Evolving the system

Change the numbers. Keep the controls

The repo ships tuned for a mid-market B2B SaaS shape: pipeline-denominated economics, named segments in the thousands of contacts, quarterly budgets with a reserve. Your business is probably not exactly that. Almost everything is adjustable, through four specific levers. A short list of controls is not — they are what makes the output trustworthy.

01

What never changes

Remove any of these and the memos stop being evidence. What is left is ordinary AI-generated text in a folder structure — confident, unaccountable, and useless at audit time.

02

The four levers, in escalating order

Work top-down. Most businesses are fully served by levers 1 and 2; reach for 3 and 4 only when the defaults describe a world you don't operate in. After any change: re-run personalize the marketing office (it regenerates every synced copy), then run the smoke test.

Lever 1 — vocabulary, in config/

The KPI unit, segment names, channel list, funnel stages, benchmarks. This is not "adapting" so much as filling in the truth: if your KPI is booked demos, the whole system denominates in booked demos with zero edits — the hurdle, the stops, the tranche gates all read "cost per primary-KPI unit". Most perceived misfits dissolve here.

Lever 2 — numbers, via the rails-overrides table

The "Rails overrides" table in config/budget-and-channels.md rewrites any default in budget-rules.md at personalize time, retagged with the date it was set. Caps, the hurdle ratio, the reserve floor, tranche shape, WAIT validity — all fair game. Overridden values stay visibly distinct from defaults, so an auditor can see what you tuned and when.

Lever 3 — behavior, in agents/ and the protocol files

Charters are markdown: rewrite a Method step, tighten a Forbidden behavior, re-aim a chair's scope, update the worked micro-example to your world. Edit the numbered files in agents/ — never the synced copies under skills/ or .claude/. Keep the eight-section charter format; the orchestrator depends on it. If a charter and a protocol file disagree, the protocol wins — so structural rules change in campaign-protocol.md itself, deliberately.

Lever 4 — the roster

Adding, merging, or retiring a chair touches six places; do all six or the system desyncs: ① the charter in agents/ (keep the format) · ② a skill wrapper in skills/<name>/SKILL.md · ③ a row in the isolation matrix in campaign-protocol.md, stating exactly what it may and may never see · ④ its slot in the run sequence (same file) · ⑤ a slice row in prompts/personalize.md's table so personalization knows which config feeds its context block · ⑥ re-run personalize to generate the .claude/ copies. Prefer refitting an existing chair over adding a fifteenth: every new memo dilutes the reader — you — and the mandatory nine are already a lot of reading per decision.

03

Worked example — a managed service provider

An MSP selling IT management contracts to local businesses breaks several defaults at once: revenue is recurring MRR, not one-off ACV; addressable segments are hundreds of contacts, not thousands; channels are local and relationship-heavy; vendor co-op (MDF) money arrives with deadlines. Here is the full retune. All values below are illustrations — set your own.

Where the default model doesn't fit

DimensionShipped default assumesMSP realityLever
Deal economicsOne-off ACV; pipeline = opps × ACVRecurring MRR; a deal's value is contract-months1 — define pipeline as 12-month contract value (opps × MRR × 12) in budget-and-channels.md
Primary KPISQLs at volumeA handful of booked assessments a month1 — KPI = "first assessments booked", defined per your CRM
SegmentsThousands of addressable contacts"Dental practices — metro area": maybe 300 contacts1 + 2 — small named segments; tighter fatigue and reach caps
ChannelsPaid social, syndication, webinarsLocal search, events/lunch-and-learns, referral programs, vendor co-marketing1 — the channel table drives everything; benchmarks per channel with sources
Budget shapeQuarterly budget with a 10% reserveSmaller budget + vendor MDF with use-by dates2 + 3 — MDF as treasury cash calls in reverse (funds expiring, not owed)
Climate signalsCategory momentum, competitor ad spendLocal: businesses opening, a competitor MSP acquired, insurance/compliance deadlines driving IT audits3 — re-aim the climate chair's signal inventory

The rails-overrides table (illustration)

Rail (as named in budget-rules.md)DefaultMSP override — why
Max share of quarterly budget committed to any one campaign15%25% — a small budget can't be split fifteen ways and still buy anything
Max concurrent campaigns targeting the same named segment32 — a 300-contact segment fatigues fast
Reach cap per flight (share of segment's addressable contacts)50%40% — leave room for the referral motion to work the same list
Frequency cap (touches per contact per week, all campaigns)32 — a 300-contact market remembers overexposure
Economics hurdle: projected pipeline-to-spend5:15:1 kept — but computed on 12-month contract value, per the definition above
WAIT-WITH-TRIGGER validity90 days60 days — local windows (an office move, a compliance deadline) close faster

Everything else — the 40% channel cap, the reserve floor, the tranche shape, the kill defaults — survives contact with the MSP world unchanged. That's the pattern generally: override less than you expect.

Charter refits (lever 3)

What an MSP run looks like after the retune

Same controls, different vocabulary: the hurdle fails on MRR-denominated pipeline, the fatigue cap counts a 300-contact segment, and the treasury chair is watching an expiring co-op balance instead of a conference payment. Fictional values throughout.

04

Other shapes, in brief

plg saas

Product-led SaaS

KPI = activated signups or PQLs; economics denominate in payback months rather than pipeline — redefine the hurdle accordingly (lever 2) and let the economics chair chain spend → signups → activation → paid. Segment fatigue matters less; channel concentration and creative-testing tranches matter more.

agency / services

Professional services

Few, large deals; the analyst's evidence is relationships and references, not funnel rates. Lower the volume math's weight by pasting deal-level data; tighten the brand-risk chair — in services, one damaged relationship costs more than a failed campaign — and expect "not launching is a position" to win more often. It should.

hardware + channel

Hardware with partners

Two audiences: end customers and the channel. Add partner-facing segments in config, count partner campaigns against the same fatigue caps, and extend the compliance profile with partner-program brand rules. If partner marketing is half your spend, that's the one case a fifteenth chair (partner-channel officer) may earn its seat — via all six touchpoints of lever 4.

05

Change discipline