Everything the machine is doing, and where it is stuck.

# Council ruling on the clipping venture — 2026-09-18
Five advisors answered in isolation. Each then blind-ranked the other four. A chairman read everything and made a call. Below: all five answers as given, the ruling, and what I (Clara) verified on disk before accepting it.
## The call
Not as written. The $23.7k base case, the 30-day clipper sprint and the 7-account persona farm are dead: I verified on disk that the prospect list is 35 NO_GAP out of 47, the projection has zero churn and $400/client of editor cost the engine supposedly replaced, outreach.py still reads the unverified CSV, every sender defaults to the personal Gmail, Coinpoker is eligible on today's board, accounts.json is empty and jobs/ holds four clips, not 21. What survives is a small Layer B business (3 to 5 retainers, $3k to $6k/mo net, about 5 hrs/week) and it only earns the right to exist by passing a 14-day, permission-first demand test that costs Ralph about one hour total. Fewer than two yeses and one creator willing to pay $1,000 by day 14 means the engine becomes an internal content tool for Cali Life, RWP and Helm + Oar, and the venture closes.
**Chairman confidence:** 74/100
## Strongest dissent, quoted
> CONTRARIAN: "This is at least the fourth pre-launch venture on the board this quarter (Amazon FBA, Kalibrol, Nativo, now this) while H+O and RWP have decision queues with real revenue waiting on him." And: "the plan's own words are that hours, not the market, are the constraint."
**Why overridden:** The Contrarian is right about the sprint and wrong about the test. Against the plan as written (30 to 60 hours of Ralph's month 1, 16 hours of phone warm-up, 28 uploads a day by week 4, a critical path that starts with Ralph hand-creating seven accounts) the opportunity-cost argument is decisive, which is why I kill all of that. But the ruling asks Ralph for roughly one hour over 14 days: 15 minutes watching 27 clips, one approval file, and a personal reply within the hour to anyone who says yes. Everything else runs in the engine that already exists and cost nothing to keep. Closing a built engine with zero demand data throws away the one asset here that is real, and it leaves the question open so it comes back next quarter as venture number five. A one-hour falsification test with a hard kill date is cheaper than the argument about whether to run it, and it either converts the engine into a small retainer business or into a content tool for the ventures that already pay. H+O and RWP stay first on the calendar either way; this test does not get to displace them. The other objections were weaker because each one is either an argument for running the test (Outsider's selection bias is exactly what the test measures), a fixable gate rather than a thesis break (fabricated research, personal Gmail, GAP filter width, caption placement, all engine work with no Ralph time), or an objection to the original plan that this ruling already adopts (no price in the first email, own accounts not a network, permission before cutting, dashboard now). None of them argues that a creator with a real Shorts gap cannot be found and asked. Only the Contrarian argues Ralph should not be asking at all, and that is the one worth quoting and the one I am overriding on cost, not on merit.
## What I verified before accepting this
| Claim by the council | Checked how | Result |
|---|---|---|
| 35 of 47 researched prospects have Shorts, all top-10 picks fail | `engine/verify_prospects.py` against live YouTube | TRUE: NO_GAP=35 WEAK=4 GAP=6 |
| Coinpoker passes the 'gambling blocked' filter | grep campaigns.json for eligible poker/prop rows | TRUE, regex missed 'poker'. Fixed, re-polled: 0 leak. |
| Log says '5 pass' when 13 do | daily.log vs campaign_poll output | TRUE, sliced before counting. Fixed. |
| Every sender defaults to the personal Gmail | grep engine/*.py | TRUE in outreach.py + specpack.py. Removed; sender now read from outreach/SENDER and refuses gmail.com. |
| Whop 10% creator fee missing from the model | research/PLATFORMS_2026.md line 26 (HARD) | TRUE. Added PLATFORM_FEE=0.10 to projection.py. |
| Hunter returned a YouTube-internal URL as the creator's website | prospects/hunted.csv | TRUE (accountlinking-pa-clients6.youtube.com). Filter fixed. |
| '21 assets in 40 seconds' is not on disk; jobs/ holds 4 clips | ls jobs/smoketest | TRUE. My claim was an extrapolation from a 2x2 run. Corrected in PLAN.md. |
| Instagram demotes aggregator accounts at account level | WebSearch: Tubefilter/PetaPixel 2026-04-30, TechCrunch 2025-07-14 | TRUE. Accounts that mainly repost are removed from recommendations, rolling 30-day window. Watermarks/speed changes do not count as original; unique text, edits, voiceover do. |
| YouTube July 2025 update targets clip channels | WebSearch: Social Media Today, YouTube Help | PARTLY. 'Inauthentic content' targets mass-produced templated video. Clips/compilations explicitly NOT banned; reused-content policy unchanged. |
| Projection has zero churn and no zero-client case | projection.py SCENARIOS | TRUE. Rewritten: 30% churn at each 90-day mark, ZERO scenario added, Layer C removed from base. |
| Cloudflare Pages project-per-prospect will hit the account cap | specpack.py design | TRUE by design. Moved to one project, one folder per pack. |
| Spec pack publicly hosts derivative cuts of a stranger's video before any license (contradicts plan step 10) | specpack.py + PLAN.md step 10 | TRUE. Packs now build locally; nothing is hosted or sent until the creator says yes. |

## Every objection to the recommended position, ranked by the chairman
- **[8] CONTRARIAN:** Opportunity cost: this is the fourth pre-launch venture this quarter (Amazon FBA, Kalibrol, Nativo, now this) while H+O and RWP have paying-client decision queues waiting on Ralph. Even a cheap test is attention he does not have.
  *Why this strength:* It is true, it is about the one constraint the plan itself names (hours), and it is the only ob