Every record ends with “true on recording day — verify before relying” notes. Here they all are — 756 claims across 209 records — oldest record first, so the ones most likely to have drifted are at the top. Prices, model names, limits, and ‘the latest’ anything belong here. The date is when the record was last written, not the recording date.
n8n UI updated just before the session (new AI assistant panel) — screenshots and menu paths will drift; workflow logic is durable.
'Opus 5 just released' (2:24:38) and the GPT Image 2 preference ('gives much better results') are point-in-time; image-model choices in this pipeline are the most perishable component.
Apify free-plan generosity, the actor's ~29k user count, and the 14-day n8n trial are as-of-recording commercial terms.
Model labels in the AI-node segment ('GPT 5.5', 'image 2') are ASR-blurred; treat exact versions as unverified rather than current fact. n8n's ~$2 built-in OpenAI credit allowance is an as-of-recording term.
The Ad Library's legal-transparency basis and the longevity heuristic are durable; everything about which model generates the image is not.
Interface names are flagged perishable by the source itself: Claude's Visuals toggles, Lovable/Bolt free tiers, Krea's free plan and unit estimates, n8n Cloud trial terms, LinkedIn's PDF-export menu path.
Workbook 5's model choice ('a low-cost capable model available at the time') and fraction-of-a-cent cost estimate are dated; the method — record price/date, set a ceiling — is the durable instruction.
The five build methods, checkpoints, test suites, and failure indexes are durable; every named product and price decays.
The roadmap is the May 2026 source edition and says so; every model name, benchmark, price, ranking, product surface and market claim is a labeled snapshot requiring verification — the record preserves the labeling, not the claims' currency.
Specific dated snapshots include: the April–May 2026 release table (Opus 4.8, GPT-5.5, Gemini 3.5 Flash, Composer 2.5, Kling 3.0…), the video leaderboard, the pricing cheat sheet (dated May 31, 2026), the Claude nine-surface table, and job-market/policy claims — all 'not independently verified for this portal.'
This is the portal's most durable part: definitions, taxonomies, diagnostic methods and safety boundaries — decay: durable nearly throughout.
Perishable residue: plan/feature availability claims (the 'feature shown in class is missing' guide exists precisely because these drift), product names in examples, and n8n/hosting pricing — all flagged by the guides themselves as requiring current verification.
The 'source: Day 1/Day 2 chat' attributions tie terms to the weekend event's two live sessions — provenance, not currency.
The portal text is itself dated by design: the twelve-tool directory is a May/July 2026 snapshot and says so; Session 5's model landscape is a June 2026 snapshot; both instruct the reader to verify current availability, pricing and rights.
Custom-GPT-requires-paid-plan, Claude-Projects-free, Gemini-Gems-limited are platform terms as of the source snapshot.
Frameworks are the durable layer: five-stage model, Magic Prompt, XML tags, assistant ladder, app anatomy, trigger-transform-decide-act, rule/LLM/agent/human, AI Director ingredients — decay: durable.
All product names in toolRefs decay: perishable; the job categories they instantiate decay: durable.
Source captured 2026-08-19 from a chatgpt.site-hosted SPA that may disappear; the V: archive is the permanent copy.
Model names as heard and time-bound: Opus 4.8 ('really good'), 4.7 ('not really that good'), 4.6 (execution tier), Fable ('the best one right now, costs a lot more'), GPT 5.5 / GPT 5.6 'Sol'. Version-check before reuse.
skills.sh figures as heard: 836,000+ skills, run by Vercel; shadcn skill at 230k+ installs; HyperFrames at 34k+ stars — all moving numbers.
'better-auth acquired by Vercel' is trainer speculation confirmed only by an on-screen banner ('joining Vercel') — verify before citing.
Effort menu described as six modes; slash-command surface (/btw, /goal, /usage, /permissions, /plan, /model, /effort) is version-dependent — his note that '/agent is removed' shows the churn.
Live costs quoted: $21 → $34 → $45 across the session (~1h40m API duration at the first reading) — useful as a magnitude anchor only.
Claude Notch is macOS-only at recording; a Windows port is 'vibe-codeable' per the trainer.
ClawdHub/OpenClaw injection incident and Anthropic trademark rename retold from memory — dates unstated; treat as folklore-grade until sourced.
Command churn documented live: /agents and /create-skill both removed between the trainer's prior use and this session; learners on newer builds saw a 'run skill generator' he lacked. Every slash-command claim is version-bound.
Models as heard: Opus 4.8 orchestrator with a 1M-token context window; Fable 5 present but swapped for Sonnet for doc reading; sub agents default Sonnet, options Sonnet/Haiku/Opus.
Per-model usage-limit tiering (most Haiku, less Opus) stated as plan behavior — verify against current plan docs before citing.
Context7 endorsed here and retired by session 14 one day later — keep both positions; the disagreement is course-internal evidence, not an error.
ElevenLabs free-tier commercial-rights limitation and Suno download flow are terms-of-service claims as of recording.
Token figures (16.3k system tools, 262.5k/26% session total) are machine-specific snapshots, useful as magnitudes only.
EXTREMELY point-in-time session by design: the 'graph' naming was ~24 hours old at recording; Agent Grid's plans/limits, /goal-/loop availability per CLI, superpowers, OpenClaw/Hermes/Paperclip, GLM 5.2, and the Thinking Machines 'inkling' early access are all as-of-July-2026 surfaces.
Karpathy quotes are dated deliberately (Feb 2025 vs Feb 2026) — the arc IS the content; the MIT '95% not making money' figure is cited secondhand, unverified.
Durable layer: the five-part loop, doer/checker separation, four loop types, hidden-test discipline, per-role model routing as a QUESTION, the four-file orchestrator pattern, harness components, rules-vs-hooks, different-model judging — these outlive every named product.
The trainer's product claims (NEM stats, grant, CV-TAS) are self-reported and dated; his 'Forge' harness was shown but not shared.
TITLE vs CONTENT: the LMS title promises app security; the delivered security model is agentic (tests, reviewer agents, guards, checkers). Recorded here so future searches for 'security' land correctly.
App-version-specific UI throughout: comment-on-plan, plus-icon slash menu, plugins browser, context-percentage icon — all desktop-app surfaces that move quickly.
Models as heard: Opus 4.8 (high effort) for planning/building, Sonnet for small edits, Fable named among 'smart' tier; plan tiers quoted at $20/$200, trainer on a $2,000/month budget.
Compact thresholds (40–50% manual, auto at 100%) are behavior claims of this build — verify against current docs before citing.
'Superpowers' and 'front-end design' attributed to Anthropic as plugin publishers — plugin provenance worth verifying when registering these tools.
Cowork ≈ Perplexity Computer comparison and the Emergent/Lovable equivalence are point-in-time market takes.
Auto mode described as stopping on destructive commands — permission semantics are version-dependent.
VERIFY BEFORE CITING: his claim that skill prompts are 'sent to Anthropic's team for their training' is stated flatly and drove his keep-IP-out advice. Training-data policies (especially for paid/enterprise plans) generally say otherwise — record as his claim, not fact; the keep-secrets-out advice stands on its own regardless.
Plan gating as stated: Claude Code needs Pro ($20) or Max ($200); he's on an enterprise plan. Tiers and gating move.
Fable 5 'released recently' timestamps the session; models named Fable/Opus/Sonnet/Haiku; first reply came from Sonnet.
The Shift+Tab mode cycle included an 'agents mode' in his build — mode sets churn version to version.
'2 million plus skills' on the marketplace and skills.sh's positioning are point-in-time figures.
Effort slider labels ('extra', 'max') are app-version-specific.
All dollar figures are trainer anecdotes as-of recording: $30k + $15k/mo chatbot deal, $6k P&L dashboard, $4k/mo phone reps, 11.5¢-style platform pricing not cited here but adjacent claims (Apollo, lease values $44k-100k) will drift.
The 391% five-minute statistic and the wine study are cited from memory without sources — display with as-of framing. (BC6 cited the same speed-to-lead stat as 21x — the two sessions quote different renderings of the same research family.)
'Anyone can build a SaaS overnight' and commoditization-speed claims ('3-6 months and the price goes to zero') are market-state assessments as of the session.
Promised future sessions (week 6 lead-gen masterclass, week 6-7 copywriting, 12 hours of outreach depth) are schedule claims — verify against the LMS as weeks land.
All referenced tools current as of extraction (2026-08-17).
Tooling preferences are explicitly volatile: 'if you would have asked me 2 weeks ago, I would have said Claude Code' — the Codex lean, the $100 tier, and the team migration are as-of recording.
Model claims point-in-time: GPT-5.5 tiers, Opus 4.7/4.8 references, 'ChatGPT image generation is better than Nano Banana at this point', Gemini video model naming.
Ecosystem drift certain: skills.sh install counts, motionsites.ai paywall state, 21st.dev catalog, Hallmark/taste-skill availability, the Vercel security incident ('now it is fine').
The skills-replacing-n8n market thesis and 'skills space getting commoditized slowly' are the trainer's market reads as of the session.
All demonstrated tools current as of extraction (2026-08-17).
The Fable lament is dated color: he built harshad.com with Claude's Fable model, lost access mid-second-page ('there's an issue with the selected model'), and repeats 'I miss Fable' — plus a garbled 'US banning Fable' aside. Point-in-time model-availability drama, amusingly extracted BY Fable.
Instantly pricing/UI is of-the-day: ~$15/domain pre-warmed, the $585 bundle quote, AirMail as a new untested provider, sender-name glitches during the live demo.
Sam Altman and VAPI-CTO anecdotes are personal receipts as told; the VAPI podcast episode was recorded but unreleased at session time.
skills.sh star counts (Humanizer 24k+, 698k skills) and .ai-domain price commentary are point-in-time.
Cameron's intent-signals session cross-reference left open — 'ask in the LMS'; F5Bot offered as his personal method meanwhile.
The SpaceX/X-acquires-Cursor claim, 'Grok will power Cursor going ahead,' and Composer 2.5-on-Kimi-K2.5 are all as-stated in-session (June 2026) — verify before citing; the KB records the claim, not the fact.
His tool rotation is explicitly month-stamped: Codex main, Claude Code fallback 'this month' due to reliability — expect drift.
Free-tier numbers are of-the-day: Scrape Creators 1,000 credits, Parallel's 'very good limits,' extract-design-system's 124k installs, yt-dlp's 174k stars.
Loop engineering deferred as unproven and token-hungry — a likely future-session topic if it matures.
Session 2 of the Cursor arc (next session) promises: the app-building harness, sub-agents, .env walkthrough, debugging/test skills, what-to-build framework.
Trainer taught through a viral infection with planned voice breaks — session pacing reflects it.
Model/platform claims are of-the-day (June 2026): Composer 2.5 on Kimi K2.5, Opus 4.8 for planning, the Sam Altman announcement aside, and another Fable lament ('it gets me') — all as heard.
