at most five per post or reel)
Facebook	Target not declared, or declared as a profile, personal account, user, timeline or group; text over the conservative 4,500 cap (no official limit was verified)	—
YouTube	Missing title; title over 100 characters; description over 5,000 UTF-8 bytes; `<` or `>` in title or description; tags over 500 characters (commas count, and spaced tags count with quotes); non-video format	Empty description; description links (clickable only for eligible channels)
TikTok	Caption over 2,200 UTF-16 units; non-video format	A URL in the caption
All	Placeholder/TODO text, control characters (including DEL), bidirectional override/embedding/isolate characters and other invisible format characters (the same Trojan Source rule `policy.validate_post` applies to live posts; ZWJ, ZWNJ and LRM/RLM stay allowed), a non-public `link_url`, secret-like keys, platform mismatch	Text over ReachWeaver’s conservative `policy.py` cap (the official limit is what applies to a manual post)

Media

Media requirements are not duplicated here. With --media, the manifest is validated by reachweaver/media.py (validate_media_manifest): real headers, the sourced CONSTRAINTS table, alt text, SRT/WebVTT captions, rights, provenance and locales. Its errors and warnings are copied into the package, prefixed media:. The package then lists each asset that targets the platform (threads, facebook, instagram, tiktok, or youtube/youtube_shorts), with its path, alt text, caption files and detected dimensions and duration.

image, video and carousel formats need a matching asset that targets the platform. A carousel needs at least two.
media.py has constraint tables for instagram, tiktok and youtube_shorts. For Threads, Facebook and long-form YouTube it warns that no verified table exists.
Without --media, a media format only produces a warning that nothing was validated, plus a checklist step to run media-check.

Decisions only the account holder makes

Each one is listed in human_decisions with selected: null, required: true and its source.

TikTok: privacy_level (options PUBLIC_TO_EVERYONE, MUTUAL_FOLLOW_FRIENDS, FOLLOWER_OF_CREATOR, SELF_ONLY; only those your account offers apply). Also interactions (comments, Duet, Stitch), commercial_content and ai_generated_label. Public is never pre-selected. The note records the Content Posting API audit status from account metadata (default unknown). The audit applies only to API clients; ReachWeaver has none for TikTok.
YouTube: privacy_status (private, unlisted, public), made_for_kids (a legal declaration) and altered_or_synthetic_content.
Threads, Instagram and Facebook have no extra platform-assigned decisions in this slice. The checklist still covers account, rights, consent and disclosure.

Verified facts (2026-10-08)

Each fact is stored in manual_export.FACTS with its value, status, source, date and a short quote. unknown means the official page did not establish the fact. conflicting means official pages disagree; ReachWeaver never relies on such a fact.

Platform	Fact	Status	Official source
Threads	Text posts limited to 500 characters; emoji counted by UTF-8 bytes	verified	Threads posts
Threads	At most 5 links; one topic tag per post; tag 1-50 characters	verified	Threads posts
Threads	Permissions `threads_basic`, `threads_content_publish`; App Review before users without an app role can grant them	verified	Threads get started
Threads	Account-type requirements	unknown	Not stated on the fetched page; re-checked 2026-10-15, no official page limits publishing to a business or creator profile (see THREADS.md), but no receipt shows that every profile can publish
Instagram	Caption: 2,200 characters, 30 hashtags, 20 @ tags	verified	IG User media
Instagram	In the app: at most five hashtags per post or reel, rolled out gradually. Separate from the API’s 30, and only a warning	reported (Instagram quoted by Social Media Today, 2025-12-18; Instagram’s help page returned HTTP 400 on 2026-10-17)	Social Media Today
Instagram	API publishing needs an Instagram professional account; Business vs Creator split not stated	verified / unknown	Content publishing
Instagram	Instagram Login: `instagram_business_basic`, `instagram_business_content_publish`. Facebook Login: `instagram_basic`, `instagram_content_publish`, `pages_read_engagement`	verified	Content publishing
Instagram	Caption links tappable	unknown	No official statement found; the Help Center documents a profile website link
Facebook	Page posts: Page access token; tasks `CREATE_CONTENT`, `MANAGE`, `MODERATE`; `pages_manage_posts`, `pages_read_engagement`, `pages_manage_engagement`, `pages_read_user_engagement`, `publish_video` (video)	verified	Pages API posts
Facebook	Creating a post on a user (profile) feed is not possible on that endpoint	verified	User feed reference
Facebook	Groups API permissions (incl. `publish_to_groups`) removed 2024-04-22	verified	Graph API v19.0 changelog
Facebook	Page post text limit	unknown	Not stated; ReachWeaver’s conservative cap applies
YouTube	Title at most 100 characters; description at most 5,000 bytes; no `<` or `>`; tags at most 500 characters	verified	Videos resource
YouTube	`privacyStatus`: `private`, `unlisted`, `public`; the API defaults to public when omitted	verified	Videos resource
YouTube	Every creator must declare whether videos are made for kids (COPPA)	verified	YouTube Help 9528076
YouTube	Upload scopes: `youtube.upload`, `youtube`, `youtubepartner` or `youtube.force-ssl`	verified	videos.insert
YouTube	Uploads from unverified API projects created after 2020-07-28 locked to private	conflicting	The revision history entry of 2020-07-28 says yes, until an audit; the videos.insert page (last updated 2026-10-08) says such uploads “are not restricted to private viewing mode”. It concerns API uploads only, not Studio uploads
TikTok	Caption at most 2,200 UTF-16 units; `privacy_level` must match the creator’s `privacy_level_options`	verified	Direct Post reference
TikTok	“All content posted by unaudited clients will be restricted to private viewing mode”; an audit lifts it; `video.publish` scope needs approval	verified	Content Posting API get started
TikTok	No default privacy; interaction toggles unchecked; commercial-content toggle off; caption editable; no promotional watermarks	verified	Content sharing guidelines
TikTok	Draft-upload scope name; caption links tappable	unknown	Not on the fetched pages

