The changeover
point.

The wider project includes a September 7, 2026 domain purchase, migration from my existing Adobe Portfolio, later site expansion and article distribution. I recall a couple of days for the initial migration; exact migration dates are unconfirmed. September 19–October 2 is the fourteen-calendar-date publishing sprint supported by the selected project records, not the entire website’s lifetime.

I estimate 30–40 hours of dedicated personal effort across the work, including 20–25 hours during the launch week and 1–2 hours producing and revising this case study. These are recollections, not timesheets. Earlier creative assets, industry experience and the original ICT software were existing inputs. Their original production is not priced as new work in this comparison.

The calculator’s compressed preparation period is a hypothetical schedule for the same effort. It divides preparation hours by an eight-hour workday and retains a five-business-day launch window. It does not rewrite the actual chronology or suggest that launch-week work happened instantly. Business days and calendar days are distinct.

The supplied Reddit Post Insights PDF and matching CSV contain ten EVGA-related posts across seven subreddits, published September 28–October 2. Their rows sum to 181,471 post views, 218 shares and 109 comments. The report’s overall 181,701 views also includes an unrelated July 2024 résumé post with 230 views, which is excluded here. Views can overlap across posts, crossposts and readers; they do not represent unique people or website visits.

The report is dated October 5, 2026. Its timezone and precise cutoff are unspecified, so that source date is preserved. It supplies a snapshot rather than a daily growth history. A retained row does not establish that a post remains publicly accessible. Upvote percentages are not averaged because vote denominators are unavailable.

The LinkedIn aggregate export covers September 26–October 2: 3,267 impressions, 2,346 members reached, 13 engagements and seven new followers. The September 28 EVGA post accounts for 2,993 impressions. Discovery totals reconcile with the daily and listed-post impression sums. “New followers” does not establish net growth because unfollows are absent.

On October 4 I also reported roughly 5,000 LinkedIn post hits during the week, about 140,000 website requests and about 10,500 website unique visitors. These newer figures are estimates supplied directly in the project conversation; no replacement export, website date window or provider definitions have yet been verified. They are shown as separate snapshots rather than added to the earlier export. Neither export separates paid and organic response or provides campaign-attributed conversions and revenue. Reddit views and LinkedIn impressions are therefore kept separate. The reports do not establish ad return, a deduplicated audience or a causal effect from AI or paid promotion.

The saved October 2 publication register records 55 outlets and separately published regional editions. Related editions and secondary coverage are included; this is not 55 independent investigations or a readership total. Examples checked on the publishers’ own pages are TechSpot, Club386 and Tom’s Hardware. The broader register is a saved snapshot, not a fresh verification of every entry.

The $759.70 provisional total comprises a full $100 monthly ChatGPT subscription, $384.32 in additional AI purchases, $22.16 domain registration, $1.50 Cloudflare payment, $173.72 LinkedIn promotion and $78 Reddit promotion. It combines documented purchases with my identification and allocation of expenses; it is not a complete bank reconciliation or a measure of credits consumed by this project.

The eight supplied AI transaction amounts total $380.32. An additional $4 ChatGPT purchase brings the AI-purchase subtotal to $384.32. Funding emails classify $80 as OpenAI API funding and $160.32 as ChatGPT credit reloads. The separate $140 charge has no matched invoice or verified credit quantity. The LinkedIn amount is supported by the supplied transaction screenshot and my campaign identification.

The $4 ChatGPT and $78 Reddit items appear together in an $82 App Store purchase record. A screenshot supplied October 4 confirms the line items: 100 ChatGPT credits for $4 and Reddit Promote Post for $78. The subsequent mention of an $82 charge referred to this same order, so the two line items are counted once, with no additional $82 expense. The Cloudflare invoice and corresponding payment are one $1.50 expense. The subscription notice corroborates the monthly amount without creating another charge. No extra tax or unrelated personal purchase is added.

Equipment, earlier Adobe subscriptions, original creative production and unrecorded services are outside this ledger. Personal time is valued separately in the calculator. Purchasing credits does not establish their opening balance, remaining balance or allocation among tasks.

The local audit covers 45 selected Codex chats and 89 descendant sessions: 134 sessions in total. Recorded activity runs from September 19 through October 2, Pacific time. The total is 1,722,537,435 tokens: 1,718,641,438 input plus 3,895,997 output. Input includes 1,660,741,339 cached tokens; output includes 804,051 reasoning tokens. Those subsets are not added again.

The processing total sums 13,001 unique response-level usage records, including 90 context-compaction responses. It reconciles to the earlier cumulative-counter audit of 1,700,537,307 tokens plus 22,000,128 compaction tokens. Compaction reorganizes context during long conversations and is real processing overhead. Cached input and reasoning output remain subsets; the separate cache-write field is not added to the total.

A separate October 4 snapshot adds this case study’s production: 54,095,193 tokens across this chat and three supporting agent sessions, through 8:56:54 p.m. PDT (October 5, 03:56:54 UTC). It contains 399 unique response records, including three compactions: 53,856,250 input tokens plus 238,943 output tokens, with 51,957,248 cached input tokens. No session or response IDs overlap the earlier audit. Work after the cutoff is excluded.

