CONFIDENTIAL. Distributed separately from the technical presentation. Planning estimates, not quotes. Not an offer, salary, invoice, debt or obligation.
Private Financial & Execution Proposal
AI Football is an idea conceived by its two promoters: 22 autonomous virtual players in an 11v11 simulation, a sealed match record, and later 3D production. This document sets out what the first phase would cost in cash, how the engineering could be executed, how long it may take, and how a separate visual investment would be released step by step, only after results.
At a glance
| Item | Amount | Paid in cash? | Basis |
|---|---|---|---|
| A First funding tranche, maximum | 2,400 | Yes, if authorised | 2 months × 1,200 ceiling; no automatic renewal |
| A Operating cash, central scenario | up to 8,400 | Yes, by tranches | 7 months × 1,200 (range 7,200–8,400); a maximum, not expected spend |
| B Value of the proposed lead-engineering contribution | ≈ 58,000–97,000 | No | ≈ 6.5 months × 9,000–15,000; form of any compensation not proposed here |
| C Hiring the same work instead (labour) | ≈ 63,000–347,000 | Only in the hiring route | 1,260 h, from a remote senior contractor to an agency; comparison only |
| D Future visual investment to pilot | ≈ 112,000–1.05M | Not now | Central ≈ 370,000; orders of magnitude, released by gates A, B, C |
B (the value of the engineering contribution) and C (the cost of hiring that work) are two ways of measuring the same work. They are never added together, and neither is added to the cash in A. If hiring happens, C becomes cash and replaces B.
Every amount in this document is a planning estimate or an order of magnitude, with its assumptions shown. None is a quote. No figure here authorises spending by itself. Any funding, hiring or spending requires express, written authorisation from the project's promoters, under the decision rules they establish. The author of the technical assessment acts as technical feasibility adviser and does not hold that authority.
Four different amounts, never summed
The project's economics are easy to misread if four different things are added together. They answer four different questions.
| Concept | What it is | Cash the company pays in the first phase? | Central figure |
|---|---|---|---|
| A. Operating cash | Money the project actually pays: AI tools, isolated cloud machine, storage, services | Yes, as a maximum budget released by tranches | up to 8,400 |
| B. Contribution value | Estimated economic value of the lead-engineering hours and expertise, contributed in kind | No. Not a salary, invoice, debt or obligation to pay | ≈ 58,000–97,000 |
| C. Replacement cost | What hiring professionals to do the same engineering would cost | No, unless the hiring route is chosen: then C becomes cash and replaces B | ≈ 63,000–347,000 |
| D. Future visual investment | Animation specialists, motion data, 3D art, rendering | Not now. Separate budget, released only through gates A, B and C | ≈ 370,000 to pilot |
Labels used throughout: strong evidence weak evidence missing evidence derived here. "Derived here" marks arithmetic this document performs on v1.3 inputs that v1.3 itself does not print.
USD 1,200 a month, as a ceiling
The planning budget is a set of maximum monthly allocations, tax included. They are funding envelopes, not list prices or invoices. Every real expense is recorded with its receipt and summarised monthly against the allocation. Money not spent from an allocation is not spent elsewhere.
| Item | Area | Maximum allocation | Expected spend |
|---|---|---|---|
| ChatGPT, intensive engineering use | AI engineering tools | 250 | 100–200 + tax |
| Claude and Claude Code, intensive use | AI engineering tools | 250 | 100–200 + tax |
| Isolated Linux machine for simulation | Cloud infrastructure | 300 | 80–260 |
| Storage, backups and logs | Cloud infrastructure | 100 | 5–50 |
| Tools, testing and extra capacity | Tools and services | 300 | 0–300 |
| Total | 1,200 | ≈ 285–1,010 + tax |
Released by tranches
| Duration | Maximum cumulative cash |
|---|---|
| 2 months (first tranche) | USD 2,400 |
| 3 months | USD 3,600 |
| 4 months | USD 4,800 |
Recommendation in v1.3: authorise at most two months (USD 2,400), conditional on results. Extend only after the milestones in §07 are verified. There is no automatic renewal.
