Nothing behind the checkout
Read the rule, the files, and the real scorecards. Then decide.
Backbrief Business OS is plain Markdown files, so there is nothing about it that has to stay hidden until you pay. This page publishes the three things a careful buyer would ask to see first: the grading rule exactly as it ships, the complete file tree of the download, and the two real grading runs that ship in the bundle. Neither run ends in an A, and that is the point.
Everything on this page is generated from the shipped Backbrief Business OS 3.4.1 payload. Page updated August 21, 2026.
The rule that does the grading. Verbatim.
Every advisor on the panel obeys this file. It is the whole contract of the grade: the rubric, the citation discipline, the three-pass cap, and what happens when a plan plateaus. It is published here in full so that when a scorecard tells you no, you can read exactly why it was allowed to.
# Grading (Backbrief Business OS) The grade is a structured stress-test, not a certification. An A- means "no confident weaknesses left un-named against this rubric." It never means the business will succeed, and no agent may present it that way. ## Rubric — six dimensions, scored 1-10, equal weights 1. **Clarity of offer** — who pays, for what, stated so a stranger understands it in one sentence. 2. **Market realism** — evidence demand exists (named buyers, comparable products, real signals — not "everyone needs this"). 3. **Financial viability** — the unit economics close using the owner's own numbers from the brief. 4. **Execution feasibility** — for THIS operator: the time, money, and skills stated in the brief, not a hypothetical team. 5. **Risk coverage** — the top risks are named, each with a mitigation or an explicit owner acceptance. 6. **Differentiation** — a concrete reason to pick this over the obvious alternative (including doing nothing). ## Scoring discipline - Every score cites the plan's own text as justification. No citation, no score. - Tag every external figure that moves a score: **verified** (checked against a named source), **unverified** (plausible, unchecked), or **vendor claim** (the source sells something). Quoting the plan is not verification. A plan quoting its own market-share or membership number is still an unverified figure, and the justification cell has to say so. Apply this consistently across dimensions, not just where it happens to be noticed. - Distinguish a plan claim from a plan claim corroborated by an outside reference. Both satisfy the citation rule; only the second is strong evidence. A score resting on the plan's own self-description is weaker than it looks, and the scorecard should read that way. - Jurisdiction-specific legal, tax, and licensing claims must have gone through the researcher before they can count as risk coverage (agent-routing rule). An unretrieved statute is an assumption, not a mitigation. - Token discipline: advisors receive the plan and rubric only (plus, on regrade passes, their own prior scores), run on the mid-tier model, and keep each dimension's justification to about 60 words. The grade's integrity comes from citations and distinct lenses, not from length. - The five advisors grade through their distinct lenses; unanimous scores without distinct justifications are a defect, not a signal. - Letter mapping: 9.0+ A, 8.5-8.9 A-, 8.0-8.4 B+, 7.0-7.9 B, below 7.0 C. Pass = 8.5 average. ## The loop - Maximum 3 grade->revise loops per plan. A loop is one full grading pass; the third grading pass is the cap. Each revision may address only the two lowest-scoring dimensions, then the full rubric is regraded. - Early exit: if the owner declares a lowest-scoring dimension unfixable (for example, declines the research it needs), stop immediately and present the owner's options — do not spend remaining loops revising around it. - Below A- after the cap: STOP. Output the final scorecard, the top 3 unresolved weaknesses, and the owner's options: proceed at its real grade, pivot, or kill. Never auto-proceed. Never loop past the cap. - A plan that plateaus is a result, not a failure of the system. Report it plainly.
Every file in the download.
This is the complete layout of the folder you unzip. Agents, commands, and rules are each one readable Markdown file. The skills are bundled with their licenses. The scaffold is what gets copied into your own project. If a folder is not listed here, it is not in the product.
