<- all tokdocs
2026.10.02@_vibe_academy

CD3 is cost of delay divided by duration, exactly as the video says; the video then lists "rough effort tags" as an input, and in Reinertsen and Arnold the denominator is calendar duration rather than effort

Business
Watch on TikTok

View on TikTok ->

A 68.266667 second vertical clip, 720x1280 at 30/1 fps, delivered as a QuickTime/MOV-family MP4 (format_name mov,mp4,m4a,3gp,3g2,mj2, format id bytevc1_720p_154846-1), carrying one HEVC Main profile video stream at level 120 and pix_fmt yuv420p alongside one HE-AACv2 audio stream at 44,100 Hz in stereo; ffprobe reports an overall container bitrate of 154,847 bps against a video stream bitrate of 83,885 bps and an audio stream bitrate of 64,174 bps, 2,048 video frames and 1,474 audio frames, and a filesize of 1,321,366 bytes, which matches the filesize field in metadata.json exactly. The post timestamp is epoch 1790978484, which is 2026-10-02T22:01:24 UTC, consistent with the upload_date field 20261002. The uploader handle is _vibe_academy with uploader id 7587180583122797623 and channel nickname "Vibe Academy"; the audio credit is original sound by artist "Vibe Academy", so there is no licensed music track to trace. The description runs 231 characters and ends in six hashtags: #ProductOwner, #BacklogGrooming, #ScrumAI, #AgileAI, #Prioritization, #LearnAI. At the capture time recorded by the metadata.json mtime, 2026-10-19 06:06:22 local, the counts stood at 3,381 views, 78 likes, 0 comments, 27 reposts and 101 saves. I read 11 of the 34 extracted frames: 001, 003, 005, 007, 009, 012, 016, 020, 025, 030 and 034, which at the 2 second extraction interval correspond to roughly 0 s, 4 s, 8 s, 12 s, 16 s, 22 s, 30 s, 38 s, 48 s, 58 s and 66 s. Every one of those eleven frames shows the identical static layout: a cream card headed LESSON 03 on the left and 03 / 10 on the right, a crimson eyebrow reading PRIORITY MECHANICS, a serif headline reading CD3 = Cost of Delay ÷ Duration, then three labelled monospace blocks, COMPUTE / "Paste backlog → AI computes CD3 scores → ranked shortlist.", CHALLENGE / "Add the team context AI doesn't have: commitments, tech debt, stakeholder reality.", and DEFEND / "A ranking you can explain — not gut feel dressed as a number.", with a cut-out talking head of a bald man in a black crew-neck shirt in the lower third, a burned-in rolling caption beneath him, and a footer reading vibe.academy | AI FOR PRODUCT OWNERS. The rolling captions I read verbatim were "Cost of delay" (frame 001), "turns gut feel into a" (003), "This is lesson four of ten" (005), "So you paste your backlog," (007), "tags, any revenue or support" (009), "surface the top contenders." (012), "to challenge the scores with" (016), "than its score suggests, or the" (020), "The output is something you" (025), "list of half-formed" (030) and "don't see until sprint day" (034). No spreadsheet, chat window, terminal, backlog table, score column or arithmetic of any kind appears in any frame I read, so the AI step the card describes is asserted and never demonstrated. The spoken argument across 16 transcript segments and 179 words is that CD3, defined as cost of delay divided by duration, converts prioritisation instinct into a defensible number; that a product owner should paste ticket titles, rough effort tags and any revenue or support cost signals into an AI tool, have it compute CD3 scores, and then override the ranking with the commitments and tech debt the model cannot see. The transcript-versus-audio check is clean: the final segment ends at 68.12 s against an audio.mp3 duration of 68.336327 s and a container duration of 68.266667 s, so nothing is timestamped past the end of the audio and there is no Whisper hallucination tail to discount.

Verified: the formula the video states is the formula the CD3 primary source states

At 0.00 s the video opens with "Cost of delay divided by duration." At 2.70 s it continues, "That's CD3 and it turns gut feel into a number you can defend." The on-screen headline in every frame I read renders the same thing symbolically as CD3 = Cost of Delay ÷ Duration, and the TikTok description repeats it as "CD3 = cost of delay ÷ duration. Highest score ships first."