Free tiers quoted: Clerk 500 user sessions, Neon free at project scale, Razorpay/Stripe live-site review requirements — verify at build time.
AI Edge shipped with known bugs (question repeat, missed popup) and auth bypassed; he promised the fixed Vercel link 'in the next few days' — the artifact may exist by now.
His 'vibe security will become very popular in the coming months' is a dated prediction; Semgrep's Replit/Lovable usage as stated.
Hermes deep-dive promised as a future masterclass; 'loop engineering' still deferred.
Trainer taught through fever, worse than the prior day — pacing and Q&A depth reflect it.
Session-time model claims (SWE-bench ~86%, 'GPT 5.4 at 400k context', 'Claude Opus 4.7') are point-in-time and shift fast — display with as-of framing, do not treat as current facts.
All demonstrated tools Current as of extraction (2026-08-16).
Dense point-in-time model claims: Claude Opus 4.7/4.6/4.1 pricing ($10/$25/$75 per M tokens), GPT-5.5 'extra high', Sonnet 4.6, Qwen 3.6 'on par with Opus 4.6', Gemma 4 variants, NVIDIA Nemotron 'updated a week ago', Tencent 'HY3' topping OpenRouter weekly rankings — all as-of recording, display with as-of framing.
skills.sh described as launched 'a couple of months ago' with 100,000+ skills; OpenRouter '400+ models' — counts will drift.
n8n product surface (AI assistant, build mode, native Firecrawl nodes, 10,000+ templates, 14-day trial) is current-as-of-session; local installs lack the AI assistant.
Mac mini M4 'under $600' and 16GB base RAM claims are point-in-time retail facts (learner corrected trainer's 18GB claim live).
All demonstrated tools current as of extraction (2026-08-16).
65 questions, 5 unscored, 2 hours, 70% to pass, in person, no endorsement. - STATUS: CURRENT as taught and long-standing, but confirm against the FAA testing page before booking.
About 700 FAA testing centres nationally. - STATUS: APPROXIMATE and changes. Check the current PSI/FAA locator.
Written exam at 14, certificate application at 16, result valid two years. - STATUS: CURRENT as taught; verify against the present rule.
Pilot Institute pass rate over 99.9%, fewer than 100 failures, 0.03% failure rate, 55,000-plus (and elsewhere "hundreds of thousands") trained. - STATUS: MARKETING CLAIMS from the vendor, self-reported and internally inconsistent between lessons. Not independently verified.
Magnifying glasses may be refused by proctors. - STATUS: The instructor confirms he raised this with the FAA and that PROCTOR DISCRETION is final. Plan accordingly.
Figures increasingly shown on screen rather than in the printed supplement. - STATUS: TRANSITIONAL at time of recording. Expect on-screen figures; still practise with the supplement.
Platform details - Copilot AI, Quizbank, community app, offline download. - STATUS: VOLATILE product features, and the lessons here (9053-9061) are visibly newer recordings than the course body. Verify against the live platform.
Mavic 2 endurance figures (32 min claimed, 25-27 min achieved). - STATUS: POINT-IN-TIME product detail. The teaching point - manufacturer figures are laboratory best-case - is stable.
Testing supplement structure and figure numbering. - STATUS: Verify against the CURRENT FAA testing supplement; appendix and figure numbers are revised periodically.
Remote ID compliance begins September 2023 and "only one drone" is approved for operations over people. - STATUS: POINT-IN-TIME. The course was recorded before those dates. Remote ID deadlines have passed and the Declaration of Compliance list has grown. Re-check the FAA DOC list rather than relying on the count given here.
FRIAs will become available "later in 2022". - STATUS: POINT-IN-TIME. FRIAs exist now; verify current application status.
The FAA Administrator is Dixon. - STATUS: OUTDATED. Personnel reference only; it does not affect any rule.
Session ID for remote ID has "no method to make that happen" yet. - STATUS: POINT-IN-TIME as of recording. Verify current capability.
Category 1 over an open-air assembly requires meeting remote ID "before September 2023". - STATUS: POINT-IN-TIME framing. The underlying requirement stands but the transitional dates have elapsed - read the current rule.
The instructor states the exam asks for both 87 knots and 100 mph. - STATUS: STABLE as regulation, but exam emphasis is anecdotal.
Local magnetic variation is about 11 degrees at Prescott. - STATUS: DRIFTS. The instructor says so himself; treat any specific variation value as point-in-time.
The Chart Supplement example is the November 2018 Southwest edition. - STATUS: EXPLICITLY OUTDATED and used only to demonstrate structure. Always download the current cycle.
NOTAM retrieval URLs (pilotweb.nas.faa.gov, fltplan.com). - STATUS: VERIFY. FAA NOTAM delivery has been reorganised since recording.
"Notices to Airmen" - STATUS: The FAA has since renamed NOTAM to "Notice to Air Missions". The acronym is unchanged.
Flight Service Stations have been closing and are less relevant. - STATUS: DIRECTIONALLY TRUE and continuing. Verify current Flight Service offerings if you intend to rely on them.
Specific frequencies quoted (Prescott 125.3, UNICOM 122.95, MULTICOM 122.9). - STATUS: MULTICOM 122.9 is a national standard and stable. Airport-specific frequencies change - always read the current chart or Chart Supplement.
Remote pilots may not transmit without an FCC licence. - STATUS: STABLE - this is FCC rather than FAA regulation.
About 700 airports and roughly 80% of controlled airspace are in LAANC. - STATUS: POINT-IN-TIME as of recording. Coverage has expanded - check the current UAS Facility Map.
Kittyhawk and Skyward are the recommended LAANC providers. - STATUS: VOLATILE. The LAANC provider market has changed repeatedly; verify which providers are currently FAA-approved before relying on either name.
AirMap is unreliable for stadium TFRs. - STATUS: POINT-IN-TIME opinion from the instructor. Treat as a caution about relying on any single app rather than a current product assessment.
One depicted zipper-line area "doesn't exist anymore" and is now 1,200 ft. - STATUS: Illustrates that chart data changes. Always use a current chart.
Specific Class B ceilings - Phoenix 9,000, Miami 7,000, Orlando 10,000. - STATUS: Verify against a current sectional; the teaching point (they are NOT all 10,000) is what matters.
Stadium TFR NOTAM number FDC 7/4319. - STATUS: NOTAM numbers are reissued. The 3 NM / 3,000 ft AGL / one hour either side / 30,000 seats parameters are the stable content.
Sectionals are valid 56 days, having previously been six months. - STATUS: CURRENT as taught, but confirm the present cycle - chart cycles have been revised before.
Specific chart examples (Phoenix, Miami, Jacksonville, Las Vegas sectionals). - STATUS: Chart data changes every cycle. Use them as technique examples only, never as current airspace information.
The instructor's 2008 paper Miami sectional is used as a prop. - STATUS: Explicitly historical - he notes the information has changed but the format has not.
Terminal Area Charts exist only around Class B. - STATUS: STABLE, but the specific list of TAC coverage areas changes.
Standard atmosphere values, lapse rates, fog classifications, cloud families, thunderstorm stages and front types. - STATUS: STABLE physical science; unchanged and not subject to regulatory drift.
The density altitude and pressure altitude formulas. - STATUS: Approximations taught for intuition. The instructor states twice they are NOT required for the exam. Use a published density altitude calculator operationally.
"2 degrees per 1,000 ft" lapse rate. - STATUS: Standard approximation (actual ICAO value is 1.98 C). Fine for the exam.
Hover is a dying app; Kittyhawk is recommended. - STATUS: OUTDATED. The weather/LAANC app market has changed considerably. Verify current FAA-approved LAANC providers.
TAF issue times and validity (4x daily, 24-30 hours). - STATUS: STABLE.
METAR and TAF code structure. - STATUS: STABLE international standard.
"Notices to Airmen" terminology used throughout. - STATUS: The FAA now says "Notice to Air Missions"; the acronym NOTAM is unchanged.
Advisory Circular 00-45H is the current revision. - STATUS: VERIFY - AC revision letters advance. The document title is stable.
The last US commercial airline crash was Buffalo, New York, in 2001. - STATUS: INCORRECT AS STATED. The Colgan Air Flight 3407 crash near Buffalo was February 2009. The teaching point (US commercial aviation is extremely safe) stands.
80% of aviation accidents are human-factors related; ADM training reduces judgement errors by 10-50%. - STATUS: Long-standing FAA figures repeated in the PHAK. Directionally stable; treat the precise percentages as approximate.
Hazardous attitudes, antidotes, IMSAFE, PAVE, DECIDE, 5 Ps, 3 Ps. - STATUS: STABLE - these are fixed ACS/PHAK constructs and are examined verbatim.
Night operations without a waiver, effective 6 April 2021, subject to knowledge and lighting requirements. - STATUS: CURRENT as taught and still the governing framework, but confirm against the present rule text.
Night airspace authorization must go through FAADroneZone because LAANC does not support it. - STATUS: LIKELY OUTDATED. LAANC coverage has expanded. Verify before assuming a DroneZone request is required.
Kittyhawk app behaviour for after-hours airspace. - STATUS: OUTDATED. The app landscape has changed; the underlying LESSON - cross-check more than one source, apps may not show after-hours reversion - remains sound.
Specific airport examples (Prescott, Winslow) and their tower hours. - STATUS: POINT-IN-TIME. Always read the CURRENT Chart Supplement entry.
Dark adaptation timings, eye anatomy, illusions, nav light convention. - STATUS: STABLE physiology and international lighting standards.
2.4 GHz carries video and 5.8 GHz carries control. - STATUS: UNRELIABLE AS STATED, and the instructor hedges it himself. Band assignment varies by manufacturer and model; check your own aircraft's documentation.
DJI drones have no maintenance programme. - STATUS: POINT-IN-TIME. Manufacturer support has evolved; check current documentation.
DJI TB50/TB55 battery cell issue. - STATUS: HISTORICAL example, resolved by firmware/hardware updates at the time.
No flight-hour requirement exists for remote pilots. - STATUS: CURRENT as taught. The instructor speculates one may appear; verify against current rules.
FAA Order "91 30.34D" reference. - STATUS: PROBABLY GARBLED. Verify the order number before relying on it.
Exam costs $160 ($96 at a PSI-owned centre plus a $64 third-party fee). - STATUS: OUTDATED. FAA knowledge test fees have increased. Check current pricing at booking.
PSI is the only FAA testing provider. - STATUS: POINT-IN-TIME and the instructor says so explicitly. Verify the current provider.
Topic weighting percentages. - STATUS: Published in the ACS and reasonably stable, but confirm against the CURRENT ACS.
Practice bank of about 350 questions. - STATUS: POINT-IN-TIME vendor detail.
Figures are only in the paper supplement. - STATUS: CHANGING. The Course Introduction section (a newer recording) states figures are INCREASINGLY shown on screen. Prepare for both.
