LLL · autonomous client-review build

Verification and Constraints

● RUNNING

One coherent journey. Twenty real ads, four Gemini landing pages, complete follow-up and a review Rita can understand.

Owner: Astra High · 01a0dd39-83d4-7d11-9035-af0c5d20f4ebRun: KL-FUNNEL-20260926-v3Updated: 26 September 2026

Goal and finish line

End goal: Give Matt every agreed client-review asset in a coherent Hungarian funnel, ready for one comment round and presentation to Rita.

Done means: One long commentable HTML contains persona and targeting, all 20 Facebook creatives, four complete Gemini 3.1 Pro LPs, the email and Stripe simulation, fourteen daily units and the testimonial kit. Distinct alternatives, a clear recommendation, real public URLs and independent deployed QA make selection easy. Two unsent emails explain what Rita should inspect.

0 / 44criteria verified

11deliverables

20 · 4 · 14ads · LPs · daily units

Rules that stay fixed

Matt has authorized the Astra High build now. Earlier Prepare-only and Sol defaults are superseded. No additional launch gate.
Current website brand, Rita’s actual corrections and every section of the supplied landing template. One broad meta-persona. Draft, critique and then polish.
Clickable review demonstrations. No real payment, upload, client send, ad activation or logged-in platform. No invented approval, results, date or scarcity. Shared metered allowance: USD 15.
Bare public review URLs. Preserve unrelated work and comments. Keep private source material outside public assets.

Proof standard: Each criterion needs the named observable result. GUI proof includes an actual screenshot, independent blind Luna High description and comparison. Missing proof stays unchecked.

End goal system

Two entry routes, one consistent offer

Both source groups inform the whole package. The free-video route reaches the cohort through useful email. The direct route reaches the same cohort offer immediately.

Input: latest Rita meeting + Figma commentsInput: current website + full template + persona workbook

One meta-persona

Common situations, motivations, objections and targeting.

20 Facebook ads

8 Figma selections + 12 ChatGPT-generated concepts.

D4–D5 · branch

Two entry paths

Video → feedback → useful email, or direct challenge ad → offer.

D4 · converge

4 complete LPs

A/B per offer, all 19 sections, three key CTA designs per offer.

D6–D8

Enrollment + experience

Stripe demo, confirmation, 14 daily units and closing/testimonial kit.

D9–D11

One review + handoff

Matt selects and comments. An unsent email guides Rita. Later loops implement.

Route to the end state

Internal quality checks, autonomous execution

The agent chooses implementation details. A convincing reference journey prevents weak design or copy being multiplied across the package.

1 · grounded

Reconcile sources

Read the full meeting, comments, workbook and visual references.

2 · reasoned

Draft and criticize

Copy, wireframe, reader thinking, objections and next-step value.

3 · quality

Finish one reference

A full convincing route and one usable lesson guide the rest.

4 · complete

Build the whole set

Exact counts, meaningful alternatives, one shared design system.

5 · verified

Inspect actual URLs

Independent editorial, functional and deployed screenshot checks.

6 · usable

Curate the handoff

Recommended choices, comparison, Rita email and future prompts.

Parent ↔ child messages

Meaningful changes only

PARENT · AUTHORIZED CONTRACT26 September 2026

The latest prompt explicitly authorizes one pinned Astra High orchestrator. This contract includes the workbook-based reader analysis, complete ad-to-payment journey, exact generator requirements, all template sections and a single parent check after 39 minutes. Source approval remains separate from draft execution authority.

Sources and authorityExact evidence stays private

Current client meeting and Figma reports, the original persona workbook including its dynamic questionnaire, current public website, the supplied landing template, and the prior local/cloud source gists ground the work. Matt’s new scope controls execution. Rita’s explicit comments and commitments control what can be called client-approved.

Private goal: plans/rita-funnel-20260926/goal.md. Each source fact, proposed term and unresolved conflict has an asset destination and evidence record. The old v5 frame or IMPLEMENTED label alone is not approval.

Decisions and testing mindsetMake choices without multiplying complexity

One meta-persona

Use common needs for the LPs, with different everyday moments for ad angles. Thoughts and emotions remain hypotheses.

Small first experiment

Compare a small recommended set. Twenty creatives are a useful library, not a requirement to split a tiny pilot twenty ways.

