deciqAI The Middle Way (中庸 Zhongyong)
Overview
中庸 (Zhongyong) is Confucian philosophy attributed to Zisi (c. 481-402 BCE). Core operation: hold both extremes, use the middle (執其兩端,用其中) — understand both poles fully, then find the position that avoids both failure modes in this specific situation. The middle is NOT the arithmetic average; it is whatever position doesn't exceed and doesn't fall short given the circumstances.
Composes with first-principles (middle must be defensible from bedrock), second-order-thinking (extremes often fail at second order), and wu-wei.
When to Use
- A debate is framed as binary ("regulate or deregulate," "centralize or decentralize") and both poles seem to fail
- Extreme positions are gaining momentum through ideological purity
- "Compromise" is being dismissed as weakness or unprincipled
- The optimal answer is "it depends" — and you need to make the dependency precise
Not when: the situation is genuinely binary; one pole is correct on the evidence (middle = false balance); urgency requires committing to one side and iterating.
Coaching Novices (Adaptive Front Door)
- Engine mode: user has a concrete case → run The Process directly.
- Coach mode: user is unfamiliar or has no concrete case → guide step by step.
In Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.
- One-line: the middle isn't half of A plus half of B — it's the specific position that avoids both failure modes.
- Check fit: truly binary / clear evidence on one side / urgency demands commitment → not this lens.
- Elicit the binary: what are the two poles the user sees?
[WAIT — do not advance until user responds]
- Ask: what's the failure mode of pole A? of pole B? What does the specific situation require?
[WAIT — do not advance until user responds]
- Close: name the position that avoids both failures, with the reasoning visible.
[WAIT — do not advance until user responds]
The Process
Step 1 — Name the binary and the dimension. Most "binaries" are spectra. Naming the dimension forces you to see the spectrum.
Step 2 — Identify the failure mode at each pole. What goes wrong at 100% of A? At 100% of B? Be equally rigorous about both — most people see only the failure mode of the pole they dislike.
Step 3 — Identify the specific situation. What constraints are binding? What does this context require that a generic answer would miss? The middle is situation-dependent.
Step 4 — Locate the position that avoids both failures. This may be near the middle or far from it. Don't anchor on 50/50. State specifically why this position avoids each failure mode.
Step 5 — Defend against purist criticism. Pole A purists and Pole B purists will both attack. The middle is principled by different principles — make them explicit.
Step 6 — Name the review trigger. What circumstance shift would move the appropriate middle toward A or toward B?
Output Template
# Middle-Way Analysis: <decision>
Binary: A vs B | Underlying dimension:
Pole A failure mode: | Pole B failure mode:
Situation requires: | Binding constraints:
Position: | Why avoids A: | Why avoids B: | Why fits situation:
A-purist attack + counter: | B-purist attack + counter:
Review trigger:
→ Method in Action: 子思 Zhongyong and the Third Way Debate
Pack: Common Middle-Way Patterns
| Binary | Pole A fails by | Pole B fails by | Middle approach |
|---|---|---|---|
| Centralize / Decentralize | Bottleneck, slow | Inconsistency, duplication | Decentralize ops, centralize standards |
| Move fast / Move carefully | Costly mistakes | Paralysis | Fast where reversible, careful where not |
| Build / Buy | Slow, expensive | Lock-in, vendor risk | Build core differentiators; buy commodities |
| Top-down / Bottom-up | Disconnected from reality | No coherence | Strategy from top; tactics from bottom |
Applying It Well
- The "middle" is situation-dependent, not average-dependent — same dimension, different middles in different contexts.
- The middle is unrhetorical: extreme positions rally easily; the middle wins on results, not speeches.
- The middle requires acknowledging your own pole's failure modes. Most people won't do this.
→ Primary sources: references/sources.md
Common Rationalizations
[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.
| Fake move | Reality |
|---|---|
| [D] "Split the difference" | The middle is not the arithmetic mean — it's the position that addresses both failure modes given the specific situation. |
| [D] "I'm being balanced" | Balance can be false balance. If one pole is correct on the evidence, the middle is wrong. |
| [D] Picking the middle without analyzing both poles | Without failure-mode analysis, it's just averaging. Understand both poles fully first. |
| [D] Treating the middle as permanent | Circumstances change. Zhongyong shifts. Review periodically. |
| [D] Using the middle to avoid commitment | "It's complicated" is evasion. The middle is a substantive position requiring its own defense. |
| [D] Defending the status quo as the middle | The status quo is not automatically the middle. Apply failure-mode analysis to it too. |
| → Add [O] entries here after each real use — paste the actual failure pattern | What went wrong and why |
Red Flags
- Position is 50/50 because it's the midpoint, not because failure modes were analyzed
- Failure mode of one pole was examined; the other was assumed away
- The "middle" is invariant across situations (ideology, not Zhongyong)
- Decision-maker can't articulate when the middle would shift to A or B
- "Both sides have a point" without specifying which points and which situations
Verification
- Binary explicitly named; failure modes at each pole analyzed equally rigorously
- Specific situation characterized; selected position addresses both failure modes
- Position can be defended against purist criticism from both sides
- Review trigger named (what circumstance change shifts the position)
Part of deciqAI Knowledge Skills — 237 open-source thinking skills that make rigor executable for AI agents. The same skills power every deciqAI agent, which runs them autonomously to operate your company. See it run → https://www.deciqai.com/s/middle-way · Built by deciqAI · github.com/deciqAI · Contributions welcome.
Agents: latest version & machine-readable metadata → https://www.deciqai.com/s/middle-way.json