Skip to content

index.md

Latest commit

 

History

History
425 lines (327 loc) · 43 KB

index.md

File metadata and controls

425 lines (327 loc) · 43 KB
title BMad Method Sample Data
purpose Curated, hand-shaped sample inputs and artifacts for testing and demonstrating BMad Method skills.
total_entries 110
last_updated 2026-06-09

BMad Method Sample Data

A deliberate cross-section of real-shaped inputs and artifacts spanning the full BMad funnel. Each entry is hand-shaped to test something specific. The repo exists to:

  1. Test BMad skills against realistic inputs at every density and shape (one-liner intents through 900-line approval-ceremony PRDs).
  2. Train downstream skills (architecture, UX, ticketing) on the range of upstream artifacts they will actually receive.
  3. Validate the spec-kernel thesis — does a five-field kernel carry enough signal for downstream, or is the full PRD wrapper irreducible?
  4. Demonstrate the BMad Method publicly via case studies built end-to-end from selected entries.

How the repo is organized

Folders are ordered roughly upstream to downstream along the BMad funnel.

Folder What lives here Count
intents/ Minimum-density inputs: one-liner intents, tweets, SMS-style threads, bullet-list-only requirements, README stubs, pre-written spec kernels 7
braindumps/ Founder-energy raw inputs: rants, urgency memos, stealth blog posts, shower-thought streams, multi-day chat dumps, customer-coffee post-mortems 9
transcripts/ Verbal-source captures: voice memos, Loom-style narrations, scribed meeting notes, podcast monologues, Zoom interviews, walking-meeting two-voice transcripts 6
threads/ Multi-voice conversation captures: Slack threads, email threads, X/Twitter threads, GitHub issue discussions 6
research/ Research outputs feeding the PRD altitude: customer-interview synthesis, market analysis 1
bmad-spec-kernel/ Pre-written five-field Spec kernels — both clean exemplars and deliberately-violating studies 8
brds/ Business Requirements Documents — corporate-style "the system shall" docs, light on product framing 6
briefs/ Product briefs (one altitude above PRDs); standalone briefs and brief→PRD pairs 6
prds/ Formal product specifications at the PRD altitude: PRDs, PRFAQs, OSS RFCs, JTBD job-stories specs 53
brainstorming-samples/ Completed bmad-brainstorming sessions — intent + full HTML output — across consumer products, services, and BMad ecosystem topics 8
architectures-solution-designs/ Downstream architecture and solution-design artifacts (placeholder)
projects/ End-to-end project examples (single intent walked through the full BMad chain) (placeholder)
other/ Catch-all (placeholder)

Density spectrum across the corpus: 3 lines (one-liner intent, three-line pitch) to 890 lines (HiveLink maximalist PRD). Median ~130 lines.

Quick stats

Metric Value
Total entries (active) 110
Languages represented English (primary), Spanish (1 PRD)
Entries with planted Spec Law violations, contradictions, or hallucinations 10
Brief→PRD altitude-fidelity pairs 2 (PomodoroLite, HiveLink)
Demo-worthy build-and-publish candidates 12 (flagged in catalog)
AI-native products 6
AI as primary feature 7

