Pov

/pov — Point-of-View Synthesis

/pov is a turn manager, not a one-shot. Each turn produces raw output, declares its successor, and stops. This is unfolding — prevents the model from simulating conclusions without doing the generative work.

When to use

• Decision synthesis between 2-N options
• Strategic alignment across competing perspectives
• Terminology lockdown, IR design, vocabulary choice
• Conflict resolution where framing diversity > model independence

When NOT to use

• Facts under dispute → /cdo (model independence catches blind spots)
• Code review / audit → /review or /security-review
• Single-chain reasoning → /think
• Parallel model comparison → /tribunal

The 4-turn protocol

Turn 1 — Unfold

Print 20-35 raw POVs. No synthesis. Each item:

• 1 sentence, ≤ 20 words
• Framed through a lens (persona or method)
• Marked (safe) | (wildcard) | (tbd)

Lenses — pick 6-10 per invocation:

• systems_thinking, first_principles, figure_storming
• 10x_dev, elon_musk, karpathy, sam_altman
• buffet_bezos_dimon, einstein_turing
• jesus_gandhi_god
• pre_mortem, reverse_brainstorm, second_order, opportunity_cost

Exit gate: declare Turn 2 is next. Stop.

Turn 2 — Synthesize

• Conflicts — find pairs/groups that genuinely oppose. Cite items by number. State the opposition evidenced.
• Convergences — find items where 3+ agree on the same direction. State the signal strength.
• Resolution hints — for each conflict, suggest "resolves as X" with evidence.

Exit gate: declare Turn 3 is next. Stop.

Turn 3 — Reduce

Collapse to ONE of:

• 1 concrete sequence — numbered steps with exit gates per step
• 1 axiom — compressed invariant that makes the decision self-evident

Exit gate: declare Turn 4 is next. Stop.

Turn 4 — Conclude

• Top-expert recommendation — name the domain, cite what practitioners in that field would do
• Evidence anchors — link back to Turn 2 convergences
• Exit menu — max 5 next-step options using the follow-up template

Exit gate: /pov cycle complete.

Axioms

1. Forced-framing diversity > persona identity — one model producing N frames matches N models producing N similar frames, at 1/N the cost.
2. Synthesis before enumeration — 30 items without synthesis is noise; synthesis without 30 items is bias.
3. Name theater before claiming rigor — call out which thinking systems actually fired vs which were prompt-stems.
4. Unfolding prevents simulation — compressed turns let the model hallucinate conclusions without doing the generative work. Turn boundaries force real output.

Costs & limits

• Tokens: ~2000-4000 per full cycle
• Time: 4 turns, no subagent round-trip
• Blind-spot risk: high (single model) — do not use for facts under dispute

vs other skills

Worked example

2026-04-18 dogfood decision: /pov decided whether to ship /pov.

• Capture: .lev/pm/workstreams/lifecycle-engine/captures/20260418-pov-pattern-and-c2-licensing.yaml
• 4 turns · 30 POVs → 4 conflicts + 5 convergences → 5-step sequence → Unix-philosophy conclusion
• Self-validation: /pov reasoning about /pov ship produced an actionable decision inline, single session, no subagents

Provenance

Discovered organically during DNA-hardening work (2026-04-16 → 2026-04-19) as an emergent pattern in next-steps menus. Formalized after the user recognized the pattern outperformed /cdo for synthesis tasks at ~5% of orchestration cost.