Architecture

A.1 The chain at a glance · measured 2026-08-10

Five stages, fifteen steps, one direction. Every number below was read live from the Notion Outreach Prospects DB and from the D1 database byforja on 10/08/2026. Nothing here is an estimate.

The bottleneck is not the machine. It is the first send. Steps 01 to 05 have produced stock (34 qualified rows, 24 mockups rendered, 15 rows priced). Steps 06 to 09 have never fired once.

The three degrees of automation

Nothing in the acquisition chain is automatic end to end today. That is the single most useful fact on this page: it says exactly where the future gains are, and it says that the current throughput is capped by AL's own hours.

A.1b AL's daily operating loop · added 2026-08-24

The 4-step loop AL actually runs, every batch, underneath the staircase below (steps 01-10). Sourcing now includes the mockup as one inseparable step — a prospect is not "sourced" until both exist, since a missing mockup blocks recording outright (Grammar 03 §2b). This was found broken for an entire 10-prospect batch on 24/08/2026 (mockup step silently skipped), hence folding it into step 1 for good.

1. Source + mockup, together. An agent sources the batch and renders every prospect's mockup PNG in the same pass — staircase steps 01 and 04, never one without the other.
2. Hand-off. Claude gives AL a list of 10, each with a tailored bullet-loom already written into that prospect's Notion page (scripted open + close only, everything between improvised — cadence doctrine §7 / Grammar 03 §7, method validated 24/08/2026), plus the folder of 10 rendered mockups.
3. Record, one at a time. AL records the raw 5-to-8-minute Loom, then sends Claude the transcript ("c'est parti pour X"). Claude returns the day's email, the Loom title (fact + contrast method, Grammar 03 §7b), and drafts the T2/T4 follow-ups straight from that transcript.
4. Confirm. AL confirms the send; only then does Claude set Date Loom — never before (rule corrected 24/08/2026, cf cadence doctrine).

A.2 The staircase, 15 steps in order · last updated 2026-08-10, STALE — see A.4

Read it top to bottom. Each stage steps one notch to the right, each step inside a stage steps one notch further. The colour of the left edge is the degree of automation, the pill on the right names it.

A.3 Automation map · last updated 2026-08-10

The same fifteen steps, laid flat. One block per step, in order, coloured by degree. You can see the shape of the problem without reading a word: a blue run that produces stock, a violet wall that nothing crosses, then blue again on the far side.

The automatic band has zero width. It is not drawn because there is nothing to draw.

How to read the wall

Steps 06 to 10 are five consecutive manual steps, and they sit exactly where the pipeline is stalled: 34 prospects are ready, 0 have been contacted. Two of the five are irreducibly manual (a face on camera at 06, a live conversation at 12). The other three, sending at 07 and the two follow-ups at 08 and 09, are scheduling and text, which is the kind of work a scheduler can hold. That is where the first real automation gain sits, and this page will say so until it moves.

A.4 Open loops and contradictions · last updated 2026-08-10

Points where two sources disagree, or where a step is defined but not decided. Listed here rather than smoothed over, because a chain is only as reliable as its worst-defined link.

Resolved on 10/08 evening, in the sources themselves: Loom length (15 to 45 s, one flaw), Grammar 04 client copy (40 sec, one concrete thing), sourcing bands rewritten in USD on the two real thresholds, mockup mandatory everywhere, EUR prices in Grammars 06 to 11 replaced by live pointers, section references repointed to the numbered doctrine, overtime at 100 USD, 12 publications contractual with 15 as unpromised bonus, F.V. city settled on Le Touquet, Growth kept at 4,000 USD. Still open, AL's call: the artist run with no exit clause, and the two Estimated payment capacity labels that still encode the dead EUR grid. The toggles below are kept as the historical record.

A.5 The two profiles · summary of doctrine, read 2026-08-10

Step 02 sets the profile, and everything downstream hangs off that one call: the formula, the price, the contract articles, whether the client ever becomes autonomous. This section is a summary of what is already written in Doctrine byForja and Doctrine Pricing. Nothing here is new doctrine.

The line between the two is not artistic talent. It is the relationship to business. A pure artist lives their business as an extension of their creative ego, so every strategic decision is felt as an intrusion into their soul, and they have no mental structure to absorb a frame. A craftsperson runs a trade: a workshop, costs, customers, a production logic. Part of the operational brain is already switched on, so they can hear "here are the three positionings that survive the benchmark" without experiencing it as a violation.

Underneath, the machine is one machine. Same production pipeline, same interface, same moat. What differs is only the human piloting layer, which means how the material comes in and how decisions get validated. This is not two businesses. It is one factory with two loading docks.

Where a prospect sits on the line

What each formula includes

Base prices in USD before the market multiplier. Amounts are read live from Doctrine Pricing at execution time, never from this page. Launch cards in blue, monthly cards in green.