How to use this corpus for skill testing

Skill Recommended inputs
bmad-spec All of intents/, braindumps/, transcripts/, threads/, research/, briefs/, brds/, prds/. The universal-distiller thesis depends on extraction working across all of these.
bmad-product-brief briefs/ for shape reference; braindumps/ + intents/ + brds/ as inputs to test brief generation
bmad-create-prd briefs/ and brds/ as input; prds/ as output reference shapes
bmad-validate-prd prds/prd-junior-pm-bad-* (10 planted anti-patterns), intents/kernel-rss-revival-* (planted Spec Law violations), briefs/brief-quiet-social-* (6 contradictions), briefs/brief-llm-fitness-platform-* (planted hallucinations), bmad-spec-kernel/kernel-violations-* (4 distinct Spec Law violation studies), brds/brd-hr-onboarding-revamp-* (internal contradiction), brds/brd-compliance-audit-trail-* (impossible requirement smuggled mid-list)
bmad-prfaq prds/prfaq-snapboard-* (consumer), prds/prfaq-internal-runbook-builder-* (internal), prds/prfaq-meetingdoctor-* (internal dogfood), prds/prfaq-rendezvous-* (consumer mobile), prds/prfaq-shipnow-launchpad-* (developer platform) for shape reference
bmad-create-ux prds/* — measure whether UX output degrades when fed only the bmad-spec kernel vs. the full PRD
bmad-create-architecture Same falsification experiment as above on architecture output
bmad-create-epics-and-stories Same falsification experiment for ticketing
bmad-quick-dev Any of intents/, braindumps/, briefs/ — what does Quick Dev do with thin vs. thick inputs?
bmad-investigate prds/prd-aurora-migration-*, prds/prd-engineering-heavy-cache-* — brownfield system investigation rehearsal
bmad-distillator Any maximalist entry (prds/prd-hivelink-*, prds/prd-medivault-*), or any BRD — lossless compression test
bmad-edit-prd prds/prd-pomodorolite-* (lean, simple edits); prds/prd-junior-pm-bad-* (substantive cleanup target)
bmad-shard-doc prds/prd-hivelink-*, prds/prd-medivault-* — maximalist shard targets
bmad-document-project prds/prd-aurora-migration-* style brownfield documentation rehearsal
bmad-kernel-validate (if/when it exists) All of bmad-spec-kernel/ — 4 clean exemplars to verify pass-through, 4 deliberate-violation studies to verify detection
bmad-brainstorming All of brainstorming-samples/ — compare the HTML output against the intent file to verify the session explored the right space; test whether bmad-spec can distill a brainstorm output into a kernel
bmad-bmb (BMad Builder) router See the "BMad-bmb reroute candidates" section below — entries whose described product is actually a BMad agent, skill, or workflow in disguise. Tests whether a router skill correctly identifies "this should be built inside BMad, not as a SaaS."

Catalog — full corpus

intents/ — minimum-density inputs (7)

Entry Lines Notes
intent-podcast-sync 3 Single-sentence intent. Floor case — should bmad-spec block or distill?
intent-tweet-i-want-to-build 3 Single tweet under 280 chars; "anyone want to build this with me" energy
intent-three-line-pitch 3 Three-line elevator pitch (who/give/so-they-can)
intent-text-to-cofounder 13 SMS-style late-night text thread; product seed emerges via abbreviated dialogue
bullets-inventory-scanner 19 15 bullets, no prose. Tests "is structure alone enough?"
kernel-rss-revival 58 Pre-written five-field kernel with planted Spec Law violations (CAP-3 missing success, CAP-4 prescribes HOW, decorative constraint, partial Rule 5 violation)
intent-readme-stub 68 GitHub-README skeleton written BEFORE any code; pre-alpha self-test

braindumps/ — founder-energy raw inputs (9)

Entry Lines Notes
braindump-shower-thought-monetization 29 One-thought-explodes shape; indie maker monetizing a free public API
braindump-conference-hallway-idea 74 Excited, scattered phone notes after a SaaStr onboarding talk
braindump-post-call-memo 83 Hot-take urgency memo after a customer call (renewal at stake)
braindump-shelf-scanner 97 Shower-thought stream proposing a phone-camera bookshelf inventory
braindump-after-customer-coffee 99 Post-coffee braindump; load-bearing customer quotes + counter-hypothesis
braindump-stealth-blog 111 Stealth blog post (essay form) proposing a forward-looking finance product
braindump-standup-killer 125 Founder rant proposing an AI-async-standup tool. Demo candidate.
braindump-2am-founder-rant 192 Solo founder, 2am, after losing a customer; ALL CAPS bursts, contradictions, no clear ask
braindump-discord-paste 257 5-day Discord thread converging on a snapshot/rollback feature

transcripts/ — verbal-source captures (6)

