A worked cycle
Before you install anything, read what you'd get back. This page shows the real artifacts from cycle 3 of the shipped exemplar — a bootstrapped SaaS company facing a venture-backed incumbent — exactly as the agent wrote them.
One /boyd-cycle pass ends with four artifacts on disk. Everything quoted
below is from
examples/orchid-saas/ in
the repository — no cherry-picked marketing copy; you can diff this page
against the files.
The situation in one breath: Orchid Workflow, a twelve-person bootstrapped legal-software company (~$1.2M ARR), is being invaded down-market by LexFlow, a venture-backed incumbent with a fresh $40M Series B and a $49/month promotional price Orchid cannot match. Cycle 3 runs on 2026-05-01, just after Orchid’s first confirmed customer loss to the promo. (Both companies are fictional; the tutorial walks the same scenario from cycle zero.)
1 · The Schwerpunkt — a focus, not a target
systems/orchid-saas/schwerpunkt.md — excerpt
Establish Orchid as the default trust-based community for solo and small-firm attorneys before LexFlow’s downmarket push gets traction.
This is a concept, not a target. It does not say “hit $X ARR” or “win Y% of the segment.” It says where effort should concentrate so that every team member can make local decisions without a phone call. […]
Competing candidates that we rejected:
- “Build the most complete feature set for solo firms” — reframes the fight as feature parity. LexFlow has more engineers; we lose on attrition.
- “Match LexFlow on pricing” — a target masquerading as focus; doesn’t tell anyone what to do beyond pricing decisions; commits us to a discount war we cannot win.
The artifact argues for its focus and shows the candidates it rejected — including the two moves (feature race, discount war) that a generic strategy-bot would happily recommend. Attrition contests are explicitly refused; that refusal comes from the doctrine, not from a lucky prompt.
2 · The layered strategy — moral before mental before physical
systems/orchid-saas/strategy.md — excerpts
Moral (lead): sharpen the gestalt of small-firm-respecting design — transparent pricing, UX simplicity, community presence — into a named identity that solos can articulate when recommending us. […]
Mental: keep our orientation tighter than theirs. […] Meanwhile, on LexFlow’s mental level: do nothing that confirms their RM1 (“brand recognition wins”). Refuse the brand-recognition fight […] Compete on a different game.
Physical: ship one coherence feature this quarter, not one parity feature.
The same file works out how to operate inside the adversary’s loop — ambiguity, novelty, and menace in forms the bigger company structurally cannot answer quickly:
- Ambiguity: the all-in-cost calculator […] Forces them into a position where either they publish enterprise pricing (against their sales motion) or they look opaque. They cannot decide quickly because the decision touches every part of their org.
- Menace: community sponsorships in spaces LexFlow’s sales team cannot access. […] They cannot win a fight they cannot see.
The moral level leads and the physical level is one line — that ordering is enforced, not stylistic. And the strategy names which of its own open assumptions it is gambling on (an open mismatch, MIS-2, with a 60-day falsification test attached).
3 · The next action — with branches, sequels, and a kill threshold
systems/orchid-saas/next-action.md — excerpts
Respond personally (founder) to the Solo Practitioner Network DM within 48 hours and propose a sponsorship structure for their virtual summit. Concurrent: begin work on the all-in-cost calculator (target ship in 14 days; assign one engineer + one designer). […]
What to observe
- Reactivation rate over 60 days (MIS-2 test): of the cycle-3 churners to LexFlow, what fraction reactivate? […] Threshold to falsify M4-gestalt: < 20% reactivation by day 60.
What to do if it fails / surprises us
- If the Solo Practitioner Network doesn’t engage after our outreach: do not chase. […]
- If reactivation rate runs below 20% over 60 days: MIS-2 falsifies in favor of pricing-dominates. Cycle 5 strategy must revisit our pricing posture […]
Every recommendation ships with the observations that would prove it
wrong, a numeric threshold decided before the data arrives, and
pre-committed branches for failure. /boyd-outcome later checks
reality against exactly these expectations — that's how the loop closes.
4 · The critique — a critic that pushes back before you see the plan
Every cycle ends with the Boyd-critic subagent stress-testing the proposal. Its verdict for cycle 3 was not a rubber stamp:
systems/orchid-saas/strategy.md — critique section, excerpt
Verdict: SHIP WITH ADJUSTMENTS
Summary: The Schwerpunkt is well-formed and the moral/mental/physical layering is the strongest part. The weakest area is robustness: the strategy assumes MIS-2 resolves in our favor but doesn’t specify the threshold at which we revise. […]
- Robustness — what specific reactivation-rate threshold below MIS-2 do we treat as falsifying M4-gestalt, and what is the contingency we commit to in advance? […]
- ABA talk leverage — […] what does an attendee do the next day that they wouldn’t have done without seeing the talk? If the answer is unclear, the talk is moral signaling without operational pickup.
The critic returns probe questions, never rewrites — and its issue #1 is why the falsification threshold exists in the next-action above. The adjustment loop happened before the artifacts reached the user.
Where this quality comes from
Not from one clever prompt — from accumulated state. By cycle 3 the system holds 18 timestamped observations, six mental models with explicit confidence (one already falsified and retired), a closed mismatch and an open one under test, and a Red-side model of LexFlow whose unknowns are marked as unknowns. Each cycle is also required to run Destruction & Creation on at least one mental model — cycle 3 pulled apart the belief “pricing transparency is our moral advantage” and rebuilt it into something more testable.
The repo also ships two minimal one-cycle systems: a job search whose adversary is a process (the commodity hiring funnel), and a contract negotiation against a client's procurement org. If "adversary" made you think this tool is only for companies at war — it isn't. See the FAQ.