Acceptance status (offline slice)

This slice delivers the manual fallback and the preflight that every issue names. It does not deliver the live adapters, mock API contract tests, receipt reconciliation or analytics, all of which need the owner’s app review or audit first.

Issue	Met by this slice	Still open
#19 Threads	Permissions and review requirements documented with date; manual-export fallback; live publishing stays disabled (registry `api_gated`, adapter unreachable). Met since (mocked): successful post (`tests/test_threads.py::test_happy_path_requests_and_receipt`), rejected permission (`tests/test_threads.py::test_contract_rejected_permission_blocks_before_claim_and_keeps_post_approved`), rate limit (`tests/test_threads.py::test_preflight_429_records_cooldown_and_keeps_post_approved`), expired token (`tests/test_threads.py::test_contract_expired_token_blocks_before_claim_and_keeps_post_approved`) and preview/receipt reconciliation without duplicates (`tests/test_threads.py::test_approved_preview_matches_the_receipt_and_is_published_exactly_once`, `tests/test_threads.py::test_reconciled_receipt_cannot_be_applied_twice`); see THREADS.md	Meta app with the Threads use case, the owner as Threads Tester, a long-lived token, a supervised first live receipt and the reviewed registry change
#20 Instagram	Professional-account prerequisite and API scopes documented; structured fallback for any account; fixtures cover bad media and missing alt text	Unsupported-consumer-account, missing-permission and asynchronous-processing API fixtures; Login-flow choice; account selection and receipts
#21 Facebook Pages	Personal profiles and groups rejected at preflight; Page permissions and Groups removal documented with sources and date	Reporting granted scopes from a real connection; mocked API contract tests; receipts and analytics
#22 YouTube	Documented manual YouTube Studio fallback (pinned by `tests/test_youtube_visibility.py::StudioFallbackDocTests`); the package never claims a video is public; privacy and made-for-kids left to the creator. Met since (mocked, synthetic fixtures): chunk upload (`tests/test_youtube.py::test_resumable_upload_in_chunks_then_receipt`), partial failure (`tests/test_youtube.py::test_partial_failures_after_the_first_byte_stop_and_never_resume`), audit gating (`tests/test_youtube.py::test_bundled_registry_keeps_youtube_gated`, `tests/test_youtube.py::test_unaudited_project_private_receipt_for_a_public_request_is_not_published`), duplicate avoidance (`tests/test_youtube.py::DuplicateAvoidanceTests`) and a mocked receipt feeding the analytics sampling report (`tests/test_youtube_visibility.py::test_an_adapter_receipt_flows_into_the_sampling_report_consistently`)	Google project verification and audit; a supervised private upload receipt from a real channel; channel-authorized YouTube Analytics API retrieval
#23 TikTok	Creator-assisted flow with traceable `manual_export_not_published` status; exact privacy choice left to the creator, public never pre-selected; audit requirements documented without bypasses. Met since (mocked): unaudited client (`tests/test_tiktok.py::test_unaudited_creator_offers_only_private_so_other_levels_are_blocked`), unapproved scope (`tests/test_tiktok_epic.py::test_init_401_names_the_missing_scope_and_nothing_is_uploaded`, `tests/test_tiktok_epic.py::test_an_api_scope_not_authorized_body_blocks_init_without_an_upload`), invalid creator options (`tests/test_tiktok.py::test_invalid_creator_options_block_before_the_claim`) and 429 (`tests/test_tiktok_epic.py::test_init_429_after_the_claim_is_certain_not_sent_with_a_cooldown_and_is_never_resent`, `tests/test_tiktok_epic.py::test_creator_info_429_records_a_cooldown_and_keeps_the_post_approved`)	App registration, Content Posting API scopes and audit; a supervised live inbox receipt

Update (2026-10-09): later slices added a mock-tested adapter with mocked API contract tests for each of #19 to #23 (owner-approved 2026-10-09; see THREADS.md, INSTAGRAM.md, FACEBOOK.md, YOUTUBE.md and TIKTOK.md). Every one stays disabled in the capability registry until a supervised live receipt, so manual export is still the publishing path; app review, real receipts and analytics remain open.