Later acquisition option

After the pilot, compare advertising both offers with only growing the video list and selling cohorts by email.

Work and proof

One card per deliverable

Reconcile Rita’s source decisions

Keep promises and professional meaning consistent.

BUILDING1 / 4

Show criteria, constraints, and evidence

Observable tests

When the latest Rita meeting receipt and full transcript are checked, then its date, participants, source IDs, exact agreements and newer-meeting search coverage appear in the private agreement ledger. When relevant current Figma comments and replies are compared with the existing variants, then every instruction maps to affected assets and final copy locations in the reconciliation ledger. When the completed package is checked against the latest instruction and historical sources, then the new daily-video/daily-feedback pilot appears consistently as Matt’s proposal, older no-feedback restrictions are not mistakenly reapplied, and no invented clinical result, prior approval or guaranteed SLA appears. When proposed terms are compared across the source ledger and review wrapper, then Matt’s proposals and Rita-confirmed facts and open terms remain distinctly labelled in one canonical term-status ledger mapped to customer-facing occurrences without publishing raw private sources.

Protect

× Never treat a draft or IMPLEMENTED label as Rita’s approval.
× Preserve client commitments and private source data.

Verified non-GUI evidence: D1.1: Full cached 18 August transcript read, exact meeting receipt and recency limit retained.

Screenshot proof: Pending for GUI criteria. Attach actual post-test capture, blind Luna High description and comparison. Unavailable evidence is not a pass.

Explain reader intent at every step

One meta-persona, useful hypotheses and draft-to-final logic.

BUILDING0 / 4

Show criteria, constraints, and evidence

Observable tests

When the workbook and source personas are synthesized, then one meta-persona with problem, framing, inner narrative, everyday moments, feelings and common objections appears at the beginning of the journey HTML. When each distinct funnel step is opened, then its copy/wireframe draft, critique and final asset have specific before-state, hook, first impression, reading order, consequential sentence reactions, objections, internal narrative, desired after-state, specific copy improvements and next-destination analysis. When the value explanation is read alongside each step, then desired outcome, credible likelihood/proof, time and effort are addressed naturally, with unsupported assumptions labelled and no fabricated numeric scores. When the targeting and journey table is inspected, then both entry routes, broad adult audience, proposed current Meta geography/language, audience/placements, objective/optimization, exclusions, timing and measurement settings, opt-in, buyer suppression and open/full/closed/no-cohort transitions appear with exact next actions; qualified video submission is distinct from click and paid enrollment from checkout visit. Recommend request-accepted web Leads and later confirmed-payment Purchase as conditional hypotheses; identify the 20 images as feed previews and later placement-fit adaptations for Advantage+ use, without claiming account availability.

Protect

× One broad meta-persona, not six new funnels.
× Reader psychology and campaign performance are testable hypotheses.

Evidence: Pending. Add exact files/URLs, test inputs and observed results before checking criteria.

Screenshot proof: Pending for GUI criteria. Attach actual post-test capture, blind Luna High description and comparison. Unavailable evidence is not a pass.

Create twenty complete Facebook ads

Eight Figma selections and twelve new generated concepts.

BUILDING0 / 4

Show criteria, constraints, and evidence

Observable tests

When the creative inventory is matched to the rendered master HTML, then exactly 20 complete ad variants with actual usable images, primary copy, headline, CTA and destination appear, covering both offers. When the eight reused images are traced to Figma, then exact file/node references, candidate comparison and selection reasons prove their origin and preserve Rita’s applicable corrections. If older assessment visuals are adapted, native node IDs and original exports remain traceable with every current-offer change documented; local fallback PNGs without native provenance do not count. When the twelve new concepts are checked against generation receipts, then twelve distinct real ChatGPT image-generator outputs and exported files exist, with no placeholders, stock or provider substitutions. When a cold reviewer reads each creative, its on-image hook and matching copy in order, then the offer, reader benefit, click expectation and matching LP are clear; corrections and a recommended small first test set appear.

Protect

× No fabricated provenance, client approval or performance.
× No campaign activation; preserve consistent brand and purposeful copy.

Evidence: Pending. Add exact files/URLs, test inputs and observed results before checking criteria.