Entry Lines Notes
transcript-podcast-monologue 39 12-min solo-podcast monologue on therapist billing pain; filler words, stutters, timestamps
notes-meeting-mobile 117 Scribed meeting notes from a mobile app v2 kickoff
voicememo-dog-walk-route 131 Rambling voice memo with backtracks — dog-walk-route app idea
braindump-loom-transcript 137 Loom-style narration walking through a 1:1 manager tool sketch
transcript-zoom-customer-interview 252 Otter-style Zoom interview with creative-agency Ops Lead on project handoffs; load-bearing quote + magic-wand reveal
transcript-walking-meeting 297 Two-voice walking transcript with [wind] / [inaudible] artifacts; Product vs Eng arguing whether to kill, sunset, or invest in Pulse Reports

threads/ — multi-voice conversation captures (6)

Entry Lines Notes
thread-x-launchpad 158 30-tweet X/Twitter thread proposing an indie-hacker launch tool
thread-pricing-page-rebuild 190 Multi-voice Slack thread, decisions implicit, contradictions present
thread-email-cancellation-feature 195 7-email reply chain with accumulating > quotes; analytics platform losing a $96k account
thread-github-issue-discussion 231 19-comment OSS issue thread on a data-validation library union-type ask; ends needs-design
thread-email-roi-tool 258 3-week email thread, 5 execs, ROI-calculator decision arc
thread-slack-pricing-debate 262 5-speaker Slack debate: seat-based vs consumption pricing for B2B observability; unresolved ending

research/ — research outputs (1)

Entry Lines Notes
synthesis-customer-interviews 212 UX research synthesis from 14 small-business onboarding interviews

bmad-spec-kernel/ — pre-written five-field kernels (8)

Entry Lines Notes
kernel-warehouse-pick-paths 31 Clean exemplar — pick-list re-ordering for a fulfillment-center operator
kernel-insurance-claims-triage 31 Clean exemplar — regulated FNOL triage for a property carrier
kernel-podcast-clip-extractor 32 Clean exemplar — indie-podcaster shareable-clip surfacer
kernel-edu-quiz-builder 33 Clean exemplar — teacher-built formative quizzes from a lesson plan
kernel-violations-non-falsifiable 38 Planted violation: capabilities are non-observable ("delight", "intuitive", "best-in-class")
kernel-violations-solution-disguised 38 Planted violation: capabilities pre-bake UI choices (dashboard, sidebar, wizard, kanban)
kernel-violations-mixed-concerns 42 Planted violation: problem leaks capabilities; constraints contain non-goals; non-goals smuggle aspirational capabilities
kernel-violations-vendor-soup 47 Planted violation: capabilities name vendors/frameworks (Stripe, AWS Lambda, Pinecone, Next.js, etc.)

brds/ — Business Requirements Documents (6)

Entry Lines Notes
brd-support-knowledge-base 90 Thin/lazy KB BRD; "FR-005 the system shall be intelligent" style; zero authoring/maintenance plan
brd-sales-quote-generator 113 RevOps quote-gen; smuggled vendor preference (FR-011 names DocuSign)
brd-hr-onboarding-revamp 119 VP of People onboarding BRD; internal contradiction between FR-005 and FR-009 (equipment self-service)
brd-finance-month-end-close 128 Controller close BRD; FR-013 conflates Blackline tool with need; FR-018 is a wish ("continuous close")
brd-compliance-audit-trail 138 CCO regional-bank audit-trail BRD; planted near-impossible requirement (FR-014 retroactive backfill of historical audit events)
brd-procurement-vendor-portal 164 Procurement portal BRD; 20 FRs, throughput-oriented success criteria, "user-friendly" smuggled in

briefs/ — product briefs (6)

Entry Lines Pair? Notes
brief-coffeebean-tracker 31 standalone Thin one-pager brief (personal coffee log)
brief-pomodorolite 61 →pomodorolite Brief→PRD pair. Lean hobby altitude.
brief-llm-fitness-platform 71 standalone LLM-generated brief with planted hallucinations (fake McKinsey citation, fake competitors, fake regulatory framework)
brief-quiet-social 77 standalone Self-contradicting brief; six embedded contradictions
brief-streetside-trees 81 standalone Standalone idea-stage brief (community urban tree mapping)
brief-hivelink 111 →hivelink Brief→PRD pair. Maximalist enterprise altitude.