Black Swan Farming, the site Joshua J. Arnold publishes his Cost of Delay material on, defines CD3 as "Cost of Delay Divided by Duration is a prioritisation/scheduling method that maximises the value delivered in a given time period when you have limited capacity," and spells out the mechanic as follows: "the priority order of features (or initiatives/projects) is determined by dividing the estimated Cost of Delay by some estimate of duration: the higher the resulting score, the higher the priority." The expansion of the acronym, the direction of the division and the "highest score first" rule in the video all match that page word for word in substance.

The video contains no numbers at all, so there is no arithmetic in it to check. The primary source does supply a worked example, and it is worth stating because it shows the size of the effect the video gestures at without quantifying. Black Swan Farming compares three features: Feature A at a cost of delay of $1,000 per week with a 5 week duration, giving a CD3 score of 200; Feature B at $4,000 per week over 1 week, giving 4,000; Feature C at $5,000 per week over 2 weeks, giving 2,500. Running them first-in-first-out in the order A, B, C accrues $69,000 of total delay cost. Running them in CD3 order, B then C then A, accrues $27,000, a 61 percent reduction. Each of those three scores is a plain division of the stated cost of delay by the stated duration, and all three check out: 1,000 ÷ 5 = 200, 4,000 ÷ 1 = 4,000, 5,000 ÷ 2 = 2,500.

Correction to a common framing: CD3 is a specific form of WSJF, not an alternative to it

The video never names WSJF, and the brief I was given asked me to establish who originated CD3 versus WSJF in case the two were being conflated. The chronology from the primary sources runs in one direction and is unambiguous.

Donald G. Reinertsen published The Principles of Product Development Flow through Celeritas Publishing in 2009, ISBN-13 9781935401001, ISBN-10 1935401009, per the Open Library catalogue record for the edition. That book is where the economic sequencing argument is made and where Weighted Shortest Job First is set out as cost of delay divided by duration. Scaled Agile's own WSJF guidance page credits him directly, citing "Don Reinertsen, The Principles of Product Development Flow" and opening with his line, "If you only quantify one thing, quantify the Cost of Delay."

CD3 is the later, narrower label. Black Swan Farming's CD3 page says so in its own words: "CD3 is one specific form of the 'WSJF – Weighted Shortest Job First' queuing method." So CD3 and Reinertsen's original WSJF are the same arithmetic under two names, and the correct reading is that Arnold's CD3 is a practitioner convention sitting inside Reinertsen's queuing method rather than a competing framework.

What genuinely differs is SAFe's variant. The Scaled Agile Framework guidance states that "WSJF is estimated as the relative cost of delay divided by the relative job duration" and identifies four estimation factors: "relative user and business value, time criticality, risk reduction and/or opportunity enablement, and job size." That is a departure from both Reinertsen and Arnold in one specific respect: SAFe builds its numerator by summing three relative scores rather than estimating a cost of delay in currency per unit of time. A secondary explainer at agility-at-scale states that the four inputs are scored on a modified Fibonacci scale of 1, 2, 3, 5, 8, 13, 20 against a reference item assigned 1, and attributes the formalisation into SAFe to Dean Leffingwell in 2012. I could not confirm that date or that attribution on a Scaled Agile first-party page, because the detailed section of the SAFe WSJF guidance is behind a login, so treat the 2012 date as unverified and the relative-Fibonacci mechanic as sourced to a secondary explainer.

The practical consequence for a viewer following this video: if you ask an AI tool for "CD3 scores" and it returns a 1-to-10 value-plus-urgency-plus-risk sum over story points, you have been handed SAFe-style WSJF under the CD3 label, which is a different estimation discipline from the one the video's own formula describes.

Correction: "rough effort tags" are not duration, and the distinction is load-bearing

At 12.66 s the video says, "So you paste your backlog, ticket titles, rough effort tags, any revenue or support cost signals you have, and ask AI to compute CD3 scores and surface the top contenders."

