Solvent — Internal · Draft v0.1 · 10 Jul 2026

Content Engine

System map & technical breakdown — the version with all the tool names in it.
Companion pieces: Document 1 (internal proposal, .docx) · public deck (deck-public/ — the one PE sees).

01 — Full system map

Bold boxes = we own/operate. Dashed = feedback loops. The public deck shows the simplified version of this — EditAssist and TryPost are internal names only.

02 — Intake & archive

Get everything. Lose nothing. Tag it once.

Shared Drive

One shared Google Drive per source (PE team · founders · shoot days). Founders upload from vlog cameras/phones with zero process — friction kills founder footage, so the pipeline absorbs mess.

Sync agent

Watches Drive folders → mirrors hot files to NAS (rclone-class job on SolvServer, checksummed). Drive stays the friendly front door; NAS is the archive of record.

External ingest

Downloader agent for founder appearances elsewhere — podcasts, panel videos, other channels’ interviews (yt-dlp + queue). Everything joins the same library with source metadata.

Tag taxonomy

soundbite · hero shot · good still · expert moment + person, company, cowork/event, theme pillar. EditAssist auto-tags on ingest; humans correct, never start from zero.

Already live

Verified on Drive (10 Jul): shared “Project Europe” + “ProjectEuropeY1” folders exist, shoot-notes docs with timestamped soundbites (London ’26: Anthropic, CoreWeave, Zayne’s security talk) — the tag taxonomy is being practised by hand. The engine automates what we already do.

Year 1 corpus

Already on the NAS. M0 job: EditAssist analyses all Year 1 timelines + delivered videos → style/pacing/look model that keeps the whole fleet consistent.

Governance

Local-first editing — footage never leaves our machines. That's the GDPR answer and the client-trust answer in one.

03 — Edit layer

Two lanes: the fleet and the heroes.

Fleet lane

EditAssist drives Resolve/Premiere/AE from natural language against brand rules + motion templates + SFX library. Output: the daily 9:16 fleet, cut down the length ladder below.

Length system

Hook layer — 10–15s. Discovery, volume, reactive posts, soundbites, hook A/B. TikTok/Reels/Shorts/X, and the lead-in everywhere (LinkedIn avg watch: 13–15s regardless of length). Audience: the Gen Z scroll.
Depth layer — 30–90s. Story cuts, technical advice, GTM & expert moments — where 60–120s posts pull 3–4× the views of sub-15s, and where sponsor/investor conversion happens (LinkedIn talk cuts). Audience: founders deciding to apply, investors deciding to care.
Long-form. YouTube/LinkedIn — cowork films, founder/sponsor talks, Year 2 doc. Audience: applications marinating (YC 68%).
One asset, one ladder: film → cutdowns → depth clips → hooks. Every rung re-usable.

Gap fill

Higgsfield MCP for generative visuals, Suno MCP for music beds — used to fill gaps, never to replace real founder footage.

Hooks

Payoff line first. 20VC front-loads the best 12 seconds into the first 36; IG now natively measures 3-second skip rate. Hook discipline is Jay's edit brief, systematised.

Hero lane

Hand-edited, full creative attention: cowork films, company showcases, Year 2 documentary. Narrative development, per-channel conforms, real thumbnails per channel (DOAC/MrBeast lesson: packaging is half the game).

Cutdown rule

Every hero timeline is analysed → 5–10 cutdowns minimum per cowork/big event/shoot day. Long-form is a short-form factory; value per shoot day compounds.

Cinema thread

Verified via inbox (10 Jul): Year 1 doc premiered Wed 6 May ’26 at the 250-seat “Art of Storytelling” screening (the Curzon Soho booking, their “best storytellers” night). New PE Soho space confirmed — house-warming at 75 Dean St, 6 Jul (Luma, host Jade Yarrow). Year 2 doc premieres on a Soho screen. Kitty’s own doc brief (Drive, Apr ’26) sets the bar: “a short, cult documentary that leaves people feeling uplifted and inspired by the potential of young founders in Europe” — cold open, 3–4 founders, real tension. The hero lane is built to that bar.

04 — Scheduling & hosting

TryPost, self-hosted. As the brief specified.

Confirmed

TryPost (trypost.it) — James's original call, verified 10 Jul: open source (github.com/trypostit/trypost), hosted app at app.trypost.it, $12/workspace, 500+ creators/agencies. The overnight pass wrongly swapped it for Postiz — reverted; Postiz stays as the evaluated fallback (AGPL, 30.4K stars).

Deploy

DECIDED (James, 10 Jul): Docker Compose on a dedicated aux VPS, EU region (~€10–30/mo) — isolated from SolvServer. Hosted fallback ($29–99/mo) if ops ever bite.