prds/ — PRD-altitude specifications (53)

Baseline PRDs — pre-corpus (13)

Entry Lines Domain Notes
prd-pomodorolite 54 productivity Lean hobby weekend-build PRD. Has paired brief. Demo candidate.
prd-vendorinsight 72 procurement Clean baseline. Control case.
prd-shipctl 168 devtools Platform CLI replacing legacy bash script
prd-echocraft 184 gaming Browser game (asymmetric co-op puzzle). Demo candidate.
prd-cardiacwatch 220 medical research Research protocol PRD (AFib prediction)
prd-palletpulse 330 IoT logistics Industrial IoT with anchor customer
prd-aurora-migration 362 backend infra Rails monolith → microservices, PRD/RFC hybrid
prd-craftcircle 404 marketplace Premium curated maker marketplace. Demo candidate.
prd-plantsona 491 consumer social Per-plant social network with AI voice. Demo candidate.
prd-heart-rate-portal 493 medical (portfolio) Portfolio web app, well-formed reference shape
prd-medivault 723 healthcare regulated HIPAA records-exchange approval ceremony
prd-hivelink 890 enterprise iPaaS Maximalist 60+ FRs. Has paired brief.
prd-narrative-vtt 11 TBD Pre-discovery stub; blocked-path test

Posture variants — greenfield/brownfield/lifecycle (8)

Entry Lines Posture
prd-vendorinsight-csv-export 49 Brownfield narrow feature-add
prd-habit-share-extension 77 Brownfield hobby extension
prd-sunset-legacy-reports 134 Deprecation / sunset (inverse PRD)
prd-oss-fork 147 OSS fork with governance posture
prd-pivot-b2c-to-b2b 160 B2C → B2B pivot
prd-soc2-retrofit 163 Compliance retrofit
prd-scale-retrofit 184 Scale-out fire-drill (12K → 180K MAU)
prd-acquisition-integration 223 M&A integration (Day 1 → Year 1)

App-type variants — distribution and runtime (12)

Entry Lines App type
prd-watchos-reverse-alarm 92 WatchOS app. Demo candidate.
prd-telegram-bot-jorprice 100 Telegram bot
prd-email-newsletter-fielddispatch 101 Email-as-product
prd-browser-ext-tabtidy 104 MV3 Chrome extension
prd-vscode-ext-commitnudge 110 VS Code extension
prd-print-kit-yearmark 113 Print-on-demand physical kit
prd-discord-bot-bookcorner 114 Discord bot (slash commands)
prd-mac-streamdeck-prompts 130 Native Mac menubar app. Demo candidate.
prd-rag-obsidian-chat 150 RAG product (Obsidian plugin). Demo candidate.
prd-ai-agent-inbox 179 Pure AI agent product. Demo candidate.
prd-oss-library-rateguard 203 TypeScript library / SDK RFC
prd-firmware-leakmate 209 Bare-metal MCU firmware

Domain variants — real-world breadth (9)

Entry Lines Domain
prd-faith-quietroom 129 Faith community
prd-creator-daw-plugin-stemloop 130 Music creator (DAW plugin)
prd-birding-warble 131 Niche hobbyist (birding). Demo candidate.
prd-pet-tech-greymuzzle 140 Pet tech (senior dogs). Demo candidate.
prd-mental-health-anchor 149 Mental health (anxiety journaling)
prd-civic-curbcut 156 Civic tech / accessibility. Demo candidate.
prd-edtech-mathmentor 160 EdTech (regulated; AI tutor)
prd-climate-carbon-tracker 170 Climate / ESG (B2B SaaS)
prd-fintech-rentsplit 181 Fintech (consumer payments)

Edge / negative cases — deliberately problematic (4)