Honest reading
- The 1,200 is a ceiling with tax included. In the highest case, with tax around 20%, spending would approach it.
- If it would be exceeded, discretionary items are cut or spending pauses until an increase is approved in writing by the promoters.
- The tax rate depends on the billing jurisdiction, which is not fixed yet.
- Costs that can fall outside subscriptions: API usage, extra usage credits, taxes, cloud traffic, snapshots and backups, extra IP addresses, a temporary second machine, currency conversion for plans billed in euros, and a football adviser if paid (from the tools line).
| Scenario | Weeks | Billable months | Maximum cash (USD) |
|---|---|---|---|
| Favourable (target plan) | 16 | 4 | 4,800 |
| Central (planning basis) | 28 (range 24–30) | 7 (6–7) | 8,400 (7,200–8,400) |
| Unfavourable | 32–40 | 8–10 | 9,600–12,000, or stop earlier |
Two intensive-use AI subscriptions, USD 250 each
The engineering plan assumes AI-assisted development at about 45 hours a week. Neither provider sells a USD 250 plan, as far as could be read; each allocation covers one intensive-use plan, tax, and some API or extra usage, or a higher tier if needed.
ChatGPT · allocation 250 · expected 100–200 + tax
- Plus at USD 20; Pro tiers at USD 100 and 200 according to April 2026 press reports. press, not the price page
- One October source cites a USD 500 tier; not confirmed.
- The official pricing page could not be read (HTTP 403).
Claude and Claude Code · allocation 250 · expected 100–200 + tax
- Official pricing page: Pro USD 20 a month (17 billed yearly), Max USD 100; both include Claude Code. price page read
- Usage limits apply every 5 hours and every week; tax extra.
- Max would include USD 100 a month of API credits, per a summarised reading of the page (pending confirmation). Higher tiers were not visible.
- The 200 upper figure assumes extra usage or a higher tier not yet confirmed.
Risk R15 applies here: if AI acceleration is lower than assumed, hours rise and the calendar moves toward the central or unfavourable scenario. It is measured from month 2 (hours against milestones). Tool usage limits can also slow the heaviest days.
An isolated Linux machine, and small storage
Third-party simulation code runs only on a rented, isolated Linux machine that can be destroyed and rebuilt from the repository. The developer's Mac is used for programming only. List prices below were read on 14 Nov 2026, before tax; they are references, not accepted quotes or commitments to contract.
| Provider and plan | vCPU | Memory | Disk | Monthly list price | Note |
|---|---|---|---|---|---|
| Hetzner CCX23 (dedicated vCPU) | 4 | 16 GB | 160 GB | EUR 73.49 | Shown as unavailable when read; Helsinki price; VAT extra |
| Hetzner CCX33 (dedicated vCPU) | 8 | 32 GB | 240 GB | EUR 117.99 | Same |
| DigitalOcean General Purpose | 4 | 16 GB | 50 GB | USD 126 | Backups 20–30% of price |
| DigitalOcean General Purpose | 8 | 32 GB | 100 GB | USD 252 | With tax, slightly above the 300 allocation |
| DigitalOcean CPU-Optimized | 16 | 32 GB | 200 GB | USD 336 | Exceeds the 300 allocation |
Recommendation
Start with 16 GB and 4 dedicated vCPU; it fits the allocation with either provider. Move to 32 GB only if measurements in weeks 1–2 show memory or time per match require it, or to run two matches in parallel. Contracting remains pending authorisation.
Storage, backups, logs · 100
Object storage about USD 7–15 per TB-month; Hetzner snapshots EUR 0.0143 per GB-month; DigitalOcean backups 20–30% of the machine price. A batch of short matches is expected to take under 1 GB hypothesis. Expected 5–50 a month.
Security that the budget buys
- A provider account used only for AI Football; an exclusive SSH key.
- No credentials of other projects or AI connectors on the machine.
- Third-party code pinned by commit, run without network, separate user, memory and process limits.
Not in the first-phase budget: cloud GPUs for training (USD 0.89–1.29 per GPU-hour mid-range; 2.89–4.89 high-end). Training is not planned in this phase.