Why it fits

All 5 target platforms natively (plus Threads/Bluesky/etc.) · built-in analytics · approvals + media library + automations · REST API · official MCP server (docs.trypost.it) → Solvent + PE can query and move the calendar conversationally. Open source; calendar data stays ours.

Calendar seed

Milestones · cohort announcements · country national days per founder · public holidays · anniversaries · reactive slots weekly (Earth Day → Lavoit green-energy clip). Reactive turnaround target: hours, not days — templates + gap-fill make that real.

Delivery

Hot clips on Cloudflare CDN/R2; share links via Drive; finals archived to NAS. Dashboard hosted alongside TryPost.

Posting times

Seeded from Sprout's UK dataset (230M+ UK engagements) → re-derived monthly from our own numbers. Seed values: X Wed 4–10pm · TikTok Wed 3–11pm · LinkedIn Thu 11am–7pm · IG Sun 6–7pm · Thursday best day, Saturday worst.

05 — Measure · test · funnel

Every experiment gets one KPI.
Every clip can carry a funnel.

Dashboard

Live to PE at all times: per-channel growth, per-clip performance, funnel conversions. Michael presents the monthly read to Kitty & Jade; decisions go straight into calendar + edit briefs.

Testing policy

Trial Reels (Meta's official organic A/B — compare trials to trials) + YouTube native thumbnail/title tests (max 3 variants). No duplicate/mirrored accounts — zero evidence it works and IG's Apr 2026 originality policy penalises reposts. "Mirrored" = native per-platform re-edits + alternate hooks on the same asset. Strip watermarks always.

KPI rule

DOAC discipline: no experiment runs without a pre-declared hypothesis and a single KPI (watch time vs follows vs applications). Their cautionary result: doubling clip cadence 1→2/day reduced per-clip views; total flat. Test cadence, don't assume it.

Funnels

UTM discipline on every tracked link → applications, hackathon entries, event sign-ups, mailing list. Plumbed to PE's CRM + Luma. Plus the YC move: "how did you hear about us?" on every application from day one — content-attributed applications are the core metric, not views.

Access asks (PE)

Native analytics on their accounts (X/LI/TT/IG/YT) · CRM/Luma API keys · application-form question added. Goes in onboarding, M0.

06 — Numbers that matter (internal read)

The gap we're selling into

Channel	PE	20VC	YC
X	5,478	522K (Harry)	1.62M
LinkedIn	~26.5K	~270K (Harry) · 32.7K (co. page)*	~1.73M
YouTube	34	214K	2.3M
TikTok	—	268K	70K
Instagram	—	157K	~671K

*20VC LinkedIn resolved 10 Jul 2026: company page = 32,670 (live scrape of linkedin.com/company/20vc og:description); the old 8,924 snapshot matches no 20VC entity — treat as bogus. All counts 10 Jul 2026.

Keep on hand

68%of accepted YC founders watched YC YouTube pre-application; video = #2 source of accepted applicants

23%of 20VC's net-new podcast subs from TikTok after going in-house; 80+ deals/month sourced there

~3×fewer views for sub-15s vs 60–120s clips (Socialinsider: IG ~2.8–3.1×, TikTok ~3.0–3.15×) → run both length layers

+450%engagement/post for creators consistent 20+ of 26 weeks (Buffer) — consistency beats bursts

−31%TikTok avg per-post views YoY (volume +72%) — saturation favours hooks + reuse over flooding

30–200×Kitty's Agnessa post vs her typical engagement — founder moments are the format

PE's own wordssite promises founders "world class media and distribution across our platforms"; their Apr '26 post: builders "doing a crap job of telling the story"; Kitty on the cohort TikTok: "crazy spiking application from people we would never ever have met" — the pitch quotes all three

07 — Proof the pipeline already ran

London ’26 went through the engine in June.

Transcripts

$ sqlite3 edit_assist.db

> SELECT COUNT(*) FROM transcript_segments;

3,991

> SELECT COUNT(*) FROM clips;

374 — 25 hrs of rushes, word-level timing

Auto rough cuts (EDL/FCPXML, generated)

pe_london_2026___top_5_rough_cut.edl

30s_best_moments_roughcut_v1.edl

founders_entrepreneurship_60_90s_v4__punchy.edl

pe_london_2026___interview_moments_60s.edl

anthropic_interview_rough_cut___tighter.fcpxml

Interviews_30s_Roughcut.fcpxml

+ 28 more

Entity sweep (agent notes, 77)

cluster: CoreWeave / Ben Richardson —

"hero quote: essential cloud for AI …

AWS for AI" @ C1541.MP4

cluster: Anthropic fireside — select

5.2s @ 01:21:22 → rough cut, tighter

3,991transcript segments, word-level karaoke timing — captions come straight from the DB

34auto-generated rough cuts (EDL + FCPXML), openable in Resolve

77agent analysis notes: entity clusters, hero quotes, select reasoning

5example frames in the public deck were pulled, graded and captioned by this pipeline today

The examples in the public deck aren’t a promise — they’re output. Same pipeline, pointed at Year 2.