King Schools practice exam is poor and draws on private pilot questions. - STATUS: THE INSTRUCTOR'S OPINION at time of recording, about a competitor. Treat as such; verify independently before relying on it.
14-day wait after a failed attempt. - STATUS: CURRENT as taught, consistent with 107.71.
The exam costs $160. - STATUS: OUTDATED. FAA knowledge test fees have risen. Check the current fee with the testing provider.
Temporary certificate in 5-7 days, permanent in 30-40 days. - STATUS: POINT-IN-TIME processing estimates. Directionally useful; actual times vary.
Recurrent TEST eliminated 1 March 2021, replaced by free online training every 24 calendar months. - STATUS: CURRENT as taught. Verify against the present rule before relying on it.
The IACRA screen flow and field names. - STATUS: UI CHANGES. The sequence is stable in substance but screens are revised; expect differences.
"Hopefully by August of 2018, everybody will have LAANC across the country." - STATUS: CLEARLY DATED - this lesson is among the oldest in the course. LAANC coverage has since expanded substantially.
AirMap recommended as a LAANC provider here. - STATUS: CONTRADICTS the Airspace section, where the instructor explicitly advises AGAINST AirMap for missing stadium TFRs. The Airspace lesson is newer. Verify current FAA-approved LAANC providers.
"Section 336 dashboard" for recreational flyers. - STATUS: OUTDATED TERMINOLOGY. Section 336 was repealed and replaced by 49 USC 44809.
Registration is $5 per aircraft under Part 107. - STATUS: CURRENT as taught; verify the fee on FAADroneZone.
The instructor's E4 / E5 mix-up when describing the Prescott extension. - STATUS: ERROR IN THE LECTURE. Extensions off Class D are E4, off Class C are E3. Trust the Airspace section, which teaches this correctly and at length.
The two workarounds are living exploits: the PhantomBuster DevTools .csv trick ('unpatched 3 years') and the Apollo Instant-Data-Scraper export — either may die at any patch; the AnyMailFinder free methods already did.
Clay claims are point-in-time: Anthropic/OpenAI/Cursor GTM usage, ~2024 inception, the LinkedIn partnership hiccup, provider-auction speculation (explicitly the trainer's guess), export limitations.
Model picks and failures are of-the-day: Gemini Flash for site analysis, GPT-5 mini persona pass, Sonnet 4.6/Haiku availability, the Anthropic credit error.
LinkedIn's ~15% annual job-change stat and the '3x better match rate' figure are as-cited from LinkedIn reports and Clay case studies respectively.
Happenstance maturity ('tested two weeks ago, not great yet') and the cohort's free subscription are dated facts.
All demonstrated tools current as of extraction (2026-08-19).
Instantly UI is visibly in flux: AI-enrichment buttons 'were not present before,' the super-search interface changed mid-session, a sidebar glitch ate his pixel snippet, and lead-list views misbehaved — expect drift from these exact clicks.
Pricing/credits are of-the-day: ~1.5 credits/row work-email enrichment, 5 credits/reply agents, Nano plan ~100 credits (exhausted live on stage).
Model references are dated: GPT-4.1-mini recommended for enrichment, Claude Opus 4.8 credited for the description, plus another 'I miss this model' Fable aside.
His /consulting and /coaching pages went live DURING the session and he announced his own campaign launching — the artifacts may have evolved since.
Claude-to-Instantly integration status ('I don't think they have a plugin/MCP yet') is point-in-time; he promised to check the MCP server.
Q&A deferred to office hours (Niharika) — expect Session 8 follow-ups in the next office-hours extraction.
GitHub pricing (free: 500 MB package storage; Team: $4/user/month, 2 GB) is as-of-recording; the free-tier file-size limit was genuinely unknown to the trainer ('15 MB or 100 MB, I think') — verify before relying on it.
'AWS has about 60% global market share' is the trainer's undated figure — treat as illustrative, not current fact.
Vercel's per-branch project flow reflects its UI at recording; the branch-per-environment pattern itself is durable.
Git/GitHub fundamentals (diff tracking, branches, PRs, forking) are the most durable content in the course — decay: durable across the board.
Lovable's auto-commit behavior and its private-by-default GitHub repos are product behaviors that can change.
Lovable's product state is mid-transition in this session: Supabase partnership replaced by Lovable Cloud, external-Supabase flow degraded — expect this UX to have changed again; treat all Lovable mechanics as as-of recording.
Model/pricing claims are point-in-time: 'GPT 5.5', Claude Code $100/$200 subscriptions, Codex momentum, Cursor's fast model, Antigravity's status.
Lovable prompting guide dated 2025-01-16 and already described in-session as aging ('a bunch of these principles are still valid').
All demonstrated tools current as of extraction (2026-08-16).
Product-surface claims are fast-moving: Claude connectors directory contents, desktop-restart requirement, Control Chrome being Mac-only, single-account connector limit, ChatGPT apps catalog and one-tool-per-chat limit, Gemini lacking connectors — all as-of recording.
Model references (Haiku 4.5, Sonnet 4.6, 'Claude 3/3.5-era' hallucination anecdotes) and reliability scores (Claude 7/10 vs ChatGPT 4/10 connectors) are the trainer's point-in-time experience.
Smithery listing stats (74/100 rating, ~40k users) and the Composio-hosted Google Sheets MCP URL will drift or break.
MCP-history simplifications ('tool calling first introduced by Anthropic') are the trainer's framing, not verified history.
All demonstrated tools current as of extraction (2026-08-16).
Model claims are point-in-time: GPT-4.1-beats-GPT-5.1 for voice, Gemini 3 Flash pricing/speed, Claude cost — re-test before relying.
Retell feature surface is fast-moving: alerting, AI quality assurance, iOS/Android call-screening response all described as new; batch calling dead due to then-current Western AI cold-calling rules (regulatory landscape shifts).
Pricing figures (11.5¢/min demo agent, 4.5¢ under 3,500 tokens, $10 signup credits, latency numbers) are as-of recording.
Statistics quoted from memory (>$100B missed-call losses, 21x five-minute contact multiplier, Gartner 90%-by-2027) are trainer-cited and partly self-doubted — display with as-of framing.
All demonstrated tools current as of extraction (2026-08-16).
Builder and platform names (Lovable, Bolt, Replit, Custom GPT builder, Claude Projects) are as-of-source options; every project explicitly says to find 'the closest equivalent' when interfaces change.
The safety architectures — draft-only mail, label scoping, injection preambles, idempotency keys, deterministic scoring, evidence rules — are durable; they are the lab's actual content.
Model-node output envelopes (e.g. $json.output vs message.content paths) drift with n8n versions; the projects instruct mapping from real execution output, which is the durable practice.
PLAN TIERS ARE LIVE AMMUNITION: the Essential-plan limits (no custom fields/CPT/CSS/Woo) were current at recording and explicitly newer than the course's start — re-verify against elementor.com/pricing before acting.
The 'new editor' was alpha at recording and expected to become default — UI-location advice is version-bound both directions.
Elementor Hosting judgment ('not yet — keep an eye on this page') is explicitly provisional.
All hosting prices quoted are promo/first-year figures; Hostinger 10% code may expire.
Discord was mid-migration from the Facebook group — channel structure 'still figuring out.'
82 of 83 course lessons have transcripts; the one gap is outside this section (flag when reached).
UI split runs through the whole chapter: old vs new theme builder AND old vs new editor toolbar — every 'over here' is version-dependent, and Elementor 'is changing their interface' mid-course.
ACF's post-types feature was new enough that his prior recommendation (CPT UI) is retracted in-course; the ACF free/Pro boundary (gallery = Pro) may move.
Two 'Reno from the future' retro-updates (custom permalink rescue; prepend/append) mean episode content is deliberately non-frozen — expect further silent updates.
The 999→empty items-per-page change is itself a documented behavior change; cap stated as 1000.
Gutenberg competitiveness and WooCommerce auto-category behavior are point-in-time observations.
Container status was 'release candidate' at recording with full rollout expected — the stability caveats and default-boxed annoyance may both be resolved for later viewers.
The Grid (CSS grid) container was beta and explicitly deferred — 'the next version of the container which is coming'; grid-based episodes were promised as updates.
The responsive-shadow gap and icon bottom-space gap are vendor-fixable any release — both workarounds should be rechecked.
An unnamed upcoming Elementor feature ('part of the stealth app', as heard) was said to add fit-to-icon sizing.
Breakpoints: 3 default, up to 6 available, his setup runs 4 — counts are configuration- and version-dependent.
The rem-based responsive padding system is teased for a later chapter — this chapter's pixel padding is transitional.
Elementor versions on screen: plugin 3.15 vs a 3.8 Pro zip — his point stands that version skew at install time is fine, but every settings screen shown will have drifted.
The experiments page 'updates all the time': container/grid/top-bar defaults are version-dependent; l1498093 already shows the global-colors panel improving mid-course.
The classes feature is Elementor's own roadmap item — 'I heard they are working on a classes feature' — which would retire both CSS snippets when it lands.
ASE was 'kind of new' at recording with an active feature-request channel to the developer; its option set will have grown.
ManageWP free tier: monthly backups at recording; tier boundaries move.
Gutenberg vs Classic Editor is a moving argument — the 5M-install datum is a 2023-era snapshot.
InstaWP/TasteWP expiry windows moved even during recording (2 days vs the 6 shown) — treat all ephemeral-install lifetimes as current-offer-dependent.
Hostinger's 100-subdomain second tier and 'cloud startup' package are 2024-era plan shapes; tiers rename constantly.
The optimizer-plugin picks are explicitly provisional ('at this time') and host-dependent; the role-limiting plugin pick is withdrawn pending a promised YouTube video.
Independent Analytics was under active personal testing, not a settled recommendation.
Elementor UI drift visible twice (role manager relocated; maintenance mode under Tools) — find features by name, not position.
Loom free-tier limits quoted from memory (5 min, ~25 videos) — verify current tiers.
The whole chapter is a dated capability snapshot by design — carousels, the new menu, Woo widgets, and text stroke all arrived 'recently' at recording; several add-on recommendations expire as core catches up (his own Crocoblock verdict demonstrates the decay).
The new menu widget may need enabling under features/experiments 'depending on the time you're watching.'
Overlay responsiveness, form-steps required-field behavior, and new-menu archive URLs are all vendor-fixable gaps — re-check per release.
Woolementor has since rebranded (CoDesigner); Happy Addons and Crocoblock pricing/tiers move.
Two standing Discord asks: dynamic archive URLs in the new menu; a cleaner mobile-menu pattern.
Every price in the chapter moved during or since recording (SiteGround cloud up on camera; ManageWP add-on pricing; Solid Security tiers) — the layer logic outlives all of them.
The lazy-load feature was 'kind of like a new feature' — likely default-on in later Elementor versions.
His site rebuild (CPT UI + ACF + JetEngine → one system) was pending; the JetEngine lean may since be visible on livingwithpixels.com.