Two ways to get the same engineering done
The first-phase engineering (simulator integration, the AI Brain, data contracts, testing, debugging and the isolated infrastructure) can either be hired, or carried out by an internal technical capability: a lead engineer (role) contributing the work in kind. Operating cash (A) is paid in both routes.
| Route 1: hire professionals | Route 2: internal technical capability | |
|---|---|---|
| Brain and integration engineering | Paid in cash: ≈ 63,000–113,000 (remote senior contractor) to ≈ 130,000–189,000 (US employee), plus 10–20% QA/DevOps | Contributed in kind by the lead engineer (role); no salary in the first phase |
| Recruitment | None for contractors; 27,000–40,000 and 2–4 months of delay for employees | None |
| Technical direction | Someone still has to direct the hire (not costed) | Included in the role |
| Tools and infrastructure (A) | Up to 8,400 | Up to 8,400 |
| Cash the company pays | ≈ 71,000–122,000 (remote) to ≈ 165,000–237,000 (US), plus QA/DevOps | Up to 8,400 |
| Value of contributed work (not cash) | Not applicable | ≈ 58,000–97,000 |
| How the work is compensated | Salaries or fees | Not proposed here; to be discussed privately with legal advice (§05) |
The contribution value (B, ≈ 58k–97k) and the hiring cost (C) measure the same engineering. They never appear together in a total. Route 1 turns C into cash; Route 2 keeps cash at A only.
A sturdier variant of Route 1 would be two engineers (game AI; simulation and integration), at a higher monthly cost and with more coordination; v1.3 does not cost it. Route 2 does not make the visual production cheap: that is planned separately (§08–14).
Value of the proposed lead-engineering contribution
The role covers, in one person: architecture and technical direction, the AI Brain and agents, simulator integration, data contracts, testing, debugging and the isolated infrastructure, at about 45 hours a week (≈ 195 hours a month, 1.2 times a standard 160-hour month).
Proposed value: USD 9,000–15,000 a month
About USD 46–77 per hour at 195 hours a month. Value of contributed work by scenario:
| Scenario | Dedication months | Estimated value (USD) |
|---|---|---|
| Favourable | ≈ 3.7 | ≈ 33,000–55,000 |
| Central | ≈ 6.5 | ≈ 58,000–97,000 |
| Unfavourable | ≈ 7.4–9.2 | ≈ 66,000–138,000 |
Dedication months = hours ÷ 195. Not added to the cash requested.
How it compares with the market
The range is conservative: it sits at the low end of a single remote senior contractor, for a role that combines several profiles. Evidence would support up to about USD 17,500–18,000 a month (remote senior at 45 h/week); it was deliberately not raised. A UK machine-learning contractor at the same hours would be GBP 9,800–19,200 a month.
The value was proposed by the person being valued. The promoters may want to check it independently.
Topics to settle with legal counsel before any work starts
A negotiation checklist, not legal or tax advice.
- IP assignment to the company. The Brain, software, data contracts and documentation assigned to the company from the first line of code, by explicit assignment; treatment of any work done before incorporation.
- Attribution and recognition of contributed work. How hours and deliverables are recorded (the repository already leaves a trail), and whether they are valued by a fixed formula or a dynamic model.
- Form and timing of any compensation. Cash, deferred, equity or a combination; whether and when a paid arrangement would start, for example after a funding round.
- Milestones, responsibilities and commitment. Hours, milestones H1–H4, and what each party commits.
- Vesting, only if agreed. If any equity element were agreed: by time or by milestones. Multi-year vesting with an initial minimum period is common practice; an observation, not a recommendation.
- Exit. What happens if the engagement ends: IP, code access, and anything already earned.
- Future hiring. Who decides to hire engineers, with what budget.