Entry Lines Edge type
prd-vibes-fashion-app 131 Vibes-only PRD (brand-heavy, behavior-light)
prd-spanish-mercado 133 Full Spanish-language PRD
prd-junior-pm-bad 211 Junior-PM PRD with 10 planted anti-patterns
prd-engineering-heavy-cache 312 Engineering-only PRD (tech-heavy, user-light)

Non-PRD shapes at PRD altitude (7)

Entry Lines Shape
prfaq-meetingdoctor 106 Internal-tool PRFAQ; manager meeting-triage; HR-surveillance guardrails
prfaq-internal-runbook-builder 108 Internal-tool PRFAQ
prfaq-rendezvous 130 Consumer mobile PRFAQ; meet-in-the-middle weekend trip planner
prfaq-snapboard 131 Amazon-style consumer PRFAQ
jobstories-pantry 149 Strict JTBD job-stories format
prfaq-shipnow-launchpad 170 Developer-platform PRFAQ; "repo to revenue in 30 minutes"; cannibalization/ICP questions
rfc-oss-bigissue 287 OSS RFC + emulated comment thread

brainstorming-samples/ — completed bmad-brainstorming sessions (8)

Each entry contains a brainstorm.html (full session output) and, where present, a brainstorm-intent.md (structured session summary with top directions and key insights). Two entries have only a .memlog.md inside the folder (no separate intent file).

Entry HTML lines Topic Notes
brainstorm-delightful-laundry 305 Making laundry a delight instead of a chore 66 ideas; 4 techniques; intent file present. Strong consumer-product direction (container system, heated folding bed). Demo candidate.
brainstorm-digital-campfire-software 340 Software that makes the internet feel cozy and human again Intent file present. Antidote to algorithmic feeds / engagement-farming.
brainstorm-er-waiting-room-anxiety 270 Reducing ER waiting-room anxiety without digital/software interventions 62 ideas; intent file present. Physical space, lighting, human-presence role. Non-digital constraint makes it a good scope-limitation test.
brainstorm-hospital-discharge-experience 381 Redesigning the hospital discharge experience 62 ideas; 4 techniques; intent file present.
brainstorm-surprise-70th-birthday 483 Planning a 70th birthday for someone who "wants nothing" Intent file present. Non-product, life-planning topic — tests skill on non-software domain.
brainstorm-photospot-caching 923 Geocaching successor using photo-viewpoint alignment Intent file present (from .memlog.md). Most detailed session; BMad YouTube demo candidate (father-daughter series). Demo candidate.
brainstorm-creative-grilling 570 Wild, scroll-stopping grilling ideas for social media No separate intent file; session log in .memlog.md. Social/viral content topic — tests skill on non-product creative domain.
brainstorm-new-vertical-modules 686 New BMad Builder modules for non-software verticals No separate intent file; session log in .memlog.md. 49 module ideas across 3 techniques. BMad-reroute candidate — the described product is BMad skills/modules.

Cross-cutting indexes

Brief → PRD pairs (altitude-fidelity test set)

Brief PRD What the pair tests
brief-pomodorolite (61L) prd-pomodorolite (54L) Are the kernels equivalent at hobby altitude?
brief-hivelink (111L) prd-hivelink (890L) Are the kernels equivalent at maximalist enterprise altitude?

If kernels are equivalent across pairs, the brief carried the load-bearing signal and the PRD wrapper is ceremony for downstream purposes. If kernels diverge, the PRD altitude adds irreducible signal.

By AI involvement

AI-native products (the product is AI-powered)

AI as primary feature

AI is the planted test signal (LLM-generated artifacts with detectable issues)

Demo-worthy build-and-publish candidates

These are flagged as strongest BMad-Method-case-study candidates: charming enough to actually build, narrow enough to ship, audience-resonant.

Spec Law test focus (for bmad-spec validation)

Entries deliberately stressing specific Spec Law rules:

BMad-bmb reroute candidates

Some entries describe a "product" that, on inspection, is really a request for a BMad agent, skill, or workflow. A well-built bmad-spec or routing skill should flag these for the BMad Builder (bmad-bmb) instead of marching them down the PRD pipeline. These entries are the routing-decision test set.

