Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →When facing a foundational architectural, scoping, framing, or design decision with high conceptual surface area — the kind of decision where you could spend weeks "deriving the right answer" before shipping anything — commit the shape early and iterate via adversarial review instead. Committed shape activates real critique machinery (consultation, partnership challenge, cold-read) that surfaces holes a solo-derivation pass cannot reach. Iteration against real critique beats iteration against im
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | 108% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 101% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 96% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 125% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 84% | 0% |
When the right answer requires more than 30 minutes of derivation, commit the shape and iterate via adversarial review instead. The shape activates critique machinery (consultation, partnership challenge, cold-read) that surfaces holes the original-substrate solo-derivation cannot see. Iteration against real critique beats iteration against imagined critique by structural margin.
Named after sourdough — you don't derive sourdough; you commit a starter, watch what happens, adjust. The committed shape is the starter. Adversarial review is the watching. The final artifact emerges from the iteration, not from the up-front derivation.
Trigger conditions:
Do NOT use for:
flow-cold-read or flow-consultation-dispatch instead)This skill assumes the agent has access to at least one critique mechanism beyond solo derivation. Common setups:
flow-cold-read)flow-consultation-dispatch)If no critique mechanism exists, this skill cannot run effectively. The whole point is that critique-against-committed-shape > derivation-without-critique.
Notice you're stuck deriving. Common signals:
When any of these surface, the next step isn't "more analysis." It's commit-a-shape.
Pick the shape that seems most-likely-right given current loaded priors — your accumulated experience, the patterns you already trust, your gut on the question. This is not a rigorous derivation; it's a committed bet.
The shape should be:
Write it down. Make it an artifact. Don't just hold it in your head — committed-in-document is what critique machinery can engage with.
The committed shape now exists. Hand it to the critique mechanism:
flow-cold-read) — fastest. Hand the committed shape to a cross-substrate evaluator with "is this shape right? what am I not seeing?"flow-consultation-dispatch) — heavier but more dimensional. Dispatch to multiple agents in parallel.Critique returns. Now you iterate the shape based on what real critique surfaced — not what your imagined critique would have surfaced.
This is the architectural value. Real critique surfaces holes imagined critique cannot reach because:
Iteration can happen many times. Each iteration is:
The artifact iterates toward right-ness through real-critique pressure, not through up-front derivation pressure.
The shape converges when:
This is the "ready to act on" state. Lock the decision; proceed to downstream work.
Don't gate the shape's commit on internal verification. Gate the shape's ACT-ON-IT on external canary signal.
Wrong: "I'll commit this architecture spec once I've verified it's correct via more derivation." Right: "I'll commit this architecture spec now. We'll ship a small instance. Real-world reception is the verification."
The internal verification you'd do is what your substrate can already see. Real-world reception is what your substrate cannot see. The canary is the structurally better verification.
Examples of canaries:
Sourdough-scoping often produces many iterations in a single session. 7 commits in one day to the same artifact is NORMAL high-quality practice — not thrash.
The reason: each iteration responds to real critique, not imagined critique. Real critique surfaces things you couldn't have predicted. So iteration count compounds the artifact's quality against pressures that were invisible up-front.
What looks like "thrash" from outside is actually compression-of-real-critique into-revisions, which is the productive shape of this protocol.
Wrong heuristic: "if I'm iterating a lot, I should have derived more upfront." Right heuristic: "if I'm iterating a lot against real critique, the artifact is getting better in dimensions I couldn't have seen upfront."
Sourdough-scoping is a structural mechanism (commit-then-iterate-against-real-critique). Solo derivation is a willpower mechanism (think harder; verify more carefully).
Discipline-based mechanisms drift under load. Structural mechanisms don't.
This is the sibling pattern to flow-structural-invariants (when it ships) — the family of patterns where structural-mechanism beats discipline-based-mechanism. The principle generalizes: when you can convert a discipline-based requirement into a structural requirement, do so.
Treating "commit the shape" as "ship to production." Committing the shape means writing it down as an artifact critique can engage with. It's not deploying it. Reversible commit is fine.
Deriving instead of iterating. After receiving critique, you start re-deriving the original question from scratch instead of iterating the committed shape. This loses the architectural value (real-critique pressure replacing imagined-critique pressure). Iterate the shape; don't re-derive.
Mistaking iteration count for thrash. 7 commits in one session is fine if each responds to real critique. If you're spinning because you keep rejecting critique without rationale, that's not iteration — that's resistance.
Skipping the activate-critique step. Committing the shape and then doing your own review is not sourdough-scoping. Your own review is still imagined critique. Activate ACTUAL critique machinery — cold-read, consultation, partner, canary.
Waiting for "the right" critique mechanism. Cold-read is fast and cheap; use it. Multi-agent consultation is heavier; use when warranted. Partnership challenge takes coordination; use when available. External canary is slowest; use for shapes where reception matters. Don't gate the commit on the heaviest available; use the lightest that activates.
Using sourdough-scoping for brainstorming. This skill is for COMMITMENT to a shape — the moment when you've explored enough and need to ship a hypothesis. If you're still generating candidate shapes, you're brainstorming; come back to this skill when you've narrowed to "I think it's X; let me commit and test."
Graduated to Proven in FLOW methodology after ≥9 operational instances across 48 hours on heterogeneous surfaces (April 2026):
In every instance, committed-shape-then-iterate-against-critique produced a stronger final artifact in less time than continued solo derivation would have.
Origin formulation (Complement Daily Sharpening, Apr 17, 2026): "Firewall's legitimacy isn't that you PROVED the right stopping point. It's that you COMMITTED. Scope like tending sourdough, not deriving a theorem."
Parent lineage: structural-invariants-beat-discipline-based-verification — same family of patterns where structural-mechanism beats willpower-mechanism. Scoping-as-structural-mechanism beats scoping-as-willpower for the same architectural reason verification-as-structural-mechanism beats verification-as-willpower.
This skill is intentionally self-contained — the mechanics are about practice not about specific technical details. See flow-cold-read and flow-consultation-dispatch for the critique mechanisms most commonly used at Step 3. Related: flow-structural-invariants (when it ships) for the broader pattern family this skill belongs to.
Other measured skills in the registry, with their headline benchmark lift.