What hiring the same capability would cost
The realistic replacement is one senior multidisciplinary engineer plus 10–20% QA/DevOps support, not five hires. The engine comes from an existing open-source project; the work is integrating and validating it, not writing physics from scratch.
| Profile | Separate hire? | Reasoning |
|---|---|---|
| Senior AI / Game AI engineer | Core of the role | The Brain is the original part of the project |
| Simulation engineer | Combinable | Integrate and validate a third-party engine in Python/JAX |
| Software architect / lead developer | Combinable at this stage | Small codebase; but a hire still needs someone to direct them |
| Integration / backend engineer | Combinable | Adapters, record and 2D viewer are bounded in scope |
| QA / testing / DevOps | Part-time, 10–20% | Isolated-machine security and automated tests justify occasional support |
Hiring cost explorer
Formula: hours × hourly range (+ QA/DevOps) + operating cash + recruitmentLabour uses the Annex F1 reference hours: 720 h (favourable), 1,260 h (central), 1,800 h (unfavourable, the 40-week end). Total cash if hired adds the scenario's maximum operating cash and, for a US employee, a 27,000–40,000 recruitment fee. QA/DevOps adds 10% to the low end and 20% to the high end. Totals other than the central remote and US figures are derived here with the same formula.
| Option | Reference | Source | Evidence |
|---|---|---|---|
| Senior remote contractor (E. Europe, LatAm) | USD 50–90/h | Arc.dev freelance rate survey, n=5,302, May 2026; lemon.io calculator (search summary) | strong |
| Senior global freelancer | USD 75–150/h (staff/principal 120–220) | Arc.dev, May 2026 | strong |
| US employee, total employer cost | ≈ USD 103–150/h: base 136,000–200,000 + ≈ 46% load | US BLS Occupational Outlook (May 2025); BLS ECEC (June 2026); Levels.fyi (summary) | strong |
| UK machine-learning contractor | GBP 575/day (P10 402, P90 788) | ITJobsWatch, 6 months to 14 Nov 2026 | strong |
| Development agency (NA / W. Europe) | USD 165–275/h | A vendor blog, unverified | weak |
| Recruitment | 20–25% of first-year salary (up to 30% for scarce profiles); 2–4 months to start | Recruitment-firm blogs, unverified | weak |
| Option | H2 central (≈ 395 h) | H3 central (≈ 790 h) | H4 central (1,260 h) | H4 favourable (720 h) | H4 unfavourable (1,800 h) |
|---|---|---|---|---|---|
| Remote senior contractor | 20–36 | 39–71 | 63–113 | 36–65 | 90–162 |
| Global senior freelancer | 30–59 | 59–118 | 95–189 | 54–108 | 135–270 |
| US employee | 41–59 | 81–118 | 130–189 | 74–108 | 185–270 |
| Agency | 65–108 | 130–217 | 208–347 | 119–198 | 297–495 |
Add, depending on the option
- QA/DevOps support: 10–20% on top.
- Employees: recruitment fees of 27,000–40,000 (20–25% of 136,000–160,000) and 2–4 months before starting.
- Always: weeks to learn the domain, and someone to direct the hire.
- Tools and infrastructure (A) cost the same in any route.
What one engineer rarely covers
- Expert football judgement (hence a football adviser)
- Infrastructure security hardening, if not a specialty
- Rigorous statistical evaluation of acceptance criteria
- Full rules and referee correctness
- Large-scale machine learning, if rule-based players are not enough
- Sustaining 45 hours a week for months with no backup
- All audiovisual work, which needs other specialists (§08)
16, 28 or 32–40 weeks to a technical prototype
Weeks count from a future authorisation to develop; no date is committed. The central scenario, not the favourable one, is the basis for cash and expectations. A milestone reached is a technical result, not proof that the football is convincing.
Scenario explorer
Every output shows its formula| Hiring option | Formula | Labour only (USD) | Total cash if hired (USD) |
|---|
Billable months = weeks ÷ 4.33, rounded up (calendar months of operating cash). Dedication months = hours ÷ 195 (months of 45-hour weeks). In the unfavourable scenario the contribution uses 1,440–1,800 h, while replacement cost uses 1,800 h, as Annex F1 does.