The two profiles are not a binary. They are the two ends of one spectrum. A real client sits somewhere on the line, and the interpolation point is fixed by exactly two things: skill and stability. The table below shows the poles. Reality interpolates between them.

What actually changes between the two

Both profiles pass the same solvency gate, unchanged by typology: 420,000 USD of annual client revenue to be eligible for a retainer at all, roughly 200,000 for a one-shot launch with no monthly engagement. The price is never lowered to make somebody eligible. The retainer is removed, or nothing is signed.

A.6 What AL delivers, and at what cadence · summary of contract and D1, read 2026-08-10

Contact cadence first, then what is actually handed over, then the clauses that keep the run survivable. The cadence and the launch pack come from the contract, articles 4.1 and 4.2. The monthly pack is read from the D1 table client_monthly_plans, which is what the tool is built to ship.

Contact cadence, the 6 week launch

Deliverables and rules, told apart

The launch pack, handed over once

The monthly pack, handed over every month

Seven artefacts, one row per client per month in client_monthly_plans. The shape below was built from the plan made for the first client, and it is the contract of the generator at step 15.

The rules that keep AL protected

What the client leaves with, and what never leaves

Left column, the machine: the full version, AL’s property, consultation only. Right column, the extract the client actually receives, with the step where the extraction happens and the week it lands. The client owns the output, AL owns the machine that produced it.

Applied to the first client, the boundary is exact: the eight cover PNGs are hers, the system that generates them is AL's. The month plan is hers, the 15 cell grid logic behind it is AL's. She left with a complete month and could not rebuild a second one alone. That is not a trick, it is the definition of the offer: the operational autonomy transfers, the strategic piloting does not.

The Pilot nuance. A Pilot graduates on the operational layer and ends up publishing, editing and updating alone. Even fully graduated he never touches the strategy or the moat. The doctrine states it plainly: he ends up doing everything in the middle, never at either end.

A.7 The delivery staircase, 6 weeks · time measured against the first client, 2026-08-10

Same wall as A.2, on the other side of the signature. Every step carries what it produces, what it costs AL in hours, who owns it, and whether it exists for both profiles or only for one. The hours come from the valuation sheet of the first delivered client: 159.5 hours worked for a contract of 1,350 EUR, which is 8.46 EUR an hour. That number is why this wall exists.

The line that decides the price is not the deliverable, it is the camera. Everything above the shoot is a system that runs in hours; everything from the shoot down is AL's own time, hour for hour, forever.

What this wall says about the price

Sixteen steps, and only two of them carry the weight: the shoot at 11 and the edit at 12. Together they are twenty of the thirty hours. Every other step either runs on a script or costs a fixed hour of call.

So the pricing question has a mechanical answer. An offer that stops at step 10 costs AL ten to fifteen hours and can be sold from 1,500 USD without disrespecting his time. An offer that includes 11 and 12 costs thirty hours and cannot go below 3,500 USD. There is no middle: the camera is the cliff.

A.8 What makes a prospect feel understood · writing rule for every Loom script, set 2026-08-10

This is not theory and it is not a style note. It governs how step 05 gets written from now on, and the opener it produces is what carries into the email at step 07.

The wow never comes from the diagnosis. Listing a flaw only proves that somebody looked, and anybody can open a website. The wow comes from naming a tension the prospect lives with and has never put into words. On an artisan there are four of them, and they are always the same four.

The four tensions, always the same four

The three beat structure

The first beat is the one every cold emailer skips, and it is the one that makes the difference. It has to show that what he makes was understood, not only what is broken on his site.

Five openers from the first wave

Quoted as written, from the alternative openers recorded on the five prospect pages. Not rewritten.

One level up, the verifiable prediction

A sentence that asserts something the prospect knows to be true and has never said out loud.

I would bet your best sales still come from word of mouth and from markets, not from the site.

On an artisan that is true nine times out of ten. When it lands, he does not think "he looked at my site", he thinks "he knows my trade". That is a different category of credibility, and it is the only line in the script that buys it.

The three things never to do

A.9 The firing map: step → deliverable → prompt → what the client receives · built 2026-08-11 after the Buckleather rehearsal

A.7 says what each step produces. It did not say which prompt fires it or what artifact the client actually holds, which is why the Buckleather rehearsal folder looked richer than the staircase. It was not richer: the same six deliverables were simply named by week instead of by D number. This table is the missing join. It is the only place where a D number, a Grammar, a Prompt and a real file sit on the same row.

Related pages: Actions board for the tickable version, Pricing calculator for step 03 and for the A.5 formulas, Call Cockpit for the live formula decider and the post-call score, Discovery questionnaire for where the profile is recorded, CRM for what happens after step 14, All links for everything else.