Recoup Internal — Marketing (the daily run)

The loop for a day of marketing on one of our own accounts. This skill is the orchestrator: it owns deciding what to make and proving it is grounded in evidence. It delegates production to the video-format skills and publishing to recoup-internal-social-ship-posts.

The goal — read this before anything else

Convert a free content viewer into a paying Recoup subscriber — and know which post did it.

That sentence is the job; any step that stops serving it should be changed rather than performed. Likes, views and followers are leading indicators, not the result. The destination is the other half of every conversion; you own the path from first frame to subscription. If the result cannot be measured, say so in writing, every run. You are allowed to not post: when the bottleneck is downstream, saying so is the day's highest-value marketing act. Steps 2a and 5 exist because on 2026-07-30 a full day of craft shipped into a page with zero trials in eight weeks (references/conversion.md). Do the steps in order.

Step 0 — Identify the account

Read ACCOUNT.md in the account workspace for the artist_account_id and the connected socials (they live on the artist account, not the token's default identity). The personal sweetman and the Recoup official accounts have their own IDs; confirm which one today's post is for.

Step 1 — Read the workspace (never skim)

Also read references/video-pipeline.md and references/voice.md before generating audio. Output of this step: name the arc and the character today's post serves. A post that serves no arc is an ad. No NARRATIVE.md in the workspace: say so rather than inventing a story.

Step 2a — Read the funnel BEFORE the feed

Open every run with one number: trials/subscriptions started since the last run. Pull it from recoup-internal-sales / recoup-internal-funnel-valuation-pipeline (Privy signups, Stripe trials and cards, credits, Attio stage), not from the social scrape, which cannot see it. Then state, in one line each:

• trials/subscriptions since the last run;
• whether the destination for today's likely CTA converts at all (gate in step 5);
• whether conversion is readable: visit capture is live (verified 2026-08-18; Vercel Web Analytics records utm_* on both properties). Run references/conversion.md → The attributed-visits pull. The visit→signup join stays unreadable (chat#1889 row 29); declare that gap in writing.

If trials have been flat at zero across several runs, that is the finding, and it outranks today's post. See the funnel guardrail below.

Step 2 — Scrape the account's own socials

Do not skip this and do not substitute the posts-log for it. The log is what we thought happened at the last re-pull; the scrape is what is true now (2026-07-28: a full video was built off the log alone, and the weak recent hooks a scrape would have shown surfaced only afterwards). Scrape and poll per references/learn-from-socials.md → The scrape.

Then rank recent posts by engagement and name the differences between the top performer and the flatliners: hook type, format, collaborator tagged, AI disclosed, length, posting slot. Those differences are today's brief. Report deltas, not all-time totals. A platform whose numbers have structurally collapsed outranks today's post (see guardrails).

Step 3 — Pick today's topic

Three legitimate sources, in preference order:

1. Continue a live arc. Best default; the canon asks recurring segments to end on a cliffhanger.
2. Pull a logged content idea from the workspace or NARRATIVE.md's upcoming canon dates.
3. Draft a new idea that advances an arc and is worth adding to the canon. A new storyline needs its own arc-proposal doc and an owner ruling before it is canon.

Gate zero, before every other gate: name the KIND of piece (launch, incident, or report) from the source, not from what scored last week. Our strongest register is confession, so a shipped feature gets pulled toward a story about our own mistake; that cost a full rebuild on 2026-08-12. Structures and the beat-budget test: references/topic-selection.md → Gate zero.

Gate the pick on all of these before building (details in references/topic-selection.md):

• Why named, theirs first, then ours. One sentence on what the person in the piece gets, one on what we get, both at the top of the plan doc. If you cannot write the first sentence, do not build the piece. Where it publishes an artist's real numbers, state the risk to them in the same breath, then let them decide. Ours is allowed to be commercial; say so plainly.
• Arc + character named.
• Destination named. Say where a convinced viewer goes before you build. If the honest answer is "the homepage," this is a trust/awareness beat, not a conversion one; both are legitimate, conflating them is not.
• Collaborator named, or explicitly none. Featuring the artist as a collaborator and inviting them to co-post is the largest measured lever we have: the 2026-07-22 collab reel took 45 likes against 1–3 for every non-collab reel after it. "None" is allowed and costs reach.
• Numbers. Any figure attributed to a real artist must be measured and verified; never publish a number you cannot audit, including about ourselves. Measured-vs-estimated disclosure is one later beat, never part of the opening figure (references/hooks.md → One flat figure).
• NO AI disclosure in body copy. Owner ruling 2026-07-28: never spend caption, tweet, description or on-screen text saying the visuals are AI generated; use the platform's own AI-content toggle at upload where labelling is wanted. This reverses the 2026-07-06 A/B guidance; if an older workspace doc still says "always disclose", this ruling wins.
• One idea per day. Do not ship two competing assets into the same slot.