| Scenario | Weeks to H4 | Assumption | Hours |
|---|---|---|---|
| Favourable (target plan) | 16 | The engine works with minor adjustments; AI accelerates substantially; few behaviour iterations | ≈ 720 (low end of the earlier 640–1,840 h effort estimate) |
| Central (planning basis) | 28 (24–30) | Engine or environment adaptations; more behaviour cycles; weeks of lower dedication | ≈ 1,260 (1,080–1,350) |
| Unfavourable | 32–40, or stop | The candidate engine fails and must be forked or replaced; behaviour needs redesign | ≈ 1,440–1,800 |
| Weeks | Objective | Hours | Verifiable milestone |
|---|---|---|---|
| 1–2 | Evaluate an existing simulator experimentally (1a isolation, 1b engine) | 90 | H1: stage 1b verdict on short matches |
| 3–5 | First agent controlled by our AI Brain | 135 | H2: one player moved by our Brain through contract 1 passes its scenarios |
| 6–10 | Intelligence for 22 players; first 5-minute 11v11 simulation in 2D | 225 | H3: a complete, visible 5-minute 2D match; quality not yet evaluated |
| 11–16 | Correct, evaluate behaviour, consolidate records, demonstrable prototype | 270 | H4: technical prototype evaluated against pre-set criteria |
| Total | 720 |
Strict order
- Simulator
- Brain
- 2D match
- Animation feasibility
- Small visual test
- Only then, larger production
Weekly capacity of 45 h (planning split): engineering 20, testing 6, debugging 7, validation and documentation 4, external dependencies 4, risk reserve 4.
What could move the calendar
- Engine compatibility (its CI uses Python 3.11; current JAX needs 3.12+): days to weeks
- Engine CPU performance unknown: may need 32 GB or smaller batches
- Behaviour quality: the most likely cause of delay in weeks 6–16
- AI tool usage limits on the heaviest days
- The calendar depends on one person's availability
Twelve visual components: reuse, develop, or a specialist
Reasoned estimates to size a future, separate investment. Not budgets or quotes; nothing here is funded before its gate. Scope assumption: the lead engineer (role) keeps building and maintaining the Brain, the simulation, the record and the integration, including the record export. None of it exists yet; that assumption removes Brain engineering from these figures and reduces the need for a dedicated graphics engineer.
| # | Component | Approach | Price reference |
|---|---|---|---|
| 1 | Technical animator with sports experience | Specialist central profile | US senior technical animator ≈ 113,000 average base (summary); animators overall median 102,030 (BLS). no freelance rate found: 50–150 USD/h used |
| 2 | Unreal Engine / real-time graphics engineer | Specialist part-time, or covered by #1 | UK median GBP 65,000 base; contractors 35–54 USD/h senior median, US average 60; Berlin ≈ EUR 112/h |
| 3 | Motion matching, IK, rigging, foot–ball contact | Develop on existing Unreal tools | Within #1 and #2. This is where the technical risk concentrates |
| 4 | Licensed animation libraries | Reuse | Packs USD 20–110 on Fab; a 308-clip football pack USD 109.99. Many forbid AI training |
| 5 | Motion capture | Specialist or own kit, if libraries fall short | Studio typically 1,500–3,000/day (500–5,000); clean-up 2–8× capture time at 75–150/h. Own kit 4,395–4,495; markerless 18–490/month |
| 6 | 3D player models | Reuse bases and customise; specialist finish | 1,000–10,000 per game-ready character; hero EUR 2,500–8,000; AAA 25,000+ vendor blogs |
| 7 | Kits, stadium, lighting, cameras | Reuse and customise | Store stadiums 40–1,499; customisation by a 3D artist |
| 8 | GPU rendering | Develop the pipeline; rented or own compute | 0.89–4.89 per GPU-hour; workstation from ≈ 8,700; pro GPU ≈ 16,000 |
| 9 | Video storage and processing | Reuse services | 7–15 per TB-month; 6–23 TB uncompressed per 96-minute programme at 60 fps |
| 10 | Editing, commentary, sound, scoreboard, TV graphics | Reuse tools; occasional specialists | Editing 15–35/h (conflicting data); motion graphics 28–325/h; synthetic voice 22–299/month; commentator 200–350/h. Scoreboard and graphics generated from the record |
| 11 | Visual quality control and fixes | Develop tools + specialist hours | Animator hours; errors per minute not yet known |
| 12 | Rights, licences, legal advice | Specialist, occasional | US lawyers average 362/h, corporate 461/h. Unreal licence: no fee below USD 1M group revenue including funds raised (provisional) |
A video editor is not a technical animator
The hard problem is movement and contact, not editing. An editor cannot do the animator's work. A qualified technical animator, by contrast, can cover components 1–3 and part of 8, supported by Unreal's tools and the lead engineer's integration work.