The companion design and business courses were waitlist-only at recording — both may now exist (the LWP Design course harvest, OI-078, confirms the design course shipped).
reCAPTCHA versions (v2/v3) and the Elementor integrations flow are Google-API-drift-prone.
HIGHEST-DECAY SECTION OF THE COURSE. Recorded before Remote ID's Sept 2023 enforcement - that date is now past; verify current compliance state and FRIA availability
OOP-approved drone list ('only the eBee X') was explicitly expected to grow - re-check the DOC list before any over-people work
FAA Administrator named 'Dixon' (as-heard) at recording - leadership changes
Dollar figures ($5 registration, $500 FAA / $25,000 NTSB thresholds) are rulemaking-adjustable
Recurrency regime (free online, since 3/2021) replaced the $160 test - watch for further changes
Session-ID Remote ID 'no method at the moment' and module-BVLOS predictions were open questions at recording
The dead 5-mile airport rule is cited as dead - itself worth reconfirming against current ecfr.gov text
LAANC coverage (700+ airports, ~80%) and 10-second turnarounds are recording-era stats; the program has only expanded
Provider landscape shifted: Kittyhawk became Aloft; AirMap has since shut down; B4UFLY was later rebuilt through partner apps - all app judgments are dated
Grid colors change ('as of about a week ago... it's actually green now' - his own example of staleness)
Stadium TFR NOTAM FDC 7-4319 parameters could be reissued; Disney TFRs persist but verify
All specific tops/frequencies (Phoenix 9,000, W-174G times, Albuquerque Center 134.45) are chart-snapshot data
PSI-only administration, $160/$96 pricing, and ~350-question practice pool are recording-era snapshots
The bonus videos are OLDER than the core course: 'hopefully by August of 2018' LAANC rollout and AirMap (now defunct) date them - the WORKFLOWS hold, the UIs and providers have moved
IACRA screen order and field labels track the FAA site's UI at recording
Currency regime (March 2021) is the current framework - watch for further rulemaking
King School criticism is a dated vendor assessment
The workspace tool ('Macro', ~4,000 stars, 'explored literally just last week') and the harness ('just came out') are bleeding-edge - names unverified, landscapes shifting; verify before adopting
Praful's ~50% subscription saving is one user's testimony at recording
Web MCP is hackathon-stage; Harshit's own verdict pending his test
Drive-embed posting is NEW for Catalyst recordings - the harvest pipeline now handles it (download + local Whisper), but watch whether it persists
Product Hunt launch session (Session 25 slot) announced for the following weekend
AS HEARD, UNVERIFIED: 'GrokBot has been built by the Cursor team... Cursor got acquired by SpaceX'; Grok 4.6 'trained with Cursor data', Grok 4.7 'releasing soon, trained with SpaceX engineering data'; coding bots = Cursor Cloud Agents
GrokBot is in beta - 'it worked three days back, now they changed the back end and it is breaking'; connector set, memory behaviour and group chats may change
Prices at recording: $30/month floor, Cursor Ultra $200/month bundle; weekly usage meter
Secondhand stat relayed by the newsletter bot: 'Spotify cut Claude Code tokens 90% by routing predictable generation to Gemini Flash' - unverified
$498 / $198 (60% off) 'for a limited time'; '$30,000 per week' vs '$129,000 first month' for the same ad; Next Level Creators '$2M' vs '$3M over two years'; 'over 4,000 students' - all as of recording and internally inconsistent in the live take
Feedback channel named as the Facebook group at recording
AS HEARD, RECALLED NOT VERIFIED: the CSCW hackathon study (~2020, ~590 projects, 35% / 17% / 3.5% activity at event, six days, five months) and the 'MIT 2025 report' (95% of 300 AI projects with weak P&L impact). The speaker hedges the second with 'I think'. Confirm both sources before reusing the numbers with a client.
Programme specs quoted at recording: Generative AI Engineering Fellowship 6 months / 100+ live hours / 50+ office hours; AI Catalyst 6 months / 70+ live sessions / 24+ office hours; both stated NSDC approved; Catalyst stated to require a prior Outskill accelerator or fellowship. Cohort-specific, expect change.
Operational claims: recording posted within about 12 hours; support ETA 24 hours at hi@outskill.com; next session 'tentatively' Tuesday 7:30pm IST.
Captions begin at 0:15 - the opening ~15 minutes of the recording are absent from the transcript, so nothing before 0:15 is represented here.
Model names and capabilities are as of recording: Nano Banana Pro at 9:16 / 2K / three variations, Kling 3.0 five-second clips, HeyGen 9:16 / 1080p, Seedance 2.0 first-frame/last-frame. PromptEdit.com described as subscription-free pay-as-you-go, and as NOT supporting dubbing - all four are the kind of thing that changes quarterly.
'Mr Beast's Spanish channel has over 26 million subscribers' - stated at recording, will be wrong by now.
Blackmagic Camera described as 'by far and away the most highly recommended camera app on the Internet', free on iPhone and Android; CapCut's free desktop tier described as including auto-captions.
The community platform is named 'Content Creator Machine'; the migration off Facebook groups is presented as recent.
App prices as of recording: Filmic Pro moved from a single $14.99 payment to $2.99/week; BeastCam on sale at $1.99 and 'typically under $10'; Pro Camera by Moment $6.99; ProTake about $20/year; Manual Camera $4.99 one-off. Blackmagic Camera free. All volatile.
Hardware named as current at recording: iPhone 15 Pro Max as 'the newest smartphone', Galaxy S23 Ultra as the Android flagship, DJI OM 6 as 'newly released'. Godox SL60 'just a little over $100'.
Business figures are undated sales narrative: over 100,000 students across the catalogue, 35,000 in this programme, $10M generated over four years, ~600,000 views/month yielding ~$4,500 AdSense, $4-5k/month affiliate, a $3,000 first brand deal, over $400,000/month in own products, a $69 first ad spend returning 4x. None verifiable here.
'23.978 fps' as heard is the well-known 23.976 (24000/1001) - the transcript has the number wrong, not the instructor necessarily.
Every phone named is superseded: iPhone 13 Pro Max, Galaxy S21 Ultra and OnePlus 9 Pro are presented as the current flagships at about $1,000, $1,000 and $900, and the lesson even anticipates the S22 Ultra's release. A later addendum lesson covers the iPhone 17. Use the six criteria, not the models.
Dynamic-range figures stated as fact: Sony A7S III 16 stops, the human eye 'around 20 stops', Canon EOS R about 12.5, newest iPhones about 14. Unverified here.
Prices as heard: PoP Voice Pro lav $14 on Amazon; diffusion panel about $20; Yongnuo YN300 II under $100 at 2,280 lumens; Godox SL60 just over $100 at 4,500 lumens; Moment lenses about $100 each; smartphone torch about 50 lumens.
Gimbal specs as heard: Insta360 Flow Pro about 10 hours of battery against roughly 6 for the DJI OM6; DockKit tracking is iPhone-only; Android loses resolution and frame-rate control inside the DJI Mimo app.
The Day 3 gear links are affiliate links on Kit.co, described as updated weekly - the picks are commercial as well as editorial.
The DJI OM5 and OM6 are presented as the current recommendations; DJI's line has moved on. Mode logic survives, model numbers do not.
Android behaviour inside DJI Mimo - varying by manufacturer, 30fps lock, install from DJI rather than the Play Store - is tied to the app version at recording.
The '70% of Hollywood shots' figure is stated without a source.
Prices as heard: Yongnuo YN300 about $68-70; PoP Voice Pro lav under $15; Apple Lightning adapter about $15 ('highway robbery from Apple'); five-in-one reflector about $20; C-stand about $90; Westcott Solix Apollo about $400; Godox SL60 'a couple hundred dollars'; RODE Wireless GO II roughly $200-300; LG 40in ultrawide about $1,500; HDMI capture card about $20.
The studio tour is explicitly a 2024 snapshot and says so in its title. The M2 MacBook Pro is presented as top-tier; Apple silicon has moved on. Riverside is mentioned as something they were 'thinking about testing' rather than using.
iPhone 14 and iPhone 13 Pro Max are the demo phones, with Anthony joking the 14 will be outdated 'in like five months'.
Software UI descriptions (Filmic Pro screens, Premiere panel layout, Voice Memos) match the versions at recording.
Software pricing as heard: CapCut about $10/month standard and $20 Pro; DaVinci Resolve Studio a one-off $295; Premiere Pro about $22/month alone or $66/month for Creative Cloud. The CapCut terms-of-service licensing controversy is discussed as recent.
The four-pillars lesson pins itself to about January 2022 - it treats Netflix's Don't Look Up as current and dates a Canon R5 review to autumn 2020.
The computer buying guide discusses Apple M1 and M2 as current, with used-market examples (an iMac Pro at $900, a 2019 MacBook Pro at $600) and new MacBook Pro configurations from $1,700. All of it will age fast.
Content Creator Machine quoted at $75/month or $750/year for members against $97/$970 list, alongside a claim of previously paying 'over $24,000 per year' for ActiveCampaign. contentcreatortemplates.com is described as being on a limited-time price.
Editor scores are one instructor's opinion at one moment, not a benchmark.
App tiers as of recording: CapCut's free tier is stated as sufficient, with many caption styles Pro-badged; VN's free tier is stated as having no forced watermark and a version-specific maximum track count. Both are exactly the kind of thing these apps change.
The CapCut terms-of-service content-licensing controversy is discussed elsewhere in the course as recent - worth checking current terms before recommending it to a client.
Platform claims: Instagram supporting 16:9, 1:1, 4:5 and 9:16, and an approximate 200MB upload threshold before heavier compression.
Hardware note tied to a generation: 'one thing they introduced with the new iPhone 13 is a 60 frame per second refresh rate on the phone', used to justify a VN preview setting.
Epidemic Sound quoted at about $30/month depending on tier.
CapCut's free and Pro split as of recording: optical-flow slow motion in the free tier (which surprises him), 'remove filler words' and some animation presets behind Pro. Expect movement.
The free Apple Log LUT and the student discount at contentcreatortemplates.com are time-bound offers attached to this course release.
Grading examples are tied to a Sony A7S III in S-Log3 and an iPhone 15 Pro Max in Apple Log; a different camera's log profile needs a different LUT or different curve points.
LumaFusion UI descriptions and its '500 page user manual' reference match the app version at recording.
The template marketplace is pitched throughout as 'the equivalent of like two cups of coffee per month for unlimited commercial use' against competitors 'upwards of 30 to 50 to even $100 per month' - an undated price claim repeated verbatim across lessons.
'We currently spend about $2,000 a day on paid advertising... promoting our 14 day filmmaker course' - a specific, time-bound figure.
Hardware as current at recording: Samsung Galaxy S24 Ultra with a ShiftCam anamorphic lens, DJI Osmo Mobile 6, iPhone 13 and 13 Pro, a Canon EOS R shooting C-Log.