Step 4 — Build the asset

Route the format first (references/video-formats.md): pick a style from VIDEO-STYLES.md and clone its reference project's video/ dir; never start from scratch. Spoken-word ad / talking character: a lip-synced clip from a generated still, see references/video-pipeline.md → Lip-synced VO over a still. Artist music video is the separate recoup-music-video skill.

Build in this order, for EVERY video. Each stage is approved before the next begins, because each one seeds the next and a mistake compounds.

1. Concept lock. One reviewed paragraph: premise, character, structure. Not changed once generation starts (09-02: the concept moved three times mid-build and orphaned a complete 21-still pass; "$2.85 for garbage"). An imprecise vision is not fixable downstream.
2. Plan doc SCRIPT.md. Why theirs then ours (risk to a real person stated alongside), arc, character, destination, collaborator, and the hook archetype named (specific number, contrarian claim, list tease; if you cannot name one, rewrite it). Never open by qualifying the audience. Cold-feed premise test: the video states its premise in its first beats, VO or on-screen, because a cut that relies on the caption has no story for most viewers (references/hooks.md). Name the final act before the hook: every number in the last three beats belongs to something the film already introduced; for a shipped feature that is the contract and a runnable CTA, not a benchmark (references/video-formats.md → The third act). A multi-item piece gets its scaffolding here (references/structure.md). Reviewed before a credit is spent.
3. Hook line generated and MEASURED from audio (08-11): the spoken hook lands in 3s read from the generated audio's duration, not estimated from the word count (references/hooks.md).
4. Voice chosen by checking the MODEL first (references/voice.md); most "robotic VO" is a stale model_id in a cloned script (08-06). Bake off on the film's lines at equal loudness, record the winner; a recurring character keeps its canonical voice in cast/<character>/. Then audio approved before compositing: one concatenated review file, owner listens (08-07: this caught a mechanical voice before the timeline existed; a voice change later is a timeline rebuild).
5. Character sheets.
6. Location plates, only when two or more shots share one.
7. Stills, then the contact-sheet gate.
8. Motion.
9. Composite. Read the plan doc's Screen column back against the composition row by row (08-12).
10. Snapshot with npx hyperframes@0.7.5 snapshot --at <times> . (pin the version; ~15s) before any render. One run burned seven full renders on findings visible in a still.
11. Render, in the background.
12. Frames read out of the MP4 at caption times. The right duration is not evidence it looks right, and inspect samples fixed points so it misses collisions between them.

Two audits ride along. Audit the SCREEN, not only the script: run the second half of the references/hooks.md checklist over every panel; it scopes to claims and measurements (a 1/5 counter, ordinals, dates, prices shown as UI and cited standards are exempt): each figure in the VO, with its unit, from a customer rather than a test harness? A run that audited the script twice and the panels never shipped a lab number to camera. Measure audio, do not blindly normalize: prefer linear gain, reserve dynamic normalization for genuinely quiet sources.

Step 5 — Publish

Run the executable pre-flight first; it exits non-zero, so do not publish past it. Then publish, then verify the published text against the config with a strict equality check, not by eye (2026-08-12 empty-tweet-to-the-wrong-account; 2026-07-28 literal <YT link> placeholder caught). references/publish-verify.md → Gate zero and the per-platform traps.

Load the destination before you publish. Open the CTA URL today: 200 and what loads pays off the post's promise (2026-07-30: every CTA in a four-platform slate resolved fine and pointed at a homepage with zero trials in eight weeks). Declare at publish time: visits readable by utm_campaign; signup join unreadable (row 29). Write it into posts-log.md then, not at the re-pull.