Do not assume five hires
Starting requires one well-chosen central specialist, with occasional support in 3D art, capture and sound as the project advances. No single profile solves the audiovisual problem alone; motion data is a separate need.
Each step is approached only if the previous one succeeds
Orders of magnitude with explicit assumptions, not quotes. Specialist hours are an own estimate without a measured basis. A firm visual budget is only possible after V1 has been measured.
Visual investment ladder explorer
Log scale; select a step to see cumulative totals| Small visual experiment (V1) | Reusable production system (V2–V4) | |
|---|---|---|
| Purpose | Learn whether it is possible and what it costs | Produce matches repeatably |
| Order of magnitude | Tens of thousands of USD | Hundreds of thousands to the pilot, plus a cost per match |
| What remains | Importer, contact test, measurements | Full pipeline, motion library, assets |
| Decision it enables | Fund V2 or not | Fund production or not |
About the earlier USD 240,000–1.44M figure. It assumed stages 4a–8 done by a fully hired team, including the Brain and simulation engineering. It was never a quote for animation, nor what the project would spend. With the Brain kept in-house, 4a + V1 + V2 + V3 add up to ≈ 112,000–1.05M, central ≈ 370,000 (taking ≈ 10,000 for the feasibility study), before V4's recurring cost per match.
Visual feasibility study: ≈ USD 2,000–18,000
- Work
- A technical animator consults for 40–120 hours on the 2D match records: what share of the simulated contacts a body can perform, and what motion data would be missing.
- Cost
- ≈ 2,000–18,000 (40–120 h at 50–150 USD/h). Totals use ≈ 10,000.
- Released by
- Gate A, after stage 3 (the 2D match). Gate A authorises only 4a and 4b (simulation consolidation), with a spending cap.
Why it comes first
It is the figure that allows everything after it to be estimated on a real basis. It also tests whether the animation specialist can be found (risk R9) and whether licensed motion data is enough (risk R3).
First 3D test (V1): ≈ USD 17,000–135,000, central ≈ 50,000
| Work | 20–30 seconds of a real match: import the record, animated bodies, verifiable contacts, fixed camera |
|---|---|
| Specialists and effort | Technical animator with Unreal knowledge, 320–640 h; the lead engineer (role) for export and verification |
| Software | Unreal Engine (no fee below the revenue threshold; provisional) |
| Motion | Licensed packs (100–500); 0–2 capture days if key gestures are missing |
| Characters | Free or store bases (0–2,000) |
| GPU | Cloud, 25–600 h (≈ 20–2,900), or an own workstation if decided (≈ 8,700) |
| Estimated cost | ≈ 17,000–135,000; central ≈ 50,000 (480 h at 90 USD/h, one capture day, cloud) |
| What prevents a firm figure | The specialist's real rate; how many gestures the libraries lack; render seconds per frame |
| What would sharpen it | The feasibility study; a quote from the specialist |
Advance to Gate C only if pre-declared thresholds are met without frame-by-frame manual correction, and V1 yields measured costs (specialist hours per second of video, render seconds per frame) good enough to budget V2.
Short 3D match (V2): ≈ USD 75,000–670,000, central ≈ 250,000
3–5 continuous minutes: full gesture repertoire, body-to-body interaction, referee, stadium, several cameras, scoreboard, sound. Gate C is the major investment decision and uses V1's measured costs.