The transcript and Pause tools are described as a recent Adobe change with no version number given; the UI has moved since.
Adam Craig is described as 'an Olympic biker' - unverified, and possibly the instructor's own error rather than a transcription one.
The two-factor algorithm description is a simplification of a system nobody outside YouTube can inspect; the channel benchmarks quoted - above 3% CTR, 30-40% average watch time - are his own numbers at recording, not published standards.
Recording date is pinned to roughly 2022 by the gear talk: the iPhone 13 Pro 'coming out last fall', the Galaxy S22 Ultra just released, the Sony A7 IV 'just recently announced'.
Prices as heard: RODE Wireless GO II $299, the DJI wireless about $329, PoP Voice Pro lav under $15, Westcott about $400, Godox about $130, Yongnuo about $70.
Thumbnail advice about the bottom-right corner depends on YouTube's current interface, which moves.
Money figures are undated sales narrative: a student on a $10,000/month car-dealership retainer for nearly two years; a two-day shoot with three edits for $3,500; about $700 of housing saved on the Airbnb trade; a '$7,000 minimum' attributed to experienced creators; a $5,000 bet with the business partner over the car-wash commercial. None verifiable here.
The Captions app at about $9/month, and an export recommendation of 1080p at about 15Mbps, are both tied to the moment of recording.
Apple Log is described as introduced on the iPhone 15, and the commercial settings are built around an iPhone 15 Pro Max.
Auto-reframe comparisons - DaVinci free and accurate, Premiere losing the subject, CapCut behind a paywall - reflect the app versions tested and will move.
The salad analogy contains a transcription contradiction: option one is given as '$1,000' and then immediately as '$10'. The intended figures are almost certainly $10 of ingredients against a $15 pre-made salad.
Retainer pricing is presented as the instructor's own historical guidance, not current market rate: $0-500 / $500-1,500 / $2,000+ setup fees and $0-200 / $200-500 / $800+ monthly, with email at about $500/month, social $300-800 and ads $500 up to $5,000-10,000.
Product pricing as heard: Content Creator Templates $19/month with grandfathered pricing and a 10,000+ member claim; Content Creator Machine $75/month or $750/year against a $97/$970 list. Justification figures quoted - a prior stack costing 'well over $8,000 per month', ActiveCampaign at '$24,000 per year', '$41,000' spent building the included templates - are all undated sales claims.
Revenue claims are undated: about 14 retainers at a combined $24,000/month; a commercial said to produce roughly $30,000/week; 14 Day Filmmaker cited as having 'over 8,000 students'.
Gear and prices as of recording: about $2,500 for the whole rig, about $80 for the Yongnuo YN300, about $15 per practical lamp. Camera comparison is Canon EOS R against a 1DX Mark II.
The free versus Studio split as of recording - Studio adding AI tools, noise reduction, 3D, multi-GPU and above-4K export, with AI Voice Isolation Studio-only - is exactly the kind of line Blackmagic redraws between versions.
Version-specific UI warnings appear in the module itself: 'if you're editing on DaVinci Resolve 20, it might look a little bit different', and Resolve 20 is credited with adding a change-media-location option at project creation.
23.976 is presented as 'the industry standard' timeline frame rate, and 1920x1080 / 1080x1920 / 3840x2160 as the working resolutions. Reasonable now, but platform norms move.
The Content Creator Templates extension requires a paid subscription and the student discount is a time-bound offer.
Revenue claims are undated and unverifiable here: over $5,000 a month from individual videos; a LUT pack earning $20,000 a month in its first month; a single evergreen video producing over $6,000 a month; sponsorships at $1,000-2,000 per thirty seconds from about 50,000 subscribers; Amazon affiliate at 3-4%.
A third-party comparison is quoted - Dwayne Johnson's tequila brand said to have made 'over $3.5 billion' between March 2020 and March 2022 against roughly $20 million per film. Repeated as heard, not verified.
The community is described as hosted on Content Creator Machine after leaving Facebook; a domain is quoted at 'only like 10 bucks a year'.
The impression-round numbers are the instructor's model of the algorithm, not published figures.
App pricing as heard: PromptSmart Pro $29 one-off on iOS, PromptSmart Plus voice tracking $7.99/month on Android, Nano Teleprompter about $4.99. Canva's background remover behind Canva Pro.
Proof-point statistics will age: a review with about 7% CTR and 40,000 views in 24 hours; a 7.4 million-view video from a channel with 19 videos and 85,000 subscribers; a video reaching nearly a million views in seven months against 219,000 over two years for its predecessor.
His own channel figures are point-in-time: over 100,000 subscribers across roughly fourteen videos, and over $30,000 a month from the channel. CTR benchmarks of 4% or higher, with best videos at 5-8%, are this niche's numbers.
The YouTube Studio walkthrough matches the interface at recording, and the claim that YouTube has said posting time does not much matter is passed on without a citation.
Hardware named as current: Moman MT12 prompter, Sony A7S III.
Every phone named is superseded: iPhone 12/13/14, Galaxy S21/S22/S23, OnePlus 9 Pro. The lesson even anticipates the S22 Ultra's launch, which dates the recording.
Camera and accessory prices are all as of recording: ZV-E10 $799 with kit lens and called 'the only camera under $1,000 worth considering'; Canon R8 $1,499; Sony A7S III $3,400; M1 MacBook Air about $650 used.
He flags one of his own recommendations as already out of date in the same lesson - the RODE Wireless GO II, 'there are some better options out nowadays'.
The Epidemic Sound free trial month and the Kit.co affiliate pages are both time-bound and commercial.
The Apple Watch as a camera remote and preview is described as his current setup and depends on the OS version.
Every device named is dated: the iPhone 13 Pro as 'one of the best smartphones available', the iPhone 15 Pro Max as 'the newest smartphone', the Galaxy S23 Ultra as 'the flagship Android phone'.
App prices as heard: BeastCam $1.99 on sale and usually under $10; Pro Camera by Moment $6.99; ProTake about $20/year; Manual Camera $4.99 one-off; Filmic Pro moved to $2.99/week. Blackmagic Camera described as iPhone-only at recording, with hope for Android.
Microphone comparison is a snapshot: RODE Wireless GO II at $299 against a newly released DJI Mic at about $329, with DJI judged slightly better on reviews rather than testing.
The apps lesson carries '2024 Update' in its own title, which is the clearest possible signal that the list has a shelf life.
The recommendations lesson is titled '2024 Camera Recommendations' - explicitly year-stamped. Prices as heard: Canon M200 about $500, Sony ZV-1 about $650, Canon R8 about $1,500, Sony A7S III about $3,500 (sometimes under $3,000), Canon R5C about $4,000, RODE Wireless GO II $200-300, PoP Voice Pro lav under $15.
Phone picks are generation-locked: iPhone 12/13/14 and Galaxy S20/S22, with the 14 Pro Max explicitly called out as above the budget. An iPhone 15 appears on a tripod in one demo, dating that recording to late 2023 or later.
'Probably the last camera I'll ever buy' about the A7S III is a personal prediction, not a recommendation with a shelf life.
Two software claims he flags as moving targets himself: Premiere not interpreting HDR well on import, and phone menus constantly changing layout.
Prices as heard: Westcott Solix Apollo about $400, Godox SL60 about $130, Yongnuo YN300 about $68-70, five-in-one reflector about $20, C-stand about $90, Amazon light stands $20-40 for three or four, Edison bulbs about $20, under-desk LED about $10, capture card about $20, LG ultrawide about $1,500.
Devices named as current: iPhone 13 Pro Max and iPhone 14 - with Anthony himself saying the 13 Pro Max 'is going to be outdated in like five months' - the Sony A7C and a 'recently announced' A7 IV, the DJI OM6, and an M2 MacBook Pro.
One lesson is explicitly a 2022 studio tour and another a 2024 one; both are snapshots by design.
The blinds discount code and the kit.co gear pages are commercial and time-bound.
Software pricing as heard: CapCut Pro $10/month, DaVinci Resolve Studio a one-off $295, Premiere Pro $21/month, Creative Cloud $55/month.
Revenue claims are undated and unverifiable here: $12 of ad spend to the first sale, scaling to $35,000 a week within a month and later about $100,000 a week; a channel of almost 150,000 subscribers generating about $20,000 a month in course sales; one sale per 250-350 views; a single video producing $5,000-10,000 a month on autopilot; affiliate income of $5,000-8,000 a month; one lighting-video affiliate link generating over $26,000 in about six weeks; a $3,000 mid-roll sponsorship.
The lesson title says $30k a month; the body says about $20,000 a month in course sales. Recorded as heard, not reconciled.
Content Creator Machine at $75/month or $750/year for students against $97/$970 list, with a claimed $24,000/year previously spent on ActiveCampaign. Genius Link at about $70/month. Amazon commission 1-7%, direct deals 15-25%.
Computer prices and configurations are all point-in-time, and the Apple silicon generation named has since moved on.
Product pricing as heard: Content Creator Templates $19/month with a 10,000-member claim; Content Creator Machine $75/month or $750/year against $97/$970; the Captions app $9/month.
The Adobe discount lesson dates itself explicitly - 'right now I'm shooting this video in May of 2020' - and depends on a specific third-party student-verification partnership that may no longer exist. Treat the whole lesson as historical.
Internal cost comparisons quoted in the Machine pitch - about $8,000 a month on software, $24,000 a year on ActiveCampaign, $41,000 spent converting templates - are undated internal figures.
The relevance example is pinned to a filming date of 18 April, and app interfaces for CapCut, Instagram and VN will have moved.
'It is proven right now that better captions leads to better retention' is asserted without a source.
App tiers, caption-style gating and platform export guidance are all as of recording and change frequently - see cc-sm-06's freshness notes, which apply verbatim.
Free-versus-paid tiers in Resolve and CapCut, and Adobe's subscription pricing, are all as of recording; see the sister records' freshness notes, which apply here unchanged.
The recommendations lesson is titled '2025 Camera Recommendations', which ties every pick and price to that year. Prices as heard: Canon M200 ~$500, Sony ZV-1 $650, ZV-E10 $799 with kit lens, Canon R8 $1,499, Sony A7S III $3,400-3,500, Canon R5C $4,000.
Phone generations named: iPhone 12/13/14 and Galaxy S20-S23. Apple Silicon discussion is M1/M2 era.
Accessory prices as heard: PoP Voice Pro $10-15, Comica Vimo C3 $111, RODE Wireless Pro $400, NTG3 $700, Godox SL60 $139 or $220 as a kit, Samsung T7 Shield 2TB $112, Moman T12 $149, FeelWorld LUT7 $249, DJI OM6 $149, Edelkrone slider components $499/$549/$999, fluid-head tripod $139, Peak Design travel tripod $650-700.
