bottleneck

What this skill does

Finds the ONE constraint that governs everything else, using the Theory of Constraints (Goldratt), and turns it into an action strategy. The premise: a business is a pipeline, and total throughput equals the narrowest stage — Jack's Lamborghini example: 100 bonnets/week, 500 paint jobs/week, 1 engine/week → you ship 1 car/week, and getting better at paint changes nothing. Working on anything except the constraint is motion, not progress. (The pipe is a diagnostic model, not a literal equality — real stages have conversion rates, lag, and demand-quality effects; the point is finding where flow narrows most, not computing exact throughput.)

Source: Jack Roberts' Theory of Constraints business review (video 1DOLq8xy0qI, 2026-07-04, knowledge/jack-fable5-five-things-1DOLq8xy0qI.md [12:49-14:04]); his exact prompt is gated in his Skool group, so this implements the method from the video. Servy twist: we don't interview for numbers we already have — read the ledgers first.

Gate 0 — scope check (hard)

Personal and Penguin Alley / freelance business only. If the ask is about SoftServe strategy, team throughput, Jumpstart pipeline, or anything Internal+ → refuse and point to Claude Desktop (Enterprise) / Cowork. Same rail as every money session.

Step 1 — ground in the real numbers (before any question)

Read, in order:

1. money/MISSION-1M.md — the $1M ledger; only verified dollars count (standing rule for money sessions).
2. money/GERA-NOW.md + latest money/audit-*.json if present.
3. projects/REGISTRY.md — what's shipped, what state it's in.
4. Latest PA CEO brief: ls pa/ceo-briefs/*.md 2>/dev/null | sort | tail -1 — if none exist, mark PA-brief evidence UNKNOWN and move on (never fabricate the org's read).
5. context/priorities.md — this quarter's stated focus.
6. pending.md — open loops (hours leak into these).

Build the pipeline table from evidence, marking each stage KNOWN (with the number + source) or UNKNOWN:

Step 2 — fill only the gaps (short interview)

Ask ONLY about UNKNOWN stages — one question at a time, max ~5 questions, each with your best-guess answer attached so Gera can confirm or correct. Do not re-ask what the ledgers already say.

When UNKNOWN cells dominate the table, open with Nate's 5x diagnostic (6h course [2:13:30]): "If tomorrow you got 5x the customers going through your pipe, what would break first?" — it forces a mental walk of the day-to-day and surfaces where flow stops faster than stage-by-stage questions. Flag: this is the client-facing wording — the phrasing to reuse when scoping constraints with a client (Jumpstart conversations included; the wording travels even though this skill's data never does).

Step 3 — name THE constraint

Pick one stage. Not a top-3, not "several areas need work" — the whole ToC review turns on naming a single constraint. State it with the evidence chain: "Throughput is capped at X because [stage] only does Y, while every stage before it can do 10Y." If two stages genuinely tie, pick the one earlier in the pipeline and say why.

UNKNOWN-heavy guard: if the stages critical to the call are still UNKNOWN after the interview, do NOT manufacture certainty — name a provisional constraint (labeled as such, with confidence low/med) and make this week's first move instrumentation (start measuring the unknown stage), not optimization. "Evidence over vibes" outranks "pick one."

Sanity check the classic trap: Hours is usually the hidden constraint for a solo operator with a day job — verify whether the named constraint would still bind if Gera had 20 more hours/week. If not, Hours is the real constraint.

Step 4 — the five focusing steps → action strategy

Apply Goldratt's sequence to the named constraint:

1. Identify — done (step 3).
2. Exploit — squeeze the constraint as-is: what gets 20% more out of it with zero new resources this week?
3. Subordinate — everything else serves the constraint: what should Gera STOP doing because it feeds a non-constraint? (Be specific — name current activities from pending.md/priorities that don't feed the constraint.)
4. Elevate — only if exploit+subordinate aren't enough: what investment (automation, delegation to the PA agents, a tool, money) widens the constraint?
5. Repeat warning — name the likely NEXT constraint once this one breaks, so the fix doesn't overshoot.

Default Shift applies at every step: for each action, ask "to what extent could AI/the PA agent org do this instead of Gera's hours?"

Output

Write money/bottleneck/{YYYY-MM-DD}-review.md (create the dir if needed) with: the pipeline table + evidence, the named constraint + reasoning, the 5-step strategy, and this week's 3 moves (concrete, constraint-serving, small enough to start now). Summarize in chat leading with the constraint, not the process.

Then:

• Offer to log the strategy decision to decisions/log.md (Gera decides — suggest, don't auto-log).
• Append deferred moves to pending.md tagged [gera]/[servy].
• If the constraint implicates a PA agent's lane (e.g. Leads → pa-outreach gate, Conversion → pa-proposal-writer), say which agent owns the follow-through.

Notes / guardrails

• Evidence over vibes: every number in the pipeline table needs a source (file or Gera's explicit answer). No invented percentages — UNKNOWN is a valid and useful cell.
• One constraint. Resisting the urge to fix everything IS the method. If Gera pushes for a broader list, deliver the one constraint first, then park the rest in pending.md.
• Not /roast, not /audit: /roast attacks an idea before building; /audit scores the AIOS itself (Four Cs); /bottleneck analyzes the operating business's flow. If Gera brings a new idea to bottleneck, route to /roast.
• Money data stays where it lives: amounts and client names in money/ (committed, private repo) are fine; never copy them into public-facing artifacts or posts. Minimize client identifiers even internally (first name or initials once context is clear), respect NDAs/platform ToS (Workana names stay in money/), and sanitize anything that leaves money/bottleneck/ — a pending.md reminder says "follow up on the July constraint plan", not the client or the number.
• Cadence: worth re-running monthly, or after any constraint-breaking event (first paid client, a product shipping, schedule change).
• Effort: ledger-reading and gap questions run cheap (low/default); the constraint call and strategy are the "first 20% that sets all the quality" — escalate to high there (Jack's rent-Einstein rule).