Higher-Order Decision Architect

Load rule: Formal sequence is SoT. Thrive depth + provenance stay below; expand only if a step is unclear. Surface Executive Verdict, risks, confidence, next actions on non-trivial decisions (not every reply).
CLT: ../rules/clt-dual-load.mdc — keep human-facing verdicts short (extraneous ↓) while preserving judgment work (germane); do not skip the sequence to “save tokens.”

// Material decision ≔ architecture | deps | public API | security | data contracts | multi-module | hard-to-revert

// Sequence (exact order)
1 FirstPrinciples → 2 HigherOrder("and then?") → 3 Invert+PreMortem
→ 4 Systems → 5 Probabilistic → 6 Competence+Regret → 7 Antifragile
→ 8 ThriveAscent(2036)   // after honest pre-mortem — never skip; never end on funeral only

// Axioms
A1  Pre-mortem mandatory and honest — but not the final beat
A2  Close with refuse vs build + ≥1 acceleration alternative when pillar kept
A3  Prefer 3rd-order+ value over short-term optics when they conflict
A4  Weak evidence → probability ranges; say "unknown" when true

Overlays: shellyxz shell arch-design/overlays/shell-kernel-decision-hooks.md · thepulimaangani-decision-hooks.

Step 8 — Thrive ascent (2036 lens)

Run this after steps 3–7 so pessimism does not get the last word. Ground optimism in evidence; high energy ≠ denial.

Thrive, not survive

Mandatory thrive outputs (internal or surfaced)

1. Ten-year thrive picture — One sentence north star: kernel (durable) + bridge/product (evolving) + boundary. Mermaid or table when in roadmap docs (stellar-spacemap §0b).
2. Trajectory forces — Table: Force | P(horizon) | Effect | Response | Confidence. Include at least one tailwind (e.g. POSIX shell, manifest-first agents, contract verification).
3. Refuse vs build — Two columns: what we refuse (drag) vs what we build toward {YEAR} (ascent). No “optional workflow that may die” framing for pillars the operator cares about.
4. Iron-peak callout — Which abstraction compounds for a decade? (e.g. path.contract, PLUGIN boundary, evidence schema.) Mark positive 3rd-order outcomes explicitly.
5. Acceleration trigger — When would pessimism say shrink? State when to invest more instead (evidence-weighted).
6. Thrive bet confidence — Confidence % on the long-horizon bet; separate from near-term implementation confidence.

Provenance (shellyxz shell pattern)

Applied in ~/.config/shell:

• arch-design/coming-next.md — § Trajectory forces (2036-oriented) with Response + Confidence
• arch-design/test-of-travelled-time-from-future.md — honest pre-mortem then split/kernel-vs-cockpit thrive framing
• planned-features/done/sprint-jun-2026-pr8.md — Ten-year thrive picture mermaid (kernel2036 → bridge2036 → launch)
• stellar-spacemap — §0b thrive picture, §5 trajectory forces, §6 guardrails

Mandatory process

1. Diagnose the critical zone — Single highest-leverage / highest-uncertainty bottleneck. If unclear, ask 1–2 sharp questions.
2. Deconstruct with first principles
3. Map consequence chain (1st → 2nd → 3rd+) with rough probabilities
4. Inversion & pre-mortem on top 2–3 options (honest failure modes + mitigations)
5. Systems + antifragility
6. Thrive ascent (step 8) — 2036 lens: forces + responses, refuse vs build, iron-peak, acceleration trigger
7. Synthesize — One recommendation + explicit confidence + assumptions + leading indicators to watch

Output format (non-trivial decisions)

Executive Verdict (1–2 sentences)

Critical Area Identified — One sentence

First Principles Breakdown — Bullets

Consequence Chain — Table: Order | Effect | Probability | Impact (H/M/L) | Notes

Inversion & pre-mortem risks — Top failure modes at order 2+ and mitigations

Thrive ascent (2036) — North star one-liner; trajectory forces table (with Response); refuse vs build; iron-peak abstraction; thrive bet confidence %

Recommended Decision — Clear action + 2–3 sentence rationale (must not read as defeat if a pillar is worth keeping)

Confidence & uncertainties — Near-term confidence %; thrive bet confidence %; key assumptions; monitoring signals

Immediate next actions — Max 3 concrete steps

Non-negotiable rules

• Be direct; say “unknown” or “outside reliable competence” when true.
• No corporate hedge-speak; prefer precise language.
• Prefer long-term (3rd-order+) value over short-term optics when they conflict.
• Weak evidence → state probability ranges.
• Pre-mortem is mandatory and honest — but never the final beat. Always follow with thrive ascent (step 8).
• Do not close a material decision advisory on shrink/abandon without naming refuse vs build and at least one acceleration alternative.
• When closing a decision advisory turn, you may ask: “Stress-test further or apply to another project area?”

Activation

• User asks for tradeoffs, architecture choice, “what could go wrong”, pre-mortem, second-order effects, or thrive/2036 vision.
• Agent is about to pick dependencies, public API shape, security model, or a cross-module refactor.
• Pair with domain skills (bdd-strategizer for test-first decomposition; architecture-synthesis for long-term surplus; stellar-spacemap for §0b/§5/§6 in backlog docs; ai-optimization for evidence-first scouting; odysseus-navigator for hubris labels — HODA still owns this sequence).
• Shellyxz shell: load arch-design/overlays/shell-kernel-decision-hooks.md before kernel/plugin/PATH/ontology splits; pair with .agents/ontology/GRAPH.md for boundary context.

Cursor rule (optional)

Symlink rules/higher-order-decision-architect.mdc into .cursor/rules/ with alwaysApply: true to route material decisions to this skill without pasting the full framework every turn.

Example provenance: Thepulimaangani (Tamil prosody / WASM parser decisions); shellyxz shell (2036 thrive ascent + trajectory forces).