Business claims are undated: over 100,000 students across the catalogue and 35,000 in this programme; $10 million generated over four years; about 600,000 views a month for roughly $4,500 in AdSense; $4-5,000 a month affiliate; a first brand deal at $3,000; over $400,000 a month in digital products; a $69 first ad spend returning fourfold; and two named students on $200,000 a year and a $5,000-a-month retainer.
Dynamic-range and ISO figures are model-specific and generation-bound: A7S III about 16 stops and usable to about ISO 12,000; Canon EOS R about 12.5 stops and usable to about 1,000; newest phones about 14 stops.
The claim that Premiere does not interpret HDR footage well on import is software-version dependent and may no longer hold.
Device call-outs - iPhone 13, 14 and 15, and recent Galaxy phones - for manual HDR and bit-depth control will age.
The claim that the 1D X Mark II has a record-time limit while his Sony bodies do not is firmware and model specific.
Prices as heard: Westcott Solix Apollo $400, Yongnuo YN300 $68-70, five-in-one reflector about $20, C-stand about $90, Amazon light stands $20-40, PoP Voice Pro lav about $15, Filmic Pro about $15, Edison bulb about $20, capture card about $20, LG 40in monitor $1,500.
Devices are dated and he says so - an iPhone 13 Pro Max, an iPhone 14, an M2 MacBook Pro, and 'I'm sure it's going to be outdated in like five months'. One lesson is explicitly a 2024 studio snapshot.
Software notes tied to the moment: Filmic Pro's manual controls, and Zoom's basic paid plan used for streaming with an intention to test an alternative.
Software pricing and the scored rankings are as of recording: CapCut Pro about $20/month, DaVinci Resolve Studio a one-off $295, Premiere about $22/month or $66 for Creative Cloud, Clipchamp about $12/month. The claim that Resolve's free version will not handle 10-bit is version-specific.
Export rules of thumb - YouTube 4K at 50-80Mbps, social 1080p at 10-20Mbps at 20-60MB - reflect platform compression at the time.
The stock-licensing income figure (a coaching client averaging a few hundred a month) is a single dated anecdote, not a benchmark, and Blackbox is linked with an affiliate code.
Content Creator Machine at $75/month or $750/year against $97/$970, justified against a claimed prior spend of over $24,000 a year on ActiveCampaign.
The export file-size comparison in the lesson is unusable as transcribed - a ProRes 444 XQ master at 25.16GB against an H.264 export given as '891 gigabytes which is 25 times smaller', which cannot be right. The principle stands, the numbers do not.
The LUT is described as 'like a five dollar LUT... might have even been free' for Mavic 2 Pro footage - a time-bound price on one product.
Template-library marketing claims - competitors 'costing upwards of 30 to 50 to even $100 per month', a member discount, 'thousands of new templates every single month' - are promotional and will change.
Hardware framing: an iMac Pro described as 'pretty powerful' for 4K drone footage reflects the standards of the time.
The 40% speed rule for 60fps into a 24fps timeline is arithmetic for those two rates, not a universal number.
The project folder is named 'After Effects CC 2019', which dates the recording and the interface shown. The edit machine is described as a 27-inch iMac Pro.
The pricing example - $1,800 for an original one-day-shoot commercial and $1,200 for a second built from leftovers - is his own rate at the time, as is the claim that this evolved into most clients being on monthly retainers.
Composition presets are given for a 24fps US context, with a note that 25fps is common elsewhere.
The pest-control results are dated precisely and usefully: 44 phone calls in November 2019, an average deal of $150-250, and roughly $10,000 a month attributed to one commercial. Production: about $400, two hours of scripting, a four-hour shoot, two hours of editing.
Pricing quoted: $500 per project historically against a current minimum of $2,000 per commercial plus a retainer from about $1,000 a month. A $48,000 advertising campaign is cited as the context for the five-commercial day.
Gear prices as heard: teleprompter about $150, Westcott Solix Apollo about $450, backlight $75-100, 50mm about $250, desk lamps $10-15, a full tutorial lighting setup for about $300.
Distribution claims: organic Facebook posts credited with over $300,000 to the agency in three years, and a 20,000-fan business page getting only a couple of hundred views per post - a platform behaviour that changes.
The paid-ads scale figure quoted (upwards of $100,000 a month managed between themselves and clients) belongs to a separate paid mentorship, not to this course.
The upsell pricing contradicts itself: the lesson is titled '70% Off' while the video says 60% off, $198 against an everyday price of $498. Treat any price here as unverified.
Revenue claims are large, specific and undated in places: 1.5 million in first-year sales from the NLC value video; over $3 million in two years from the NLC ad; $30,000 per week from the 14 Day Filmmaker ad; $10,000 per month from a pest-control commercial; $2.5 million spent on paid distribution.
Testimonial results quoted: a 1,200% ROI in one month for Steven's tourism business, $50,000 in sales for Jason's ad. Unverified and time-bound.
The 'video increases conversions up to 80%' statistic is attributed loosely to HubSpot and one other source; it is a marketing-industry figure, not a study result, and should not be repeated as fact.
The cited study on music and memorability is attributed to a firm the transcript renders as 'Neurosite' - the name is unreliable, so the finding cannot be traced from this source.
'We currently spend about $2,000 a day on paid advertising techniques promoting our 14 day filmmaker course' - a dated spend figure, not a standing fact.
A produced commercial is described as 'currently airing on television during the show the Real Housewives', which pins the recording to that show's run.
Daniel Schiffer is credited with popularising the seamless-transition style - a snapshot of who was prominent when this was recorded.
Backstage, GigSalad and the music-licensing platform named are all subject to rebranding or closure; verify before recommending.
The Bud Light Seltzer example references a specific dated product launch.
Gear prices as quoted: EOS R rig about $2,500, G7X about $1,000, Solix Apollo Orb about $450, Yongnuo YN300 $60-80, Glide Gear TMP100 about $200, Prompt Smart Pro $15-20.
Course figures quoted in passing: 'over 88,000 students' and 'over 25 hours of content' for 14 Day Filmmaker, and in a different lesson lifetime access 'for just $48' - the two do not reconcile and neither is current.
The website pricing ladder ($0-500 through $2,000+ setup, $0-200 through $800+ monthly, ads management $500-10,000) is a snapshot of one market at one time.
The hiring anecdote of 'literally 100 plus applications' describes a single hiring round, not a standing state.
CapCut's Auto Reframe being paid-tier is a product-tier fact that changes with the vendor's plans.
The COVID-19 retainer anecdote is a period observation, useful as a structural argument rather than as a forecast.
Every price in this section is point-in-time: Templates at $19/month with a grandfathering promise, CC Machine at $97/$970 standard and $75/$750 as the member offer.
'Over 10,000 content creators have already joined' is an adoption claim tied to the recording date.
The Adobe education route depends on both an active partnership with the named institution and Adobe's current eligibility policy; the walkthrough cites May 2020 as its example date.
The comparison figures - about $8,000/month of prior software, $24,000/year for ActiveCampaign, $41,000 on template conversion - describe one company in one era and are not benchmarks.
Named comparison tools (DocuSign, ActiveCampaign, Kajabi, Close.io) all reprice and re-tier independently.
Both modules are republished across at least three ContentCreator courses under different lesson ids; if a course is restructured, the ids here may stop resolving even though the teaching does not change.
Software-version-specific detail in these lessons - panel layouts, effect names, export presets - ages faster than the workflow logic around it.
The stack snapshot is launch-window fresh: Twenty's UI/pricing, PostHog's free-tier limits (~1M events), Resend tiers, and the '3UI' library will all drift.
The Claude-vs-Codex mood readings ('attitude since Fable 5', slower, pricier) are period sentiment, not benchmarks — but they triangulate with the sprint's subsidy-math session a day earlier.
The PostHog→Twenty CRM integration was prescribed as prompts, not demonstrated — verify the wiring before relying on it.
His open-rate ladder is stated practitioner lore, not a cited study.
The promised resource pack (notes, prompts, repo links) lands on the LMS 'tomorrow in the first half' — harvest it with the next scheduled run.
Every price in this record is a point-in-time DJI list price (2022 guide; 2025 guide: Neo $199, Mini 4K $299, Mini 3 $419, Mini 4 Pro $759, Air 3/3S $1,099, Mavic 3 Classic $1,399, Mavic 3 Pro $2,199, Inspire 3 $16,499); refurbished and DJI Care prices likewise
Model verdicts are as of recording - the Mini 5 Pro and Mavic 4 Pro post-date the 2025 guide, and the Air 3S 'top pick' predates both
Air 3S availability 'customs issues', Mini 5 Pro 251.7 g / EU C0 status, Avata 2 1,600 ft altitude cap and the Mavic 4 Pro max-speed setting are firmware- and rule-dependent
All six per-drone recipes (Air 2/2S, Air 3S, Mavic 4 Pro, Mini 5 Pro, Avata 2, Air 3/Mini 4 Pro) are DJI Fly app-version and firmware-bound - menu paths, the Air 3S 'Auto-only 14 stops' behaviour, the Air 3 HLG/Normal mid-shot shift, the Mini 5 Pro autofocus sensitivity ('at least with the current firmware') and the Air 3 dual-gain cutoff above 100 fps may all change
'No Freewell NDs yet for the Mavic 4 Pro' and 'DJI isn't shipping Air 3 ND filters yet' were true at recording only
Sensor-size ISO ceilings (half-inch: 400 unusable; one-inch: 800-1600) are generation-specific
Cruise Control availability (Air 3, Mavic 3, Mavic 3 Pro at recording), the drag-to-orbit tracking control on Air 3 / Mini 4 Pro, and the C1/C2 default mappings are firmware-bound
Specific EXP (0.3) and Air 2/2S Sport-mode gimbal values (pitch 30 deg/s smoothness 5, yaw 50 deg/s smoothness 7) are his 2022-era settings
The ~30-degree upward gimbal cap before props enter frame varies by model
Air 2S RAW-interval minimum (5 s) and 4K60 crop, the Air 3 HLG mid-shot shift ('wish DJI could fix that'), the Mavic 4 Pro 6x POI limitation, and DJI Fly UI locations (Quick Transfer box, task-library icon) are firmware/app-version claims
Prices - a $135 power station, sub-$30 walkie-talkies - are sale prices at recording
'The new waypoints feature on the Air 3' dates the spec-ad lesson to the Air 3 launch window
Airline battery rules and TSA practice are as he experienced them; the country-rule sources he names can go stale
Resolve UI paths, the default append shortcut (Shift+F12), Speed Warp being Studio-only and its render speed, and Windows hardware-encoder options are version- and machine-specific
YouTube's recommended bitrates (35-45 Mbps 4K, 8 Mbps 1080p) and 'H.265 not widely supported' are as of recording
FastStone pricing ($25 commercial) and 'Windows only' as of recording
The D-Log M lesson is explicitly a '2025 Update' that supersedes his earlier 'don't use log LUTs' advice; drone-to-profile mapping (D-Log M: Mini 4 Pro, Mini 5 Pro, Air 3, Air 3S, Flip, Avata; D-Log: Mavic 3, Mavic 4 Pro) will move with new models
Color Space Transform has no D-Log M preset 'as of' the lesson; Resolve versions may add one
The Mac colour-management mismatch was confirmed on a 2021 M1 MacBook Pro; Resolve and macOS updates may change the fix
The example 'DLOGM to Rec709' LUT is the Mavic 2 Pro's
Drone casts (Air 2 green, Air 2S/Mini 5 Pro magenta, yellow greens, pink roads) are per-model colour science and may change with firmware or new models
LUT prices (Parker Walbeck ~$10-12, X-Grain ~$15 bundle) and the 'free' status of the Lutify.me and Color Grading Central LUTs are as of recording
The DCI-P3 transform trick depends on Resolve's colour-management defaults at recording; the night lesson's drone is heard as 'Mini 5 Pro' - possibly Mini 4 Pro
'Something I've been doing recently' (outside nodes) and 'brand new' (DCI-P3 PowerGrade) date the Slovenia lesson relative to the favourites reel
The LUT install path names a year-versioned app folder ('Premiere Pro 2024 or whatever the year is') and Premiere's package layout; both change with releases
The HLG default-interpretation behaviour (sequence becomes Rec.2100 HLG) and the 'LUT destroys the grade' quirk are Premiere-version claims
LUT names and the 'included in the course / sold separately' status of the Cinematic Drone LUT pack are as of recording
Jun 2025: DJI executive-order/FCC situation, Mavic 4 Pro US availability and reseller mark-ups, Care Refresh parts rumours, the US geofencing removal - all moving targets
Model names and tiers are a September 2026 snapshot: Kimi K3, GLM 5.2 (5.3 mentioned as existing), Gemma 4, Nemotron, New Spark, DeepSeek, Qwen. Free-tier availability on OpenRouter changes without notice.