Effort and duration are different quantities, and substituting one for the other changes the ranking. The agility-at-scale WSJF explainer puts the distinction plainly: "The denominator is job duration, a proxy for how long the work takes relative to other items in the backlog, not effort or cost," and illustrates it with a six-person team working two weeks, which is twelve person-weeks of effort but only two weeks of duration. The quantity that matters to the economics is the calendar time the value stream is occupied, because that is the window over which the cost of delay on everything queued behind the item keeps accruing.

This matters most where effort and duration diverge hardest. A one-day change blocked for three weeks on a vendor approval is a low-effort, long-duration item, and its CD3 score falls when you score it honestly on duration and rises misleadingly when you score it on effort. A large parallelisable build staffed by eight people is high-effort and short-duration, and the opposite distortion applies. Black Swan Farming's own at-scale writing takes duration to mean elapsed time explicitly, defining the denominator for a multi-team option as the longest duration among all the teams working on it, which is a calendar-time construct that has no sensible reading as summed effort.

The video's own closing line, at 54.26 s, is that "a ranked list of half-formed tickets is still a mess," which is an accurate warning about input quality. The input list it recommends thirty seconds earlier contains the exact substitution that produces the mess.

Partial correction: the inputs listed do not contain a cost of delay

The numerator in CD3 is a cost of delay. Black Swan Farming defines Cost of Delay formally as "the partial derivative of the total expected value with respect to time," and credits Reinertsen for the concept, quoting him: "Cost of Delay is the golden key that unlocks many doors." In practice that means a rate, money per unit of time, which is what makes the division by a duration in the same time units produce a meaningful score. The worked example on the CD3 page is consistent: every numerator is stated as dollars per week.

The video's inputs are "ticket titles, rough effort tags, any revenue or support cost signals you have." Ticket titles carry no economics. Effort tags belong in the denominator and are the wrong quantity there, as covered above. Revenue and support cost signals are the only inputs in that list that could feed a numerator, and the phrase "any ... you have" concedes that they may be absent. Nothing in the list is a per-week or per-month delay rate, so an AI asked to work from it is estimating the numerator rather than computing it.

The Maersk Line experience report that Arnold and Özlem Yüce wrote up for Agile 2013, published at Black Swan Farming, shows what the real version costs in effort. The numbers in that paper come from doing the quantification: cost of delay for a sample feature of roughly $210,000 per week, end-to-end lead time cut from a 150 day average to 108 days on one programme and 60 days on another, return on investment of $26 and $44 per dollar invested against a $4.1 baseline, 22 percent higher throughput, and a finding that the most valuable quarter of requirements was worth around three orders of magnitude more than the bottom quarter. That last figure is the actual argument for ranking by CD3, and arriving at it required estimating delay costs in money per week rather than pasting ticket titles into a chat window.

Clarification: the primary source warns against defending the score itself

At 2.70 s the video promises CD3 "turns gut feel into a number you can defend," and at 48.04 s it says "The output is something you can take into a planning meeting and explain, not a gut call dressed up as a number." The on-screen DEFEND block says the same thing: "A ranking you can explain — not gut feel dressed as a number."

Black Swan Farming's page on CD3 and proportionality narrows what is defensible. It describes CD3 as "change in outcomes per unit time, per unit time," an acceleration rather than a magnitude, and warns that "the difference in outcomes isn't proportional to the CD3 scores" when options are delivered sequentially, because "the change isn't linear" and "timeframes are different." Proportionality would hold only "if the change was linear and over an equal timeframe." Its conclusion is a ranking claim and not a score claim: "You should still prioritise the option that has the highest CD3 though, as this will accelerate your desired outcome the most."

So a CD3 score of 4,000 against a score of 2,000 supports the statement that the first item should go first. It does not support the statement that the first item is worth twice as much. The video's own caution at 25.04 s, "What you get back is a ranked shortlist, not a finalist," lands on the right side of this distinction, but the repeated framing of CD3 as a number you defend points the other way, and a product owner who defends the magnitude in a planning meeting is defending something the source says is not proportional to outcomes.

The card says Lesson 03 of 10 while the narrator says lesson four, and no AI tool appears on screen

Two internal inconsistencies are visible in the artifact itself and need no external source.