Screenshot proof: Pending for GUI criteria. Attach actual post-test capture, blind Luna High description and comparison. Unavailable evidence is not a pass.

Generate four faithful landing alternatives

Gemini 3.1 Pro, complete template and meaningful choices.

BUILDING0 / 4

Show criteria, constraints, and evidence

Observable tests

When all four LP URLs and generation receipts are opened, then two free-video and two challenge pages generated by verified gemini-3.1-pro-preview with exact returned-model and metered-usage receipts contain complete source-grounded copy and functioning review interactions. The accepted Rita instruction clip is located and placed if available, otherwise two source-grounded scripts/storyboards, on-page placement and the explicit remaining recording dependency are inspectable without claiming finished footage. When the source template inventory is compared with each LP, then all 19 meaningful sections map to final anchors in a complete 19×4 coverage table, with no missing section, fabricated proof or template placeholder. Proof-sensitive rows identify their supporting source or the truthful evidence-free treatment used. When each offer’s hero/main CTA comparison is opened, then three distinct treatments per offer can be compared and selected, and the two complete LP variants per offer test a meaningful communication difference. When the current website and relevant Figma variants are compared side by side with deployed desktop/mobile LPs, then typography, palette, imagery, component character and copy corrections align; purposeful differences have an explanation.

Protect

× Use the exact supplied template and Gemini 3.1 Pro, without silent substitutes.
× Preserve all section functions and truthful offer/proof content.

Evidence: Pending. Add exact files/URLs, test inputs and observed results before checking criteria.

Screenshot proof: Pending for GUI criteria. Attach actual post-test capture, blind Luna High description and comparison. Unavailable evidence is not a pass.

Write the complete email relationship

Full useful copy, exact triggers and two real alternatives.

BUILDING0 / 4

Show criteria, constraints, and evidence

Observable tests

When the email inventory is opened, then two full versions with subject, preheader, body, CTA, destination, trigger and exclusion exist for receipt, feedback delivery, three content-led nurtures, weekly newsletter, four cohort promos, enrollment, all 14 day-specific delivery/reminder messages, closing invitation and post-program communication. When every nurture message is read aloud from hook to CTA, then a useful relevant observation leads naturally to the challenge, with no filler, forced sentimentality or fabricated individualized assessment. When no-opt-in, weekly, open-cohort, enrolled and closed/full scenarios are applied, then the trigger map shows correct suppression, replacement cadence and destination for every message. When all email links are followed on the hosted review, then each opens the named intended asset and continues the same offer and reader expectation.

Protect

× Rita owns actual personal assessment; examples are clearly samples.
× No sending, invented deadlines, em dashes or semicolons in client copy.

Evidence: Pending. Add exact files/URLs, test inputs and observed results before checking criteria.

Screenshot proof: Pending for GUI criteria. Attach actual post-test capture, blind Luna High description and comparison. Unavailable evidence is not a pass.

Demonstrate Stripe enrollment and recovery

Show every click, purchase expectation and next delivery.

BUILDING1 / 4

Show criteria, constraints, and evidence

Observable tests

When either challenge LP CTA is clicked, then the demo shows selected offer, full price/currency, Stripe-style checkout, successful confirmation and matching next-step email without a real payment or forced account. When cancel/back, retry/decline, processing/pending, sold-out and closed/no-cohort scenarios are selected, then clear recovery actions and truthful payment/access states appear on the hosted journey. When LP, checkout, success and email are compared, then contents, one-time price, proposed dates/access and next communication match exactly; screenshot and scenario evidence records every transition. When the future payment handoff is inspected, then server-side Checkout Session payment-state verification and idempotent fulfillment distinguish paid from pending and return-page visits; it handles checkout.session.completed plus async_payment_succeeded/async_payment_failed for delayed methods, supported by current official Stripe documentation. No real Checkout Session is created in this review build.

Protect

× Simulation only, no live Stripe or customer mutation.
× Do not claim pending payment or a success-page visit proves enrollment.

Verified non-GUI evidence: D6.4: Current official hosted Checkout fulfillment documentation retrieved, payment-state and idempotency handoff checked.

Screenshot proof: Pending for GUI criteria. Attach actual post-test capture, blind Luna High description and comparison. Unavailable evidence is not a pass.