Hostinger pricing moved during his own use: he cites ~$11/mo rising to ~$14 for KVM2, $16 with tax, and a 20% referral code. Re-check before quoting anyone.
OpenClaw moved its .md files in an update days before this session and his agent could not locate them live. Any file-path instruction here is version-specific.
ClawLink's URL is unverified even by him ('clawlink.dev or claw-dash-link, just check that') and the $5/month price is as-stated.
The 'Meta's autonomous agent' attribution around 0:26 is unreliable - the ASR renders Nous Research as 'news' throughout, and the Hermes agent is a Nous Research product, not Meta's. Treat the terms-of-service anecdote as directionally true, attribution unverified.
Instinct is invite-only as of this session and he rates it above Hermes for proactive assistance. Availability and that ranking will both move.
The OpenAI benchmark sandbox-escape story is told from memory, with the benchmark name forgotten on air. Verify before repeating it as fact.
'The limit is not in the model, it's us' is attributed to Anthropic's founders from a slide, not quoted from a source. Treat as paraphrase.
The remaining C3 curriculum (LangGraph, LangSmith, Expo mobile, distribution, Founders Time, product readiness) is a plan stated on 2026-09-13, not a schedule. Check the LMS for what actually ran.
Every market figure in the Swipe AI pitch - Amazon's $3.5bn in written-off failures, Uber's ~10% failure rate, the $4.1bn market, 'three or four companies do this' - is the presenter's claim from his own industry experience, unverified on air, and the host immediately produced competitors he had not found.
The skills-audit numbers (~4,000 skills, a third problematic, 76 stealing credentials, February) are cited from a slide with the auditing organisation's name garbled. Verify before repeating.
RoleCompass's own cited statistics are from its pitch video: 59% of the workforce needing training by 2030 (World Economic Forum) and 80% of professionals feeling unprepared for 2026 (LinkedIn). Both are second-hand here.
Palak's traction numbers (~70 signups, 'payments started', team of four) are as stated in September 2026 and will move.
SkillPort's listing count, user count (15 at the time, three of them the founder's own) and free-forever commitment are all point-in-time.
The n8n-is-receding claim is a working opinion about tooling in September 2026, not a durable fact - Sarah's shipped product runs on it.
Highest-decay session in the recovered batch: media-model names, routing behavior, and credit pricing (Veo 3.1, Nano Banana, 75-credit runs, 5-15-minute renders) shift monthly; Higgsfield's catalog, HeyGen's API surface, and Codex's annotate/in-app-browser features are all weeks-old-at-recording. Sora's withdrawal claim is cohort speculation. Durable: the provider/aggregator/cloner map, replaceable-boundary design, V2/V3 scope discipline, component-bound feedback, and the entire interface-moat pricing doctrine.
Fast-moving: Codex desktop features (projects, drag-context, sub-chat spawning, status sites) are weeks old at recording and will shift; OpenAI voice model names/credentialing were unstable ON CAMERA (realtime key rejections) — resolve current model ids through the agent, as taught. Apify actor pricing/catalog and Firecrawl free-tier numbers are perishable. Durable: the loop cycle and its guardrails, worktree-per-agent parallelism, the actor/API/self-host data hierarchy with approval-gated spend, plan-first expansion, and the model-agnosticism point.
Perishable: model tier names (5.6 extra-high / Sol light), GPT Image latency/pricing, LinkedIn's ~3-week product-approval timeline and product catalog, Resend's free-tier shape, and especially the metadata-stripping trick — content-credential detection is actively moving server-side, and the pretext prompt may stop working (or start mattering legally) at any time. Durable: owned-distribution doctrine, the pipeline shape with a human publish gate, image-native slide design, harness-in-the-folder portability, and goal.md.
Fast-decay zone: OCR model names, download counts, and parameter sizes are as-heard ('Unlimited OCR', 2.8M downloads, 3B params) — verify on Hugging Face before use. The PixPipe pricing claims (fixed image tokens, $20→$4, 30-70% savings, Fable-5-only efficiency) are stage numbers tied to one moment's pricing tables; benchmark before relying and expect the loophole to move. Datalab's licensing terms and discount offer may change. The parsing-vs-scraping frame, the no-LLM runtime doctrine, the three-gate dependency check, and foundation-before-theme are durable.
The session title advertises "Gemini File Search API" but the delivered build uses Haystack + embedded Qdrant; Gemini appears only via the Google Gemini Cookbook repo tour. Model-market claims (parameter counts for Claude/GLM/Kimi, the "Laguna" model, DGX Spark pricing/variants) are as-heard live commentary — perishable and unverified. The RAG framework landscape itself moves fast: re-run the measurement procedure, not the conclusion, before building.
Evergreen mindset content with a dated tool map: the market-share figures (Elementor ~12M, Divi ~2M, Webflow ~500k), Adobe XD's abandonment, Figma Sites' limitations, and the AI-builder aside (Lovable, Bolt) are all as-heard snapshots. The psychology (resistance, imposter syndrome, originality, the five myths) and the design-first argument are durable - they are the course's least perishable chapter.
Fully evergreen chapter: psychology of unstructured work, style-follows- function, client empathy, the assumption guardrail, and the workflow preview carry no tool or version dependency. The only dated texture is incidental (the Bali recording location, the viral Instagram skit).
The method (four phases, agreement rounds, research order, extend-vs-create, colored wireframes) is durable and is the course's crown jewel. Perishable: every tool tier and price ($250 is his personal, location-based rate; Loom's 5-minute cap; Typeform's dropped free plan; Relume's paid features and in-beta style guide; the AI-builder landscape he critiques). Re-verify tool specifics before acting; adopt the order and the gates as-is.
The fundamentals (palette structure, type system, spacing math, proximity, the eleven styles) are the course's most durable material. Perishable: every tool price and tier (Creativa/Magic Pattern deals, Unsplash paywalling, Font Awesome counts), font fashion (Inter/Ubuntu/DM Sans as current tastes), the trendy-effects annex (self-declared future-dated), the EU accessibility figures (as-heard, not legal advice), and 3D-icon/AI-generation trend calls.
The method (control posture, wants/needs interpretation, four-source order, one-pager bias, screenshot direction, gates) is durable. Perishable: Figma's free-tier boundaries (the pricing change that reshaped the course), Fillout and Relume feature tiers, ChatGPT model names, and Dribbble/Behance as the inspiration venues of record. The AI-threat framing ('next 10 years') is the course's own time-stamped bet.
The system doctrine (connect-before-duplicate, token scales, components/ variants, constraints) is durable Figma craft. Perishable: stylekit.com's feature set and the plugin's approval status, Figma free-tier boundaries, font fashion (Pally/Outfit as the picks), copysection.com's library depth, and the corner tokens' exact pixel values (never stated cleanly on camera - rule reliable, numbers unverified).
The delivery method (order, cadence, versioning, disclosure language, motif extraction) is durable craft. Perishable and already-moving at recording: BatchStyler's death mid-course, ChatGPT's image quality and the stock-to-AI inversion claim, background-remover pricing, Envato's AI features, and every plugin's availability. The popup's Open Overlay wiring reflects current-era Figma prototyping.
April-2026 snapshot: model versions, plan shapes (pro exhausting in ~20-25 minutes of building), the Anthropic run-rate figure, and the Carlini/ Mythos story are all as-heard and time-bound; the ':online' OpenRouter suffix and PM Skills repo stats should be re-verified before use. The Claude outage is a dated event. Durable: the taste/systems/craft doctrine, muscle-and-mirror with disk-resident artifacts, clarity-as-token-economy, Opus-for-planning rationing, selective skill import with community-signal judgment, the model-council pattern, and the full secrets ritual.
April-2026 snapshot: every repo statistic, skill name (UI UX Pro Max, Claude-Mem, caveman), marketplace state (skills.sh listings, Anthropic's first-party testing skill), and model version is as-heard and fast-moving; the 50%-degradation threshold is empirical folklore anchored to one paper and one model generation — re-test per model. Durable: the CLAUDE.md/ memory.md behavior-vs-state split, the exit-and-resume ritual, clarity- over-compression, conflict-free skill scoping, security audits as a shipping gate, proportional use of skip-permissions, and hardware as part of the agent budget.
Model names and prices are as-heard snapshots (Opus 4.6-vs-4.8 economics, Codex-vs-Hermes token allowances, Scrape Creators $47 tier, Seedance v4); Hermes itself is young and moving fast, and its md-file conventions (soul/user/memory) may rename between versions — verify against the repo before configuring. Durable: the dynamo redesign doctrine, the employee-on-day-one file architecture, the feedback→lesson→patch loop contract, agent-as-harness operations, and access-equals-power.
Highest-decay elements: every price and model claim (DeepSeek v4 pro rates and the '75% forever' discount, Claude version numbers, Fable/Mythos availability, the as-heard 'GOLD' command and 'Glasswing' access story) — all are stage snapshots, several explicitly unverified. Hermes' desktop app was days-old and self-described buggy. The generated news brief's contents are demo output, not reporting. Durable: the loop anatomy and era frame, the three loop patterns (scheduling, delegation, auditor), the plan-approval gate, the deterministic-vs-decisive routing rule, and cheap-model-with-fallbacks economics.