business-os-pack.zip # the download, unzipped
├── README.md # install + orientation (Claude Code track)
├── COWORK-QUICKSTART.md # the same system, no terminal (Cowork track)
├── USER-GUIDE.md # daily use
├── BUSINESS-OS-GUIDE.md # the CEO loop, end to end
├── CUSTOMIZATION.md
├── RECOMMENDED-TOOLS.md # disclosed affiliate links, free alternatives named
├── LICENSE-BUSINESS-OS.md # includes the refund policy
├── CHANGELOG.md
├── CLAUDE.md # the one-instruction installer
├── VERSION # 3.4.1
├── example/ # a complete worked build-team run
│ ├── README.md ├── dogfood-report.md
│ ├── transcript.md
│ └── run/ # the run's real project: .claude/, context/,
│ # outputs/ (the built FAQ), workflows/ (the plan)
└── kit/
├── INTEGRATION.md # merge path for an existing .claude setup
├── reference-manifest.json # hashes: which shipped docs are current, which are yours
├── enforcement/ # the optional permission-rule layer (opt-in at install)
│ ├── README.md # plain-language explanation
│ └── settings-enforcement.json
├── .claude/
│ ├── CLAUDE.md
│ ├── VERSION
│ ├── agents/
│ │ ├── orchestrator.md ├── researcher.md ├── planner.md
│ │ ├── builder.md ├── reviewer.md ├── runner.md
│ │ ├── verifier.md ├── advisor-contrarian.md
│ │ ├── advisor-executor.md ├── advisor-expansionist.md
│ │ ├── advisor-first-principles.md ├── advisor-outsider.md
│ │ ├── cmo.md ├── cfo.md ├── web.md
│ │ └── ops.md
│ ├── commands/
│ │ ├── intake.md ├── business-plan.md ├── grade.md
│ │ ├── approve.md ├── grade-idea.md ├── critique.md
│ │ ├── council.md ├── setup.md ├── next.md
│ │ ├── verify-install.md ├── handoff.md ├── pickup.md
│ │ └── demo.md # two-minute replay of a recorded run
│ ├── rules/
│ │ ├── grading.md # published in full above
│ │ ├── ceo-gate.md ├── escalation.md ├── constraints.md
│ │ ├── agent-routing.md ├── verify-before-delivery.md
│ │ ├── context-first.md ├── decision-log.md ├── handoff-checklist.md
│ │ ├── session-hygiene.md ├── token-discipline.md
│ │ ├── skill-routing.md └── skill-routing-business.md
│ └── skills/
│ ├── THIRD-PARTY-LICENSES.md
│ ├── agentic-automation-designer/ # SKILL.md + reference.md
│ ├── automation-workflows/ ├── client-proposal-generator/
│ ├── cold-outreach/ ├── competitive-analysis/
│ ├── content-repurposer/ ├── customer-review-aggregator/
│ ├── deal-closer-playbook/ ├── email-campaigns/
│ ├── expansion-revenue-finder/ ├── financial-modeling/
│ ├── inbound-lead-qualifier/ ├── landing-pages/
│ ├── objection-handling/ ├── offer-creation/
│ ├── personalization-at-scale/ ├── pricing-strategy/
│ ├── sales-call-prep-assistant/ ├── social-media/
│ ├── sop-builder/ ├── stop-slop/
│ └── unit-economics/
│ # every third-party skill folder carries its own MIT LICENSE
└── scaffold/ # copied into the root of your project
├── CLAUDE.md # project template
├── .claude/memory/ # starter decision log + preferences
├── context/strategy/ # current-state.md, current-priorities.md
├── context/reference/ # your knowledgebase + SOURCES.md index
├── examples/worked-example-business.md # both real runs below
├── templates/business-brief-template.md
├── workflows/ # plan, business-plan, scorecard templates
├── inputs/
└── outputs/Real run one · the maker’s own business
NestPet: a real C, and a recorded pivot.
The author’s own pet memorial brand, run through the full pipeline from his original brain dump. Two independent panels, a month apart, graded the same material C: 6.4 on the original run, 5.9 on the August 2026 re-record. Both put the same two dimensions on the floor, and that agreement, not the decimal, is the signal. The owner’s recorded decision was to pivot, and the ops team stayed locked. Below is the re-recorded scorecard, dimension by dimension.
