Legacy Soil & Stone

The Research Journey — How We Got Here

Legacy Soil & Stone — April 2026

This document tells the story of how 29 research reports, a financial model, and a master proposal came together in roughly two weeks. It covers what worked, what didn't, what surprised us, and which pivots changed the direction of the business.

Where It Started

The idea began as a late-night conversation — "make me a tree." The original concept was simple: take cremated remains, mix them with soil, plant a tree. But the more Mark dug into the actual science, the more he realized the problem was both harder and more interesting than expected.

The first formal research push started in early April 2026, using AI-assisted deep research agents to pull together sourced, verifiable reports on every major topic the business would need to address. The approach was deliberate: don't guess, don't assume, research everything. Every claim had to trace back to a source. Every number had to come from somewhere real.

The Research Method

Each research report was generated through a focused agent run — a single session dedicated to one topic, with instructions to find primary sources, cite everything, and flag where the evidence was thin. The agent would produce a long-form markdown report, and Mark would read it, challenge it, and often send it back for deeper investigation.

This approach had three strengths:

Speed. A topic that would take weeks of manual research could be covered in a day. Twenty-nine reports in two weeks would not have been possible without agent-assisted research.

Sourcing discipline. Every report was required to cite its claims. When a report said "the Jora JK400 reaches 140°F in 48 hours," there had to be a source. When a report said "ChipDrop provides free woodchips," there had to be a link. This discipline caught several early assumptions that turned out to be wrong.

Iterative depth. The first pass on a topic would often reveal that the real question was different from the one we asked. The Pearl Method report started as "how to mix cremains into cement" and ended up being about pan granulation chemistry. The vessel design reports started as "how to insulate a barrel" and ended up being about why we shouldn't build barrels at all.

The Major Pivots

Pivot 1: Custom Barrels → Jora JK400

The original plan called for custom-fabricated HDPE barrels on steel cradles with manual insulation. Three reports (Vessel_Analysis, Vessel_Design_Detailed, Vessel_Design_Variables) were produced exploring this path. The research was thorough and technically valid, but it kept surfacing the same problem: the hidden cost of custom fabrication.

When the Jora JK400 entered the picture — a commercial, insulated, dual-chamber tumbling composter at $940/unit — the economics flipped. One off-the-shelf purchase replaced $20,000+ in custom fabrication R&D, solved the insulation problem with factory-integrated 2.25" HDPE, and came with academic validation from MSU Extension.

What we learned: The three vessel design reports are still valuable as engineering reference, but the Jora selection made them background material rather than the operational plan. Research that leads to "don't build this" is still good research.

Pivot 2: Weight-Based Pricing → Product-Tier Pricing

The original Line 2 pricing was structured around animal weight: small, medium, large, extra-large. Research into customer psychology and competitor pricing (Competitive_Landscape, Money_and_Prices) revealed that weight-based pricing feels clinical. Families don't think about their pet's weight when they're grieving.

The pivot was to product tiers — Seedling, Bloom, Grove, Legacy — named after what grows, not what weighs. Each tier corresponds to a planter size and plant selection, and the price reflects the experience, not the input weight.

What we learned: Pricing structure is a brand decision as much as a financial one. The numbers matter, but the framing matters more.

Pivot 3: Shelter Intake Fees → Zero-Revenue Contracts

The original Line 3 model assumed a per-animal intake fee from shelters ($15/animal). Research into shelter budgets (Mass_Shelter_Intake, Financial_Proforma) revealed that shelters genuinely don't have the money. A $15 fee per animal would either kill the partnership or require subsidy.

The pivot: zero-revenue municipal contracts. The shelter pays nothing. Revenue comes entirely from community soil sales at $35/bag. This reframe turned Line 3 from a marginal service into the primary revenue driver — the soil is the product, and the shelter partnership is the supply chain.

What we learned: Sometimes the business model that works is the one where you stop trying to charge the wrong customer.

Pivot 4: Two Service Lines → Four

The original concept had two lines: stones and soil. Research kept surfacing opportunities that didn't fit neatly into either. The shelter program started as a footnote in the soil discussion and grew into its own line with its own economics. The academic research angle started as "wouldn't it be nice" and became a real line when we realized that Line 3's mass operation was exactly the kind of instrumented, controlled environment that universities need and can't easily build.

What we learned: The business architecture emerged from the research, not the other way around. We didn't start with four lines and fill them in — we started with two and the research told us there were four.

What Worked

The "research button" approach. Treating each topic as a standalone, source-verified report meant that every decision could trace back to its evidence. When the financial model needed an equipment price, the answer was in Equipment_Inventory.md. When the proposal needed to explain composting science, the answer was in Process_Blueprint.md. No guessing, no "I think it costs about..."

Brain dump transcription. Mark's April 8 brain dumps (45 minutes of stream-of-consciousness talking, transcribed via OMI) captured the founder's vision in a way that no structured interview could. The Vision document, the brand direction, and several operational decisions trace directly back to those transcripts.

Adversarial review. The Claude_Analysis and Debate_Prompt reports were deliberately adversarial — "tear this apart and tell me what's wrong." Those reports surfaced real weaknesses (single-operator risk, permit timeline uncertainty, pricing assumptions) that went directly into the proposal's "what we don't know" sections.

Decision closures. When a research question had been answered well enough to move forward, we locked it and documented the closure (Decision_Closures_April2026.md). This prevented the research from becoming an infinite loop where every answer spawns three new questions.

What Didn't Work

Early .docx drafts. Several early research reports were produced as Word documents rather than markdown. These were harder to search, harder to cross-reference, and harder to update. All have since been superseded by markdown versions and archived. Lesson: markdown first, always. Generate .docx at the end for distribution.

Prefix soup in file names. The early naming convention used prefixes like "Deep_Research_", "LSS_", "Legacy_Soil_Stone_" that made files hard to scan. The April 12 reorganization stripped all prefixes and moved to clean descriptors. The folder tells you the context; the file name tells you the topic.

Trying to finalize numbers too early. The first financial model pass used placeholder numbers that cascaded into misleading projections. The fix was to flag every unresearched number explicitly ("NEEDS RESEARCH") and only lock numbers when the research backed them up. The Open Research tab in the financial model tracks what's confirmed and what's still estimated.

Overbuilding engineering solutions. The vessel design series (three reports, 100+ pages combined) went deep on thermal modeling, insulation trade-offs, and custom fabrication options. All technically excellent work — but the Jora JK400 selection made most of it academic. The lesson isn't "don't do deep engineering research" — it's "check whether the problem has an off-the-shelf solution before designing a custom one."

The Numbers

What's Next

The research phase is complete. The build phase starts now. The first bench-scale tests (Marble Method pearling run, first Jora JK400 composting cycle) will validate or revise the numbers the research produced. When they do, the research reports get dated addendums — not rewrites. The research journey is documented; the build journey starts here.

"Don't make it up on your own. Research was done for a reason and there's good information there." — Mark Barnett