Receipts from the EditAssist project DBs on JRSSD, 10 Jul 2026. Internal only.

08 — Cautions (why our editorial line is the safe one)

Hype without substance backfires.

Cluely

10–13M-view launch → $15M a16z → 700 paid clippers pushing ~800K views/day. Then: Mar 2026 public admission the $7M ARR claim was fabricated (~$5.2M real). Attention bought capital; the lie bought a permanent asterisk. (Also: the brief's "Cooley" = Cluely — auto-caption mishearing, evidence on file.)

Icon

$12M domain as a content stunt, 10M+ impressions, Founders Fund money — shut down within a year when the product missed.

Friend

26.1M-view launch, $950K subway takeover → ~4,000 devices. Views ≠ revenue. (His pivot: a self-financed documentary — stunt graduating into media property, same as Bryan Johnson → Netflix.)

Our line PE's genre is documentary-truth, not rage-bait: real founders, real numbers, chaos only when it's genuinely theirs. "Europe is the character" isn't just positioning — it's the risk model. It's also why the public deck carries one soft caution line instead of these case studies.

Other watch-items: attribution is genuinely hard (only 37% of marketers find social→outcome linkage easy — hence UTM + application question from day one) · X hides view counts from tools (need native analytics access) · TryPost is a young OSS project (mitigation: open source + exportable data; $12/mo hosted tier; Postiz evaluated as drop-in fallback).

09 — Ops & rollout

Three lanes, one loop, four phases.

James

Technical infra & build-out: intake/sync, NAS, EditAssist pipeline, TryPost + MCP hosting, dashboard, UTM/funnel plumbing

Jay

Edits & creative direction; each channel's unique strategy and copy; hero edits; hook discipline

Michael

Shoot production, footage procurement, monthly performance presentation to Kitty & Jade

All three

Recap films + Year 2 documentary, creative collab

M0Foundations — brand rules locked · MG templates built · EditAssist Year-1 analysis · TryPost live · calendar seeded · access asks to PE

M1Engine live — intake syncing · first fleet ships · daily posting starts · baselines captured · trial posts begin

M2Funnels + dashboard — tracked links everywhere · dashboard live to PE · CRM/Luma verified · first monthly presentation

M3+Steady state — monthly tune to sweet spot · cowork films + 5–10 cutdowns each · Year 2 doc thread → Soho cinema premiere

Costs: infra mostly owned; incremental = VPS (~€10–30/mo) + CDN by volume + Higgsfield/Suno credits. Commercials handled separately — not in these docs (James, 10 Jul). Table in Document 1 §9.

10 — Open questions for the morning

Decide, then we ship.

01Mount the Project Europe Content Engine folder — it wasn't attached; I worked from the brief text

02TryPost ✓ (reverted to the brief’s original pick, 10 Jul) + hosting decided ✓ — self-host on aux VPS, EU region (~€10–30/mo); Postiz = fallback

03Zuck meme dropped ✓ (James, 10 Jul) — public slide 07 now runs Kitty’s application-spike quote + reactive/signalling blocks instead

04Public deck carries one soft caution line — want the harder Cluely/Icon receipts in it, or keep those internal?

0520VC LinkedIn number needs one manual logged-in check before we quote it anywhere

06Zayne Zhang · Suna Health confirmed ✓ (James, 10 Jul) — adopted everywhere, incl. public deck

07Retainer/budget numbers for Document 1 §9

08Deploy: run the Claude Code handoff (CLAUDE-CODE-HANDOFF.md) when you've reviewed — it finishes assets, QAs, and ships deck-public to Cloudflare Pages at pitch.solvent.studio (decided 10 Jul). deck-internal never deploys.

09RESOLVED ✓ — Gmail + Drive connected (10 Jul): Year 1 premiere = 6 May, 250-seat Art of Storytelling screening; house-warming = 75 Dean St, 6 Jul (Jade, Luma); Drive intake folders already live

11 — Pricing calculator (sales only)

Set the retainer in the room.

Move the numbers live with the client. defaults — adjust in the room

Coworks filmed / year × cost per shoot (crew + travel)

× £

Editors × monthly cost each

× £

Creative direction, days / month × day rate

× £

Equipment & infrastructure amortised monthly — cameras, drives, rigs, founder vlog kits

£/mo

Documentary reserve include

£/mo

Margin 25%

25%

Monthly cost base £0

+ margin £0

= Monthly retainer £0

Annual — £0

Per cowork film

£0

Per clip (40 / mo)

£0

12-month engagement value

month 0 / 12

£0