Across both measured phases, the arithmetic total is 1,776,632,628 tokens across 138 sessions: 1,772,497,688 input plus 4,134,940 output, with 1,712,698,587 cached input tokens (96.6% of input). This combines the launch work and later case-study production without placing the latter inside the earlier launch dates.

This is broad, whole-session telemetry. Some selected chats include other work, while unselected chats, the earliest migration, image generation, other devices and other services are not fully covered. Repeated context is part of the count; these are not 1.78 billion unique words. The retained ChatGPT export does not supply token or credit-consumption fields. The absence of later records is not proof of no later usage.

Purchase evidence documents 5,110 ChatGPT credit units, $80 of API funding and the additional $140 purchase with unknown unit quantity. These are different measures. The telemetry is not an actual bill, a complete account total or an exact project-only count. No API-price conversion is used to reconstruct subscription spending, and cached input is not assumed to be free. Turn durations include tools and waiting, so they are not presented as pure model-processing hours.

The lean, base and extensive scenarios assign 172, 296 and 454 person-hours to the full project: 144 / 244 / 368 for the website and launch, plus 28 / 52 / 86 for this case study. These are illustrative effort estimates, not measured industry averages, staffing requirements or supplier quotes. They price work across functions without assuming four full departments or a store, checkout and fulfillment system.

The extra case-study hours cover data and receipt reconciliation, economic-model research, writing, design, interactive build, QA and publication. Existing campaign reporting remains in the original launch allocation and is reused. No second campaign is assumed. The role table separates these phases so that the additional work is visible.

Hourly compensation equals the May 2025 U.S. annual median wage divided by 2,080, multiplied by 46.89 ÷ 32.82. Wage anchors are web developers ($92,650) and digital interface designers ($104,000), public relations specialists ($74,750), marketing and market research specialists ($78,760), and project management specialists ($102,320).

The benefits factor uses BLS June 2026 private-industry total compensation of $46.89 against $32.82 in wages, released September 9. It is a broad approximation rather than an occupation-specific benefits rate. The BLS technical notes explain the annual-hours conversion and wage exclusions.

Full-scope wages and benefits are approximately $10,396, $17,912 and $27,432. Both paths include the same $251.72 promotion budget and $23.66 infrastructure expense. The solo path includes $759.70 of total cash outlay, covering those shared costs plus AI purchases and the monthly subscription, then adds the value of all personal hours. The reported 30–40 hours already include the 1–2 case-study hours. The $82 Apple order comprises the existing $4 ChatGPT and $78 Reddit items, so it creates no extra charge. No token-price valuation is added to purchases.

Agency margin, office overhead and unrecorded team software are excluded. This is a partial-cost comparison. Tasks are assigned once, and waiting for a review does not automatically add paid labor hours.

Default solo inputs are 35 total hours, including 22.5 during launch week and 1.5 making this case study. Preparation is (35 − 22.5 − 1.5) ÷ 8 = 1.375 business days. Adding the five-day launch window and 1.5 ÷ 8 = 0.1875 case-study days gives 6.5625 modeled business days. Total effort remains 35 ÷ 8 = 4.375 workdays. Launch and case-study allocations cannot exceed the total.

Original team preparation uses the longest of three parallel paths, at six productive hours per specialist per day: (40% of design + web hours), (editorial + half of social hours), or coordination hours. The remaining design work is assumed to overlap the delivery path. Add two, four or six blocking reviews multiplied by the selected wait. Launch and campaign reporting take the greater of five business days or (remaining social + original analytics hours) ÷ six. Both approaches retain the same minimum observation window.

The case study follows that window. Its production path is evidence and economic analysis → the longer of writing or design → interactive build, QA and publication, each divided by productive hours per day. Coordination runs in parallel; the model uses the longer of coordination and that sequence, then adds one, two or three further review waits. At the base defaults: max((14 + max(10, 8) + 16) ÷ 6, 4 ÷ 6) + 2 = 8.67 case-study business days. These are scheduling assumptions, not measured team turnaround times.

The base team takes about 26.87 business days through launch and observation, plus 8.67 for the case study: 35.53 in total, versus 6.56 solo. The full-completion difference is 28.97 days. Public-launch readiness is a separate milestone: 17.87 preparation + four review days = day 21.87 for the team, versus day 1.375 solo, a 20.49-day difference. A focused team may work faster; different staffing, review authority or availability can change the result.

Earlier-public-launch value equals the difference in readiness days multiplied by an assumed incremental contribution per day. It excludes the subsequent observation and case-study periods. At the defaults, 20.4917 × $500 ≈ $10,246. The $500 is a sensitivity input, not observed revenue, salary or a forecast. Use contribution after relevant incremental costs and avoid counting the same benefit twice. Setting it to zero removes that hypothetical value. Negative differences remain negative. Adding it to the full-scope cost difference produces a scenario comparison, not earned savings or guaranteed earnings.