Assemble fourteen usable daily units

Preserve all six modules and make each day actionable.

BUILDING0 / 4

Show criteria, constraints, and evidence

Observable tests

When days 1 through 14 are opened, then each contains purpose, complete source-supported task/instructions/cautions, usable material, completion cue and next step across all six original modules. When source documents and downloads are compared with the library, then professional content is preserved, every required source is mapped and every final download resolves. When A/B presentation and daily reminder alternatives are selected, then both complete editorial treatments preserve the same canonical content and include closing-consultation guidance. When a participant follows any daily unit, then they can inspect the exercise/guide, one daily video submission and the proposed daily Rita feedback for the first cohort. The free-video SLA stays distinct. A consistent proposed cutoff, feedback format and weekend/next-day treatment are explained. A twelve-person workload estimate uses explicit assumptions, with four-batch feedback shown only as a later efficiency alternative. The progress/remaining-needs Meet and source-supported filming guidance remain clear; no real uploads, login or clinical outcomes are implied.

Protect

× No invented exercises, clinical guarantees or unsupported daily duration.
× Do not build the real logged-in platform or live Messenger delivery.

Evidence: Pending. Add exact files/URLs, test inputs and observed results before checking criteria.

Screenshot proof: Pending for GUI criteria. Attach actual post-test capture, blind Luna High description and comparison. Unavailable evidence is not a pass.

Prepare testimonials and prior promises

A ready invitation/interview kit with honest source status.

BUILDING0 / 4

Show criteria, constraints, and evidence

Observable tests

When the testimonial review is opened, then six complete invitations organized into two coherent approaches and neutral interview questions are visible and usable. When consent choices are inspected, then participation, recording and public use are separate decisions with no fabricated permission or testimonial. When the booking CTA is followed read-only, then the intended verified service-specific booking page opens without creating an appointment. When earlier reporting and therapy-booking promises are compared with the secondary review section, then existing verified work or exact outstanding gaps appear with source references.

Protect

× No invented social proof, duration or client obligations.
× No client send, booking mutation or unrelated reporting rebuild.

Evidence: Pending. Add exact files/URLs, test inputs and observed results before checking criteria.

Screenshot proof: Pending for GUI criteria. Attach actual post-test capture, blind Luna High description and comparison. Unavailable evidence is not a pass.

Assemble one complete journey review

Every asset in order, with decisions and useful comparisons.

BUILDING0 / 4

Show criteria, constraints, and evidence

Observable tests

When Matt opens the long master HTML, then persona and targeting lead into all 20 ads and both full journeys, followed by daily/closing/testimonial materials, with every complete asset inspectable in-page and each distinct step having a standalone HTML. When recommended and alternate views are selected, then meaningful A/B and triple-CTA sections compare side by side on desktop and clearly on mobile, without incompatible offer combinations or divergent duplicate copy. When a disposable review comment and inline rewrite are saved and the page reloaded, then both persist and text/images within full asset views remain commentable; real review comments are preserved. The client master and verification board have distinct stable URLs and comment-document IDs recorded in the manifest. When the decision and experiment sections are read, then each material AI choice has a reason or alternative and the small first A/B test states hypothesis, variable, fixed facts and measurement without false certainty.

Protect

× A link list or screenshot-only presentation is not the complete asset review.
× Keep the customer preview clean and preserve stable comment/asset identity.

Evidence: Pending. Add exact files/URLs, test inputs and observed results before checking criteria.

Screenshot proof: Pending for GUI criteria. Attach actual post-test capture, blind Luna High description and comparison. Unavailable evidence is not a pass.

Verify the actual deployed experience

Independent copy, visual and functional evidence.

BUILDING0 / 4

Show criteria, constraints, and evidence

Observable tests

When all deployed sections/states are captured at desktop and 390px mobile, then the required Luna section-QA reviews, fixes and fresh postdeployment round have complete screenshot evidence and no unresolved material visual defect. When an index classifies all acceptance IDs as GUI or non-GUI and each GUI criterion is reviewed, then its actual screenshot, independent blind gpt-6-luna high description and orchestrator comparison are attached, with missing evidence left unchecked. When an independent reviewer reads the entire Hungarian package and walks both routes by mouse/keyboard, then no material contradiction, confusing transition, broken image/control or unsupported claim remains. When the final email URLs, downloads and fresh unauthenticated public responses are checked, then exact expected page identities load, review functionality works and an older host page remains intact.