Strong reroute (the described product is a BMad agent/skill/workflow):

Entry Why it reroutes
braindumps/braindump-standup-killer An AI agent ingesting dev signals (commits, PRs, calendar) and emitting an async standup. BMad agent + workflow, not a SaaS.
prds/prfaq-meetingdoctor Calendar/Zoom-metadata ingest → meeting recommendations. BMad workflow: ingest skill + analysis skill + report skill.
prds/prfaq-internal-runbook-builder Generate runbooks from incident artifacts. Textbook BMad skill, or small workflow chaining ingest → outline → fill.
braindumps/braindump-after-customer-coffee Founder explicitly wants help structuring post-coffee customer-signal interpretation. BMad skill ("post-customer-conversation analysis").
braindumps/braindump-post-call-memo "Help me distill this hot customer call before I lose the thread." BMad skill.
transcripts/transcript-zoom-customer-interview Magic-wand answer is "something that listens to conversations and generates the handoff as a byproduct" — transcript-ingest skill + structured-summary skill, not a SaaS.

Ambiguous overlap (could be a BMad skill or a real product depending on goal):

Entry The tension
prds/prd-ai-agent-inbox Claude-native inbox triager. BMad agent if the user owns the runtime; SaaS if they want to sell it to non-BMad users.
prds/prd-rag-obsidian-chat Obsidian RAG chat is genuinely a BMad-installable skill for many users. Routing depends on whether the goal is a community Obsidian plugin or a personal BMad knowledge agent.
prds/prd-mac-streamdeck-prompts Stream Deck → AI prompts is one BMad-skill thin client away from existing.
prds/prd-vscode-ext-commitnudge Commit-message nudging is a BMad developer skill in disguise; the IDE-extension wrapper is the only product-y part.

Stay as-is (genuinely product, SaaS, hardware, or regulated — not BMad work):

All BRDs (corporate ops systems), prd-fintech-rentsplit, prd-edtech-mathmentor, prd-medivault, prd-cardiacwatch, prd-firmware-leakmate, prd-palletpulse, prd-watchos-reverse-alarm, the games/marketplaces (prd-echocraft, prd-craftcircle), and most consumer-mobile PRFAQs (prfaq-rendezvous) describe products that need to ship as products. A router should NOT reroute these.

As a test set: 6 strong reroute + 4 ambiguous + the broader "stay as-is" set is a graded difficulty curve for any BMad routing or triage skill. Misroutes in either direction are calibration signals.

BRD-specific test focus

BRDs are deliberately product-light. They test whether downstream skills can ask the right product questions to recover what the BRD omits.

By density / weight class

  • Tweet (1-10L): intents/intent-podcast-sync (3), intents/intent-tweet-i-want-to-build (3), intents/intent-three-line-pitch (3)
  • Sparse (10-50L): prds/prd-narrative-vtt (11), intents/intent-text-to-cofounder (13), intents/bullets-inventory-scanner (19), braindumps/braindump-shower-thought-monetization (29), briefs/brief-coffeebean-tracker (31), all 4 clean bmad-spec-kernel/kernel-* exemplars (31-33), prds/prd-pomodorolite (54), prds/prd-vendorinsight-csv-export (49)
  • Lean (50-100L): ~14 entries including the violation kernels (38-47L), intent-readme-stub (68), brief-pomodorolite, brief-streetside-trees, brief-quiet-social, brief-llm-fitness-platform, vendorinsight, prd-habit-share-extension, prd-watchos-reverse-alarm, braindumps/braindump-shelf-scanner, braindumps/braindump-after-customer-coffee, braindumps/braindump-conference-hallway-idea, braindumps/braindump-post-call-memo, brd-support-knowledge-base
  • Standard (100-200L): ~45 entries — the meat of the corpus; includes all PRFAQs, most domain PRDs, most BRDs, and most threads
  • Mid-heavy (200-400L): ~16 entries including prd-junior-pm-bad, research/synthesis-customer-interviews, prd-cardiacwatch, prd-acquisition-integration, threads/thread-email-roi-tool, threads/thread-slack-pricing-debate, threads/thread-github-issue-discussion, threads/thread-email-cancellation-feature, braindumps/braindump-discord-paste, prds/rfc-oss-bigissue, prd-engineering-heavy-cache, prd-palletpulse, prd-aurora-migration, prd-craftcircle, transcripts/transcript-zoom-customer-interview, transcripts/transcript-walking-meeting
  • Maximalist (500-1000L): prd-heart-rate-portal (493), prd-plantsona (491), prd-medivault (723), prd-hivelink (890)