The CTA is a direct tagged link, never a comment-gate: tried twice, zero leads both times. Tag per references/conversion.md.

LinkedIn gets an IMAGE, not video, because images outperform video there (owner ruling 2026-07-28, from personal-post data). A performance decision, not a connector workaround, so do not hand-upload a video "for consistency" either. The image depicts what the copy's opening line describes.

• Publish + measure: recoup-internal-social-ship-posts owns per-platform copy, the connector mechanics, and the ~48h re-pull.
• Log it: one row per post in posts-log.md with the arc and character, the why (theirs and ours, as in the plan doc), and the sign-off line used.
• Capture what you learned in the workspace doc that owns it (HOOKS.md, VIDEO-STYLES.md, NARRATIVE.md), not just the chat.

Step 6 — Did it convert?

A post's job is to move a passive viewer to a signup.

1. Every CTA link is tagged at publish time. Convention, capture chain and per-platform caveats (Instagram cannot carry a per-post link in the caption): references/conversion.md. Signup join tracked as row 29 of chat#1889.
2. At the ~48h re-pull, read conversions alongside engagement: attributed visits, then signups via recoup-internal-sales / recoup-internal-funnel-valuation-pipeline.
3. Log both in posts-log.md, even when the answer is zero. A written zero is a finding.
4. Feed it back into step 2. Next run ranks by what converted; where likes disagree, conversion wins.

Guardrails

• No em dashes in published copy (captions, tweets, titles, descriptions, first comments, on-screen text; internal docs exempt). Rewrite the sentence rather than substituting a comma.
• No third-party provider or model names on anything published (owner ruling 09-04): no fal, MiniMax, Seedance, ElevenLabs, Grok, OmniHuman on cards, captions, descriptions or on-screen text; describe by function; label costs "as billed" or "at list price".
• No why, no build. Permission is not a reason: "they consented" answers whether we may, never whether we should.
• Nothing publishes without explicit owner go-ahead. "My goal is to post X" is a draft instruction, not authorization. Build it, then ask.
• Never re-trigger a live customer's task to test a change. Every run emails the customer.
• Real numbers or nothing, including about ourselves; publish the counts you can audit.
• Our own accounts only. Customer- or artist-facing content goes through recoup-content-*.
• A structural distribution collapse STOPS the run. If step 2 shows a platform down an order of magnitude, report it and get a decision before building for it; diagnose against a like-for-like baseline and check tagged clicks before writing a platform off (learn-from-socials.md); yt has sent the most tagged visitors from the lowest view counts. (07-28: flagged twice, built past twice.)
• A BROKEN FUNNEL STOPS THE RUN, one layer down. A dead destination burns first-time visitors and hides behind healthy engagement (2026-07-30: /pricing was the #2 marketing page with zero trials since 06-06, and a full slate shipped into it). Fix the path, do not post.
• A build that yields two assets queues the second one as tomorrow's episode in posts-log.md or NARRATIVE.md.

References

• references/learn-from-socials.md — the scrape, per-platform metric shapes, reading gotchas, the like-for-like baseline
• references/topic-selection.md — choosing the day's beat, and the gates it must clear
• references/video-formats.md — format routing table and the third act of a shipped-feature film
• references/hooks.md — hook archetypes, retention constraints, the pre-render checklist
• references/conversion.md — tagged CTA links, the capture chain, the attributed-visits pull
• references/publish-verify.md — the pre-publish gate, post-publish verification, platform traps
• references/video-pipeline.md — the shared build recipe: scaffold, rebuilt UI panels, captions, lip-synced VO over a still, render, frame QC
• references/voice.md — model vs voice vs delivery, the bake-off, audio tags, loudness
• references/seedance.md — Seedance 2.5 generated plates: constraints, the fal call, prompt doctrine, run log
• references/seedance-examples.md — every Seedance 2.5 example prompt we have, verbatim and sourced
• references/structure.md — holding a multi-item piece together after the hook
• references/character-swap-edit.md — recreating a trend by swapping the cast of a real performance clip (reAPI Seedance 2.5 edit): casting limits, the 3-reference rule, hosting, QC; script scripts/reapi-edit.sh
• artist music video lives in the separate recoup-music-video skill. Route there; do not rebuild the pipeline here.