Any future adapter needs its own security review and must keep human approval, dry run, the emergency stop, credential isolation and ambiguous-send reconciliation unchanged.

After you post a package by hand

manual-export cannot see what you do with the package. Once you have posted it yourself, record it from its link with reachweaver posted LINK --likes N (see below), or with reachweaver mark-shared --platform P --video FILE --text "..." --url LINK to also remember the video file, so the repeat guard warns you if the same video or wording is about to go out again on that platform.

Measure posts you shared by hand

The short way is one command with the link to your post and the numbers you see in the app:

reachweaver posted https://bsky.app/profile/you.bsky.social/post/3kabc --likes 12 --replies 3 --reposts 2
reachweaver posted https://www.instagram.com/p/ABC123/ --likes 40 --views 900 --at yesterday --text "Behind the scenes"

Network. Taken from the link’s address: bsky.app (Bluesky), instagram.com, linkedin.com, tiktok.com, youtube.com and youtu.be, threads.net and threads.com, x.com and twitter.com, facebook.com, t.me (Telegram), dev.to, reddit.com, pinterest.com, hashnode.com and hashnode.dev, and the server of a Mastodon account in your accounts file. For any other address ReachWeaver asks you for --platform (with the list of names); it never guesses and never opens the link.
Account. The account in your accounts file whose handle the link names, else your only account on that network. If you have several and the link does not say which, it asks for --account and lists them. With none, the post is recorded as posted by you without an account, and the output says so.
Numbers. --likes, --replies (or --comments, the same option), --reposts and --views are stored as you typed them (source entered-by-hand, observed now). A number you leave out stays unknown, never 0. A wrong value stops the command before anything is written.
When. --at (yesterday, 2026-10-19, '2026-10-19 18:30', …). Without it, a terminal asks When did you post it? [today] (e.g. yesterday, 2026-10-19) and Enter means today; a script gets now, and the output says posted today (DATE) — if not, run again with --at yesterday.
Update later. Run posted with the same link and new numbers: it adds them to the post you already recorded (“Updated numbers for … on Bluesky”) instead of recording it twice. This also works for a post ReachWeaver sent.
Correct it. Run posted with the same link and --at, --text or --account to correct a post you recorded by hand (“Corrected: posted 2026-10-20 (was 2026-10-21)”); the 7-day reminder then follows the corrected date. The record stays the same never-sendable published_manual row (its status and empty approval stay as they are), the linked mark-shared entry gets the same values, and a manual_publish_corrected event records the old and new time and account (never the text). A post ReachWeaver sent is never changed this way: its time, text and account are the record of what was sent, so only its numbers can be added. A time in the future, or after numbers you already entered for the post, is refused and nothing changes.
Text. --text is optional. Without it, reports name the post by its network and day (Instagram post of 18 Oct), and the repeat guard has no wording to compare later drafts with; add --text (or use mark-shared with --file, --image or --video) when you want that check.

posted makes the same published_manual record as mark-shared --url below, with the same limits: it can never be approved, scheduled or sent.

The longer way, mark-shared, also records images, videos and links for the repeat guard. Add the link of the post you published (and, if you like, the account and the time you posted it):

reachweaver mark-shared --platform mastodon --account my-masto \
  --url https://social.example.org/@me/109876543210 \
  --at 2026-10-20T09:00:00+02:00 --file caption.txt

With --url, mark-shared does two things and sends nothing:

It records the material for the repeat guard, exactly as without --url (without --account it counts for every account on that platform).
It creates a measurable record: a post with status published_manual (“posted by hand”). It prints its id. --at is the time you posted it (default: now; a time more than 10 minutes in the future is refused). This time is what report --since and insights use.

Then record its results as for any published post:

reachweaver metrics POST_ID --likes N --replies N --reposts N (recorded now, source entered-by-hand; add --observed-at yesterday or --source native-export when that is true), or import-metrics --csv FILE with that post_id, from the platform’s own analytics.
metrics-fetch --online reads public counters for a Mastodon link on your configured server and handle, and for a Bluesky link of the form https://bsky.app/profile/did:.../post/... (the post must be by your configured handle). For a Bluesky link with a handle instead of a did:, or for any other platform, use metrics or import-metrics; metrics-fetch lists the record as posted_by_hand_without_platform_id.

report, insights, journal and the dashboard’s Posted and Results pages include these posts and label them “posted by hand”. journal --no-by-hand leaves them out.

What it cannot do:

It cannot be approved, scheduled, sent, retried or reconciled. Every send path refuses it or does not see it.
It is your own statement. ReachWeaver does not check that the post exists. Record the same link twice and the second call is refused.
It does not change older posts. A record made without --url stays a repeat-guard entry only; record it again with --url to make it measurable.

The first time you record a post this way, ReachWeaver upgrades an older database once (it rebuilds the posts table so it accepts the new status, keeping every row and approval). Back up state/ first if you like; nothing else changes.