Experiments the corpus enables

1 · Universal-distiller validation

For each non-PRD-shaped entry (braindumps/, transcripts/, threads/, intents/, research/, brds/), distill via bmad-spec and verify a workable five-field kernel comes out. If extraction holds across all shapes, the universal-distiller thesis is supported.

2 · Kernel-enough hypothesis

For each brief→PRD pair in briefs/, distill both sides and compare kernels.

  • Equivalent kernels → the brief carried the load-bearing signal; the PRD wrapper is ceremony for downstream purposes.
  • Divergent kernels → the PRD altitude adds irreducible signal.

Extension: hand the distilled kernel (no source) to bmad-create-architecture, bmad-create-ux, bmad-create-epics-and-stories. Measure downstream output quality vs. handing them the full PRD. This is the falsification test for "is the kernel enough downstream?"

3 · Spec Law detection rigor

For each entry with planted violations (intents/kernel-rss-revival, briefs/brief-quiet-social, briefs/brief-llm-fitness-platform, prds/prd-junior-pm-bad, prds/prd-vibes-fashion-app, prds/prd-engineering-heavy-cache, and all 4 bmad-spec-kernel/kernel-violations-*), run bmad-spec and verify it surfaces the planted issues as direct findings, assumptions, or open questions. Misses are calibration signals.

4 · Density floor

For the thinnest entries (the 3-line intents, intents/bullets-inventory-scanner, braindumps/braindump-shower-thought-monetization), verify bmad-spec produces appropriate output — blocked for genuinely-too-thin, partial with extensive open-questions for sparse-but-distillable, complete for the bullet-list.

5 · Language and idiom robustness

prds/prd-spanish-mercado tests extraction without English idioms. The kernel should be functionally equivalent to what a parallel English-rewrite would produce.

6 · Demo flywheel

For each demo-worthy candidate, the entry is the starting input. Run the full BMad chain (spec → architecture → UX → epics → dev) and publish the transcript as a BMad case study. The chain trail becomes the demonstration of the method.

7 · BRD-to-product translation

For each entry in brds/, run bmad-product-brief or bmad-create-prd and measure how well the resulting product artifact:

  • recovers user journeys the BRD omitted
  • challenges vendor-as-requirement smuggling
  • flags contradictions and impossible asks
  • exposes the unstated assumptions (who maintains it, what success looks like for the end user, not just the stakeholder)

This is the "is BMad better than a typical IT-receives-BRD handoff?" test.

Conventions

  • Entries live in date-stamped folders ({slug}-YYYY-MM-DD/) so multiple revisions of the same product can coexist over time.
  • Each folder typically contains a single primary artifact (prd.md, brief.md, brd.md, spec.md, voicememo.md, thread.md, transcript.md, braindump.md, etc.) and optionally a .decision-log.md or spec.md sibling.
  • Product names are deliberately fictional; nothing in this repo represents real shipped products.
  • All artifacts are hand-shaped to test a specific thing. Avoid filler. Avoid generic samples that don't exercise an edge case.

Contributing

When adding a new entry:

  1. Pick the folder that matches the artifact's shape (or open a discussion to add a folder).
  2. Use the {slug}-YYYY-MM-DD/ directory convention with a primary-artifact filename inside.
  3. Add a row to the relevant section of this index, including the test signal the entry carries.
  4. If you plant intentional violations or contradictions for testing, note them in the entry's notes column so they remain rediscoverable.