| Dimension | Score | From the justification, citing the plan’s own text |
|---|---|---|
| Clarity of offer | 8.4 | "A pet owner or a sympathy-gift buyer pays $59 for a 6x8 inch matte ceramic tile bearing a painterly, true-to-coat portrait of a specific pet" — all five advisors found the offer stranger-clear in one sentence. Docked for the unresolved dual-buyer split and open plain-sense questions the outsider raised. |
| Market realism | 4.2 | The plan’s own words: "Revenue to date is $0" and "no search-volume data... no data on Etsy category competition." The sole price comparable is tagged unverified by the plan itself. DIVERGENCE reported: contrarian 3 ("an absence of evidence dressed as a 90-day experiment") vs. expansionist and executor at 5. |
| Financial viability | 3.6 | "The gap between 78.3% and 55%+ is real and unexplained" and "No AI-generation or compute cost per portrait appears anywhere" — by the plan’s own admission the margin floor is an underived target. DIVERGENCE: contrarian 2 ("a margin floor with no reconciled cost stack is a wish") vs. expansionist 5. The chairman’s note discounts one advisor’s mis-tagged citation, on the record. |
| Execution feasibility | 6.2 | Units 1-4 are dated, single-owner, and pass/fail-decidable, but the calendar runs on "solo operator with a day job... no hours-per-week figure exists to check that mitigation against." DIVERGENCE: contrarian 4 vs. executor 8 (this week’s units genuinely start Monday). |
| Risk coverage | 5.8 | Honest coverage where evidence exists (shipping damage economics, Etsy AI-policy compliance), but two structural risks are named-not-mitigated: "Business registration, sales-tax collection, licensing... Not addressed in the brief; no jurisdiction is named." |
| Differentiation | 7.0 | "A true-to-coat likeness of the specific pet rather than a template with the pet’s name added," operationalized in the listing’s own acceptance criteria. Held under 8 because the differentiation is corroborated only by the owner’s own QC pass, n=1. |
Average: 5.9· Letter: C· Pass (≥ 8.5): no
“This grade means: no confident weaknesses were left un-named against the rubric above. It is not a prediction or guarantee that the business will succeed.” The full scorecard, with complete justifications, the chairman’s note policing its own panel, and the owner’s recorded decision, ships in the bundle unedited.
Real run two · three loops to the cap
TrailNotes: it improved for three passes and still never passed.
Disclosure, stated plainly: TrailNotes, a $6/month local trail conditions email, is the product’s invented specimen business. Its figures are the example’s premise, but the running of the pipeline and every grading panel was real, and the panel was never told to pass it. A single C run invites a fair objection: maybe this tool just says no to everything. This run answers it. Watch what the panel does when defects actually close, and what it still refuses to do.
| Dimension | Loop 1 | Loop 2 | Loop 3 |
|---|---|---|---|
| Clarity of offer | 8.4 | 8.4 | 8.4 |
| Market realism | 6.2 | 7.4 | 7.4 |
| Financial viability | 5.8 | 7.8 | 7.8 |
| Execution feasibility | 6.8 | 6.6 | 7.6 |
| Risk coverage | 7.4 | 7.2 | 7.4 |
| Differentiation | 6.6 | 6.6 | 7.6 |
| Average | 6.9 (C) | 7.3 (B) | 7.7 (B) · stop |
Scores moved when, and only when, a revision closed the cited defect. Clarity was never revised and never moved. Financial jumped two full points in the loop where the plan adopted the worse number and computed break-even, then sat still.
The panel never drifted toward yes. The contrarian conceded exactly what the revisions earned and nothing more, and was still writing objections at the cap. The first defect was found by the product itself: its cfo agent recomputed the specimen’s own cost stack, found a 15% gap, and the panel docked its own example’s grade for it.
“Below A- at the cap of three grading passes: the loop STOPS here, per the grading rule. No further revision cycles. The owner’s options, presented without a recommendation to auto-proceed: proceed at this grade (B, 7.7), recorded as the owner’s explicit ‘proceed at this grade’ decision; pivot the plan or the business; kill it.”
“A plan that plateaus is a result, not a failure of the system.”
A real business graded C. A designed specimen plateaued at B. That is the product working.
Between them, the two runs answer the two questions every buyer should ask a grading product. What does it do when the answer is bad? It says C, cites the plan’s own sentences, and stops everything downstream until the owner decides. Can the grade be gamed by iterating? Scores move only where defects close, and the loop is capped at three passes by rule.
$79, one-time. All sales are final; the free kit is the try-before-you-buy.