First, the lesson number. The header in all eleven frames I read reads LESSON 03 on the left and 03 / 10 on the right. The narration at 8.40 s says, "This is lesson four of 10 in AI for product owners," and the burned-in caption in frame 005 at roughly 8 s renders it as "This is lesson four of ten." The card and the voiceover disagree by one. Nothing in the metadata resolves which is correct: the description, the title and the hashtags give no series position.

Second, the demonstration. The COMPUTE block promises "Paste backlog → AI computes CD3 scores → ranked shortlist," and at 17.54 s the narration describes asking an AI to compute CD3 scores and surface top contenders. Across frames 001, 003, 005, 007, 009, 012, 016, 020, 025, 030 and 034 the visual composition never changes. There is no screen recording, no chat transcript, no prompt text, no model name, no spreadsheet and no ranked output. The video does not identify which AI tool it means, and the claim that an AI can compute CD3 scores from the listed inputs is never shown working. Given that the listed inputs do not include a cost of delay rate and do include effort in place of duration, the unshown step is the one that would have exposed the problem.

Key Takeaways

  • Verified: The video's formula is correct. CD3 expands to Cost of Delay Divided by Duration, higher scores rank first, and Black Swan Farming's CD3 page states both the expansion and the direction of the division in the same terms the video uses.
  • Context: Reinertsen's The Principles of Product Development Flow (Celeritas Publishing, 2009, ISBN-13 9781935401001) is where the cost-of-delay-over-duration sequencing rule originates as Weighted Shortest Job First. Scaled Agile's WSJF page cites that book directly.
  • Clarification: CD3 is not a rival to WSJF. Black Swan Farming states that "CD3 is one specific form of the 'WSJF – Weighted Shortest Job First' queuing method." Joshua J. Arnold's contribution is the CD3 label and the practitioner discipline around it, not the underlying arithmetic.
  • Clarification: SAFe's WSJF is the variant that differs. Its published guidance sums relative user-business value, time criticality, and risk reduction or opportunity enablement to form the numerator, and lists job size among its four factors, rather than estimating a cost of delay in currency per unit time.
  • Correction: The video at 12.66 s lists "rough effort tags" among the inputs for computing CD3. The denominator in CD3 is duration, meaning elapsed calendar time, not effort. Black Swan Farming's at-scale guidance takes the denominator as the longest duration across the teams involved, which is an elapsed-time quantity.
  • Partial correction: The video's input list contains no cost of delay. Cost of Delay is defined by the primary source as the partial derivative of total expected value with respect to time, and is used in the worked example as dollars per week. Ticket titles, effort tags and whatever revenue signals happen to be available do not constitute that rate.
  • Clarification: CD3 scores support an ordering, not a magnitude comparison. Black Swan Farming states that "the difference in outcomes isn't proportional to the CD3 scores," which limits what the video's "a number you can defend" framing can actually defend.
  • Context: The worked example the video omits is on the source page. Features at $1,000 per week over 5 weeks, $4,000 per week over 1 week and $5,000 per week over 2 weeks score 200, 4,000 and 2,500. FIFO ordering costs $69,000 in delay, CD3 ordering costs $27,000, a 61 percent reduction.
  • Context: The Maersk Line case Arnold and Özlem Yüce reported for Agile 2013 records roughly $210,000 per week of delay cost on a sample feature, lead time down from a 150 day average to 108 and 60 days, ROI of $26 and $44 per dollar against a $4.1 baseline, and 22 percent higher throughput.
  • Correction: The on-screen header reads LESSON 03 and 03 / 10 while the narration at 8.40 s and the burned-in caption both say "lesson four of ten." The video contradicts itself on its own series position.
  • Unverified: No AI tool, prompt, model name, backlog table, score column or output of any kind appears in frames 001, 003, 005, 007, 009, 012, 016, 020, 025, 030 or 034. The central workflow claim is asserted and never demonstrated.
  • Unverified: The attribution of SAFe's WSJF formalisation to Dean Leffingwell in 2012 comes from a secondary explainer. The detailed section of Scaled Agile's own WSJF page requires a login, so I could not confirm the date or the attribution against a first-party source.

Resources

Published October 2, 2026. Writeup generated from a favorited TikTok.