Effort
- Technical animator 1,000–2,000 h; graphics engineer 0–960 h if the animator does not cover it
- 3D artist for 6–10 characters (two teams, goalkeepers, referee) and stadium customisation
- Capture 3–10 days plus clean-up, or own kit plus markerless capture
- 5 minutes at 60 fps = 18,000 frames: 25–600 GPU-hours per take
Risks and unknowns
- The error rate may not fall with more data
- Body interaction (tackles, collisions) may cost more than expected
- Specialist availability; asset licence restrictions on AI use
- Sharpened only by V1's measured errors per minute, render seconds per frame and animator hours per second of video
Professional pilot (V3): ≈ USD 18,000–225,000 on top of V2, central ≈ 60,000
On top of V2: commentary from match events, TV graphics, editing, visual quality control, corrections, legal review, first official match. The lead engineer (role) handles event-driven commentary, video-to-record verification and the integrity protocol.
Effort: technical animator 160–640 h of corrections; editor and motion-graphics designer 40–160 h each; legal adviser 10–40 h. Risks: residual errors visible in replays; music and voice rights; publishing-platform policies. What would sharpen it: an audience test of V2 material.
Full 90-minute match: no defensible budget, only a model
A 90-minute match plus added time; a repertoire extended to rare events; production and verification automated where possible. There is no basis today for a defensible budget. What can be given is how the cost per match would be calculated.
Cost per match = render + storage + (errors per minute × 96 min × correction minutes per error ÷ 60 × rate) + editorial review
Render calculator
96 min × 60 s × 60 fps = 345,600 frames; cloud GPU at 0.89–4.89 USD/h| Seconds per frame | GPU-hours | Cost per match (USD) |
|---|
Errors per minute and correction minutes per error are unknown; they are measured in V1 and V2. The defaults reproduce v1.3's illustrative example only (2 errors/min × 10 min = 32 specialist hours, 1,600–4,800 at 50–150 USD/h). Editorial review and the extension to rare events are not estimated.
Where money could be lost, and the control for each
| # | Risk | Severity | Detected | Control |
|---|---|---|---|---|
| R15 | AI acceleration lower than assumed | Medium-high | Month 2 | Measure hours against milestones; plan on the central scenario; unfavourable scenario stated |
| R16 | Real spend exceeds the allocation (tax, extra usage, cloud) | Medium | Monthly | 1,200 ceiling with tax included; expense log with receipts; billing alerts; cut discretionary items or pause until written approval |
| R13 | All knowledge and the calendar depend on one person | High | Always | Continuous documentation in the repository; written contracts and decisions; machine reproducible from the repository |
| R1 | Legal but not credible football | Critical | Stage 2 | Scenario ladder; blind review; stop point before further spending |
| R2 | Animated bodies cannot follow the simulation | Critical | Stages 4a–5 | Feasibility study and a minimal sequence before paying for 22 bodies |
| R3 | Motion data with rights is missing | High | Stage 4a | Prior inventory; own capture budgeted |
| R4 | Cost per match unaffordable | High | Stage 6 | Measure before scaling (V4 model) |
| R9 | Animation specialist unavailable | Medium-high | Stage 4a | Early search |
| R10 | Licences unconfirmed | Medium | Before Gate C | Legal reading of the primary licence texts |
| R12 | The audience is not interested | Dominant, not technical | Stages 3 and 6 | Show real material early; not an engineering experiment |
| Other | Currency and tax | Low-medium | Monthly | Some plans are billed in euros; tax jurisdiction not fixed; both absorbed by the tax-included ceiling or cut lines |
Financial control. Funding by milestones; a first tranche of at most two months (2,400); extensions only after verified progress. Every stage has advance and stop criteria fixed before it runs, and a maximum effort in hours. The two technical stop points are stages 2 and 5. Numeric thresholds are recorded in the repository, dated, before each stage, and are not changed after seeing results.
Small tranches, verified milestones, no automatic renewal
The only decision that would open execution and spending is to authorise step 1a (create the isolated machine and demonstrate isolation, with no third-party code) and a first tranche of at most two months (USD 2,400). Step 1b needs a separate GO once 1a has passed. No gate renews automatically.