Launch-week snapshot of a fifteen-day-old product: every quota, price, model comparison, command list, swarm limit, and MCP availability here is as-heard and will drift fast; the CLI's open-source status had already flipped once by recording. The Lovable/Replit 'mockups only' verdict is the trainer's competitive opinion, contradicted by production Lovable work elsewhere in this KB. Durable: the four-product decoder, the grill→plan→goal→verify ritual, hooks-vs-skills-vs-schedules taxonomy, the task.md swarm-coordination idea, the import-CSV privacy pattern, and speed as a distinct routing axis.
~March-2026 snapshot of the fastest-moving object in the archive: OpenClaw's name, ownership optics (Peter at OpenAI), Hostinger/Nexos plans and prices, the Anthropic tier thresholds and token limits, the OAuth ban map (Google permanent / Anthropic ambiguous / OpenAI allowed), model codes (5.2, 4.1, Sonnet 4.5, Haiku 3.5, MiniMax), and the claw-clone wave are all perishable — re-verify every number before acting. Durable: infrastructure-not-agent, the two-layer document architecture and memory-decay audits, jail-your-agent isolation, read-only-first rollout, routing-by-task-class, and 'power without security becomes risk.'
~April-2026 snapshot. Perishable: every token number (10-15k prompt tax, 8k schemas, 170k heartbeat, 90% cache discount, 80%/30% savings), all model codes and prices, ClawHub/awesome-list contents, Hostinger plan mechanics, certificate logistics, and the claim that prompt caching must be manually enabled. Durable: the seven-drain anatomy as an audit framework, the eight fixes as mechanisms, install-the-agent-where-the-work-is, screenshot-driven beginner loops, approval-gated skill installs with .env key placement, DIY-skills-as-capital, and agents-are-folders multi-agent design with per-seat model economics.
The acquisition narrative (who bought which part of Windsurf, headcounts, what shipped as Antigravity vs Windsurf 2) is as-heard storytelling — directionally useful, numerically unverified. Model menus and prices (adaptive $0.5/M, SWE 1.6, Kimi K2.6 free tier, BYO Claude) are a May-2026 snapshot of a fast-moving product. The Anthropic/OpenClaw rate-limit example is time-bound. Durable: conversation-vs-specification, the pizza GitHub model, the four spec concerns (edge cases, auth-vs-authorization with auth-last, rate limits, error handling with the ask-me default), and lineage-aware tool evaluation.
Product-state snapshot: Windsurf's Opus lockout (BYO key), Devin's pricing and trial shape, adaptive-model menus, and the Windsurf-Devin merger trajectory ('they are actively working to merge it') were all in motion at recording; the X API price and Apify actor economics likewise. As-heard names (styles.referrer.design, DeepVicky, Sourceti) need verification before citing. Durable: the IDE-vs-autonomous lane split with the sandbox model, plan-with-frontier/execute-with-cheap via in-repo artifacts, the spec/plan audience distinction, curated-source retrieval bias with budget-gated fetching, review-report-first on existing code, and legible MVP failure.
This is a news-digest snapshot of July 2025 AI-tool launches (ChatGPT agent, Genspark super agent, VO3/Grok 4/Imagen 4, Google Web Guide); every version number, price, credit limit, and launch date is as-heard from the presenter on-air and several claims (GPT-5 timing, context-window sizes, company names) are explicitly speculative or mis-transcribed — rewatch the episode or verify independently before treating specifics as current.
This is a news-snapshot digest of AI product/news claims made live in August 2025 sessions (tool pricing, feature names, credit costs, and model comparisons as stated by the host); verify current details against official sources before relying on them, as pricing and features change quickly.
This is a news-snapshot digest of tool states, prices, and feature availability as reported live in September 2025 (MAI Voice 1 free/English- only, Claude file editing gated to the $100/mo Max plan, GenSpark browser credit limits). Tool tiers, pricing, regional rollouts, and feature sets change quickly — verify current status before acting on any specific claim.
This is a news-snapshot digest of live AI-news roundup webinars from October 2025; tool names, pricing, feature availability, and legal or regulatory status (India's DPDP rollout, platform disclosure policies) may have changed materially since recording and should be re-verified before being treated as current fact.
This is a rolling AI-news roundup snapshot from November 2025 (4 episodes); tool names, pricing, feature availability (Gemini 3 US-only rollout, ChatGPT Go pricing, Pomelli's regional beta), and legal disputes (Amazon v. Perplexity) may have changed materially by the time this is read — verify current status before citing as fact.
This digest summarizes live weekly webinar recordings from December 2025; tool names, pricing, credit costs, feature availability, and country rollouts (e.g. Hedra, Suno, 11 Labs Music, Notion AI, Gemini AI mode features) change frequently and should be reverified against current vendor documentation before relying on any specific claim, price, or capability. Several product names were transcribed inconsistently by ASR (see asrCorrections) and a few speaker claims (e.g. exact credit costs, plan limits, legal/copyright interpretations) are the presenter's informal understanding as of late December 2025, not verified legal or vendor guidance.
This digest captures AI news exactly as reported live by the host across four episodes in January 2026 (Jan 7, 14, 21, 28). Tool names, feature availability, pricing, and rumored product codenames (e.g., "Garlic") reflect the host's and live chat's understanding at broadcast time and may since be inaccurate, renamed, region-limited (several features were cited as US-only or beta), or superseded. Several items (OpenAI's rumored model, xAI's "vibe coding agent", some Grok upgrade claims) were explicitly described as unconfirmed or rumored by the host. Verify current tool names, availability, and claims independently before relying on them for coursework or citations.
This digest reflects AI news and product claims as narrated live in four Weekly AI Updates sessions during February 2026 (Feb 4-25). Company valuations (e.g., Anthropic at $380B, the SpaceX/xAI combined entity at $1.25T), pricing, credit allotments, model version numbers (Gemini 3.1 Pro, Claude Sonnet/Opus 4.6, Grok 4.2, ChatGPT 5.3 Codex), and feature availability (regional rollouts, free-tier limits) change quickly and should be independently verified before being treated as current fact. Some names are host pronunciations or ASR transcriptions of fast-moving news items (e.g., "Moltbook", the OpenAI hire referred to as an "open clock creator") and may be imprecise -- cross-check against primary sources.
This digest summarizes informal, host-narrated news roundups recorded live between March 4 and March 25, 2026. Product names, benchmark figures, funding amounts, and feature claims (e.g. Nano Banana 2's language/resolution specs, GPT-5.4's OSWorld score, Sarvam AI's Nvidia funding, Claude's off-peak usage window) are as stated by the hosts on air and have not been independently verified against vendor documentation. Several product names were unclear or inconsistently pronounced in the audio (see asrCorrections) and should be confirmed before being cited authoritatively. Given the pace of AI product releases, feature sets, pricing, and availability described here may already be outdated by the time this digest is read — treat all specifics as a snapshot of what was reported in March 2026, not as current fact.
This digest summarizes live weekly-update sessions recorded in April 2026 and reflects the hosts' real-time reporting, including several leaks, rumors, and unconfirmed claims (e.g., Anthropic's Mythos/Project Glasswing, a leaked Google enterprise chat interface, GPT-5.5 speculation, and disputed claims about Claude's reasoning effort being reduced). Tool names, pricing, availability, and feature sets described here (ChatGPT Images 2, Claude Managed Agents, Claude Routines, Gemma 4, design.md, etc.) were current only as of their respective recording dates and may have changed, been renamed, or been deprecated by the time this is read. Several proper nouns were transcribed via ASR from a live audio feed and could not be independently verified; treat names flagged in asrCorrections as approximate. Verify current tool capabilities and pricing directly before relying on this content for course material.
These four episodes (May 6-27, 2026) are live AI-news roundups delivered informally by rotating hosts reading press releases, teaser videos, and audience chat in real time; several items were explicitly described on-air as unlaunched, beta/trusted-tester-only, US-only, or unconfirmed rumor (e.g., "Claude Mythos" was called speculative/unverified by the host; Google's screen-aware pointer, Mira Murati's real-time AI, and several Google IO features were teasers without public release dates). Figures such as deal sizes, chip counts, hallucination-rate percentages, and subscription prices were stated verbally by hosts without a cited source and may be approximate or mis-heard. Given the fast-moving news-roundup format and six-plus months since original recording, treat all pricing, feature-availability, and company-deal claims as a snapshot that should be re-verified against current official documentation before being cited as fact.
These four episodes are informal, live-streamed news roundups recorded Jun 3-24, 2026, delivered from the hosts' personal impressions of headlines, benchmarks, and screen demos rather than vetted reporting. Company names, model names (e.g., "Fable 5" vs. "Mythos"), pricing, valuations, and benchmark claims are as-heard and were not independently verified; several items (the Fable 5/Mythos ban, SpaceX's IPO/valuation, the Cursor acquisition, Anthropic's ID-verification policy) were explicitly described by hosts as still developing at recording time and may have since changed, been reversed, or been clarified. Treat all figures, dates, and product claims in this digest as a snapshot of what was said on air, not confirmed fact.
This digest reflects a live weekly AI-news roundup recorded in July 2026; treat specific claims (US government intervention on Claude Fable 5 and GPT-5.6, named pricing figures, "AI inflation" hardware price jumps, OpenAI's screenless device and Apple lawsuit, and any tool named only as-heard by mentors) as snapshots from that week, not verified facts. Several tools discussed were in beta, US-only, or not yet personally tested by the presenting mentors (e.g., xAI Voice Agent Builder, Gemini Spark, OpenAI's screenless companion device, Manus PowerPoint generation) — confirm current availability, pricing, and feature scope before relying on this digest for course content or advice.
This is a cheap-tier news-roundup digest of a fast-moving field; treat every product name, benchmark number, and price figure as a snapshot circa August 2026 that is likely stale within weeks. Several names are uncertain due to live-transcription artifacts (see asrCorrections) — e.g. "Quinn" for Qwen — verify against primary sources before citing any of them. The Aug 26 episode (lessonId 4276717) was posted late (published-but-empty until ~Sept 4) and was added to this digest on 2026-09-05; Stripe/OpenRouter deal terms and the Luna price cut are as reported in the episode, not independently verified.
Cheap-tier news-roundup digest; every product name, price and benchmark is a snapshot circa 2 September 2026 and likely stale within weeks. Mac mini pricing, Kimi tiers and the Fable 5.1 cost/benchmark figures are as shown in the episode's on-screen material, not independently verified. Sept 2, 9 and 16 are in; Sept 23 and 30 append here as they post. The Sept 9 'GPT-6 Astra' and Sept 16 Astra deep-dive report the presenters' claims - release dates, tiers and the OpenAI attribution were not independently verified.