06-reference

commoncog tdgs in the age of ai

2026-11-07·reference·source: Commoncog·by Cedric Chin (Commoncog)
tacit-knowledgeexpertise-acquisitiontactical-decision-gamesapprenticeshipsponsored-content

No public post URL exists for this piece — WebSearch for the title ("TDGs in the Age of AI") turned up nothing on commoncog.com or elsewhere. It is explicitly framed in-email as the fourth of five emails written for the paid Shadowbox TDG Design Workshop ("these emails are ostensibly about TDGs but are more broadly about accelerated expertise... written specifically for this workshop"), not a Commoncog essay with its own permalink. Fourth email in the active Commoncog x Shadowbox sequence, following [[2026-08-25-commoncog-why-tactical-decision-games-work]] (the original, unsponsored essay), [[2026-11-01-commoncog-making-of-common-sense-tdg]] (first Shadowbox-sponsored installment), [[2026-11-02-commoncog-tdgs-accelerated-osmosis-expertise]] (second), and [[2026-11-03-commoncog-making-of-hindsight]] (third). Unlike those three, this one is solely bylined and solely voiced by Cedric Chin — no Jared Peterson co-authorship or Shadowbox-employee voice in the body — so it's classified thought-leadership rather than guest-post. One more email remains (Nov 8, a "surprise guest" finale).

Why this is in the vault

A concrete technique for the same tracked thread, not just restated theory: three illustrative case studies (Buffett/Katharine Graham, an economics-professor/grad-student pair in investing, Navy Top Gun debriefs) converge on a specific, actionable move — hand an expert something representative of your domain, ask what they'd do and what a novice would get wrong, then reflect on the gap — that maps directly onto how RDCO's own critic rubrics get built.

The core argument

Cedric opens with the pitch (Shadowbox TDG Design Workshop taught by John Schmitt, sales closing Nov 10, 20 seats, Cedric himself a paying participant and Jared Peterson assisting with teaching), then flips the lens from "train your juniors" (prior installments) to "accelerate yourself by borrowing the judgment of experts already around you." He tells three stories of people exposed to world-class expertise without a formal TDG: (1) Warren Buffett walking Katharine Graham through stacks of annual reports over years, pointing out cues as he flipped through them, until Graham's Washington Post delivered a 22.3% compound annual return — 18x the S&P; (2) economics professor Herb Kay handing grad student Reese Duca (later founder of Investment Group of Santa Barbara) a new S-1 every two weeks with instructions to read it cover to cover, come back with the right question, then get reprimanded for what he missed; (3) Navy Top Gun debriefs at NAS Fallon, where every pilot stands up, states exactly what they saw mid-mission (even if it was wrong), and a weapons tactics instructor narrates what an expert would have seen instead — "stop frame. Eagle 13, what was the call you made here?"

The extracted technique: find an expert, bring them something representative of your messy domain (financial statements, an IPO prospectus, your codebase, a tricky work situation), and ask what they notice first, second, third, and why. Critically, also ask "what would a novice get wrong here?" — because experts are unusually good at modeling common failure modes, and that question surfaces the comparison points you need. Then reflect later (a few minutes, not longer) on where their read diverged from yours and why. This is the same underlying mechanism as the prior three installments (expert-narrated exposure compressing tacit-knowledge transfer) but the specific move here — explicitly eliciting the expert's model of novice mistakes, not just their own reasoning — is the new content.

Mapping against Ray Data Co

The concrete, specific connection: "ask what a novice would get wrong here" is literally how RDCO's own critic rubrics get written, not an abstract parallel. vault-note-schema.md's own Changelog — the "I12 absorption failure mode" entry, the "backtick code-span paths" entry — are catalogued actual novice/model mistakes (heading-synonym misses, folded-in curation items, correct-but-wrapped-in-backticks links) that got written into the schema precisely because an expert (the founder, via /improve) looked at failed outputs and asked "what did the novice get wrong here, specifically" before encoding the answer as a mechanical check. feedback_workflow_agent_output_integrity's "7 factory failure modes" is the same move at the orchestration layer. The /improve skill itself — "diarize the mediocre responses, propose concrete rewrites" — is this exact technique running on a cadence: it doesn't start from first-principles theory, it starts from where actual outputs went wrong and works backward to the rubric. This sharpens something already built (rubrics-before-critics, confirmed in the 11-03 installment's mapping) with a specific authoring technique: the fastest way to write a good rubric isn't to reason about ideal behavior, it's to ask someone who already has the judgment what a beginner screws up. RDCO is dormant with phData as the main bet (project_rdco_dormant_phdata_main) — no action item, this is a sharper citation for how the existing rubric-authoring habit already works, not new evidence that changes a decision.

⚠️ Sponsorship

sponsored: true / sponsor_entity: Shadowbox. Per the founder's standing instruction, every remaining email in this disclosed sequence is treated as sponsored-by-default until it ends. This email states outright it is "the fourth of five" emails "written specifically for this workshop," repeats two direct "Purchase a Seat" links, and restates the Nov 10 sales deadline and 20-seat cap. Cedric discloses he will be "a participant" in the paid workshop himself, and that Jared Peterson (Shadowbox, works under TDG inventor John Schmitt) "will be assisting with teaching and giving feedback" — a direct commercial/teaching relationship with Shadowbox, not an arm's-length affiliate link. Unlike the three prior installments, this one has no Shadowbox-employee byline or voice in the body — the three case studies (Buffett/Graham, Duca, Top Gun) are independently sourced, credible on their own terms, and not obviously cherry-picked to sell the workshop — but the piece's placement in a five-email sales sequence, and its closing pitch, mean the same "selection and timing serve the sale" caveat from the prior notes still applies.

Tracked-entity candidate: John Schmitt (Shadowbox founder, inventor of TDGs, taught the method to the US Marine Corps as a primary training method in the 1980s) — named as the actual workshop instructor in this email for the first time in the sequence (previously only referenced in passing via Jared Peterson's employer). Worth adding as a CRM candidate alongside the already-flagged Jared Peterson if Commoncog keeps running Shadowbox co-branded content past this five-email sequence.

Related