Funding timeline
Weeks from a future authorisationFavourable milestone weeks are from v1.3 §12.2. Central and unfavourable positions of H1–H3 are derived here by the same proportional scaling Annex F1 uses for milestone hours (central ≈ 395 h to H2, ≈ 790 h to H3; 40-week case ≈ 560 h and ≈ 1,125 h), at 45 h/week. Beyond stage 3, v1.3 gives no weeks: those depend on the visual investment and on measurements that do not exist yet.
How much weight each figure can bear
Market references were read on 14 Nov 2026 for the v1.3 planning basis. No new research was done for this document. Tags: [R] page read; [S] search summary only; [V] verified; [P] pending confirmation.
Strong
- US official statistics: BLS Occupational Outlook, software developers median base 135,980 (May 2025) [R]; BLS ECEC benefits share (June 2026) [R]; BLS animators median 102,030 [R]
- Freelance rate survey with more than 5,000 responses: Arc.dev, n=5,302, May 2026 [R]
- ITJobsWatch UK contractor and permanent rates, 6 months to 14 Nov 2026 [R]
- Provider price pages read: Claude, Hetzner, DigitalOcean, GPU clouds (Lambda, RunPod), Fab packs [R/V]
Weak
- Agency and studio rates (vendor blogs with commercial interest) [S]
- Per-character 3D model prices [S]
- Recruitment fees and time-to-hire, from recruitment-firm blogs [S]
- ChatGPT tiers, from April 2026 press; official page unreadable [S]
- Motion-capture studio day rates (a vendor blog, March 2026) [R, vendor]
- Lawyer hourly averages [S]
Missing
- A freelance rate for technical animators: visual costs use 50–150 USD/h (senior remote to senior global)
- Errors per minute and correction minutes per error (measured only in V1–V2)
- Render seconds per frame (measured in stage 5)
- Real hiring rates, motion-capture quotes, primary Epic licence texts
| Topic | Source | Date | Tag |
|---|---|---|---|
| Remote and global freelance rates | Arc.dev freelance developer rate survey (n=5,302) | May 2026 | [R] |
| Remote senior ML/Python, Ukraine | lemon.io rate calculator | 2026 | [S] |
| US developer pay | US Bureau of Labor Statistics, Occupational Outlook Handbook | May 2025 | [R] |
| US benefits load | BLS Employer Costs for Employee Compensation | June 2026 | [R] |
| US engineer base median | Levels.fyi | 2026 | [S] |
| UK ML contractor day rate | ITJobsWatch | to 14 Nov 2026 | [R] |
| Agency rates; recruitment | Vendor and recruitment-firm blogs | 2026 | [S] |
| Linux machines | Hetzner and DigitalOcean list prices | 14 Nov 2026 | [V] |
| GPU hourly prices | Lambda and RunPod on-demand pages | Nov 2026 | [V] |
| Workstation; pro GPU | Puget Systems configurator; vendor blog | Nov 2026 | [V] |
| Animation packs | Fab marketplace | 14 Nov 2026 | [R] |
| Motion capture | Studio-rate vendor blog; Rokoko; Move.ai | 2026 | [R]/[S] |
| Technical animator pay | Salary aggregator (summary); BLS animators | 2025–2026 | [S]/[R] |
| AI tool plans | Claude pricing page; press reports for ChatGPT | Apr–Nov 2026 | [R]/[S] |
What is not included, and what this document is not
Not included in the figures
- Taxes beyond the tax-included 1,200 ceiling; the billing jurisdiction is not fixed
- Company formation, accounting, and legal fees for any agreement
- GPU training for machine learning (not planned in the first phase)
- The cost of directing a hired engineer, and domain ramp-up time
- A costed two-engineer variant
- V4's recurring cost per match, editorial review and rare-event repertoire
- Marketing, platform, distribution and audience validation (no experiment designed)
- Currency movements for plans billed in euros
What this document is not
- Not a quote: every figure is a planning estimate or order of magnitude
- Not an offer, salary, invoice, debt or obligation of any kind
- Not a compensation or equity proposal: no form and no percentage is proposed
- Not legal, tax or accounting advice
- Not an authorisation to spend or execute: no budget authorises spending by itself
- Not a commitment to dates: weeks count from a future authorisation
- Not access-controlled: confidentiality depends on how this file is shared