Protect

× Never substitute green markers or HTTP 200 for exact evidence.
× One publisher preserves shared host pages; no forced destructive deployment.

Evidence: Pending. Add exact files/URLs, test inputs and observed results before checking criteria.

Screenshot proof: Pending for GUI criteria. Attach actual post-test capture, blind Luna High description and comparison. Unavailable evidence is not a pass.

Deliver Rita’s email and morning guide

Make selection easy and the later implementation actionable.

BUILDING0 / 4

Show criteria, constraints, and evidence

Observable tests

When both unsent Rita email drafts are read, then every exact final URL opens the named asset and the text tells her what to review, distinguishing proposals and demonstrations from activated systems. When Matt opens the morning overview and playable voice guide, then recommended variants, reasons, open decisions and review order are clear; the final Markdown transcript, two sequential independent Luna Max script reviews and decoded playable Hungarian Charon audio at 1.25× exist. When the two future executor prompts are opened, then campaign/funnel activation and full customer infrastructure are separately actionable from exact selected asset IDs and pending approvals, without launching either loop. When deliverables, prompts, generation/cost/source/QA evidence and completion report are compared, then every required result is accurately classified and the exact parent receives one terminal report with real URLs and any residual work.

Protect

× No automatic sending, future loop launch or guaranteed Rita approval.
× One combined final guide, no duplicate worker recaps or audio publication.

Evidence: Pending. Add exact files/URLs, test inputs and observed results before checking criteria.

Screenshot proof: Pending for GUI criteria. Attach actual post-test capture, blind Luna High description and comparison. Unavailable evidence is not a pass.

Open decisions and blockers

Build proposals now, confirm operations later

PROPOSALS FOR RITAUnconfirmed client terms

24,900 Ft, 12 paid seats, a short video of at most one minute and written feedback within three working days are Matt’s drafting proposals. Rita must confirm workload and service terms before activation. Start and closing dates remain open. The paid first cohort now proposes daily exercises, one submitted video daily and daily Rita feedback. A four-batch model is only a later efficiency option. The review makes Rita’s workload and these proposed terms clear without weakening every line of copy with caveats.

TRUTHFUL TEMPLATE ADAPTATIONAll 19 sections remain

Use verified facts, actual Rita imagery and supported proof. Adapt generic team, statistics, logos and before/after blocks to this offer without inventing people, endorsements or clinical results.

MODEL AND ACCESS REQUIREMENTSVerified route recorded privately

Gemini 3.1 Pro access has a successful minimal generation preflight. Four LPs still need actual generation receipts. Twelve new images must use the ChatGPT image generator. No silent substitution or placeholder acceptance.

Parent review window

One check after launch

39 minafter actual launch

1steering prompt

Idleparent between dispatch and check

At that single check the parent may clarify the contract and steer the same orchestrator using actual progress evidence. Then the parent sends one consolidated steering prompt and returns to idle. No repeated supervision loop. Afterward the orchestrator owns completion and its final report.

Build in progress

26 September · Mandatory sources read. Three bounded production packages are underway: four Gemini landing pages, twenty ad choices, daily materials and paired emails. The integration owner has drafted the receipt, feedback and complete payment-state demonstration. These drafts are not yet deployed acceptance evidence.

Current primary proposal: daily materials, one daily practice video and daily written Rita feedback. Proposed cutoff 18:00, reply by the next day at 18:00 including weekends. Rita’s capacity and dates remain decisions for the review.

Verification remains 0/44 until named artifacts and independent deployed checks establish each criterion.

Live build · landing and content assembled

The client journey now contains four complete Gemini landing alternatives, six hero messages, fourteen daily units, eighty-two email versions, six testimonial invitations and the simulated payment/recovery path. Independent editorial review found bounded wording/download-formatting corrections, now assigned to the original content owner. Actual deployed Luna section and interaction QA is in progress. The twenty-image advertising package remains in production.

2 / 44 checks are currently marked verified. Completed assets do not by themselves satisfy screenshot and independent-review acceptance. Metered generation so far: USD 1.613698. Final guide reservation: USD 2. Total task cap: USD 15.