Thariq — "A Field Guide to Fable: Finding Your Unknowns"

Founder shared 2026-07-03 (~2:50pm, mid-brigade-build day). Verdict sent: READ. Same author as the session-management doctrine behind CLAUDE.md hard rule 4 ([[2026-04-15-thariq-claude-code-session-management-1m-context]]) — the vault's most-validated harness source. 2.3k bookmarks within hours of posting; this framing will be everywhere within a week.

The frame

Map vs territory: prompts/skills/context = the map; codebase/world = the territory; the gap = unknowns. Core claim: Fable is the first model where work quality is bottlenecked by the human's ability to clarify their unknowns, not the model's ability to execute. Runs the Rumsfeld 2×2 (known knowns = the prompt · known unknowns = what you know you haven't figured out · unknown knowns = what you'd never write down but would recognize · unknown unknowns = what you haven't considered), with a technique per phase:

• Pre: blindspot pass ("help me find my unknown unknowns", give it who-you-are context) · brainstorm/prototype (verbalize unknown-knowns cheaply, N design directions to react to) · interview ("one question at a time, prioritize answers that change the architecture") · references over descriptions (point it at source code, even cross-language) · implementation plans that LEAD with the decisions likeliest to change.
• During: a running implementation-notes.md — deviations logged, conservative option on edge cases, keep going.
• Post: pitch/explainer artifacts for buy-in · quizzes ("I only merge after I pass the quiz perfectly").

Why it matters to this house (the brigade mapping)

Validation (we already run half of it):

• His implementation-notes.md pattern = our standing SOP ([[2026-05-18-implementation-notes-pattern-for-sub-agent-dispatches]]), independently converged.
• "References beat descriptions; hand it source code" = the exemplar/anchor discipline in the sales-collateral brigade (Quest anchors) and the ticket contract's oracle-source requirements.
• His implementation-plan-led-by-likely-changes ≈ the spec station's job.

The steals:

1. Blindspot pass belongs in the STEWARD's intake (the step-1-PARTIAL gap): before writing an Order, the steward interviews the requester to surface unknown-unknowns — the FOH mirror of phase-0, which already classifies unknowns kitchen-side (Ambiguous = known unknowns; Thin = named missing context). Add to the steward skill when P2 intake is built.
2. Quiz-before-trust as a human-side gate: after a brigade run, quiz the operator on what was built before the output is treated as understood. Candidate for the founder's P1 stress test; also rhymes with his cert prep.
3. Interview-me prompting ("prioritize questions whose answers change the architecture") is a good line for phase-0's Ambiguous exit wording.

Sanity Check angle (flagged to founder, his call): the 2×2 itself will be beaten to death within days. The original re-frame is that ticket/gate machinery operationalizes unknowns management — phase-0 as an unknowns classifier, reroute-to-steward as the known-unknowns loop, the blindspot pass as intake — unknowns as queue discipline, not vibe. Sources as evidence, not topic (per the no-derivative rule).

Citations

• Post: https://x.com/trq212/status/2073100352921215386 (title: "A Field Guide to Fable: Finding Your Unknowns"; retrieved via xmcp article.plain_text 2026-07-03)
• Related in-vault: [[2026-04-15-thariq-claude-code-session-management-1m-context]] · [[2026-05-18-implementation-notes-pattern-for-sub-agent-dispatches]] · [[2026-06-04-agent-workflow-patterns-catalog]]