bikoret/specs/shorashim_economic_model.md

# SHORASHIM: Economic Model

## Document Purpose

This is the machatzit hashekel — the half-shekel. Torah ties funding to census and census to accountability. *He'ashir lo yarbeh v'hadal lo yam'it mimachatzit hashekel* — "The rich shall not increase and the poor shall not decrease from the half-shekel" (Shemot 30:15). Equal contribution. Not proportional. Not progressive. Equal. The half-shekel funds the tamid — the daily communal offerings. Without the half-shekel, the tamid cannot be brought. Without a funding model, the system cannot be sustained.

**Upstream authority:** Torah → PRD → Architecture Spec → Construction Spec → Operations Manual → **This Document** → Sustainability

---

## 1. THE PRINCIPLE: THREE REVENUE TYPES

Torah distinguishes three kinds of economic provision for the mishkan. Each has different rules. Conflating them is like using terumah where machatzit hashekel is required.

### 1.1 Terumah — Voluntary Offering (Startup Capital)

*Kol ish asher yidvenu libo* — "every person whose heart moves them" (Shemot 25:2).

Terumah is voluntary. It funds the initial construction. It is not recurring. The golden rings, the acacia wood, the blue and purple yarn — these are given once to build the mishkan. After the mishkan is built, terumah is not collected again for the same purpose.

**Application:** Startup costs are funded by terumah — voluntary one-time contribution. This includes server provisioning, initial hosting deposit, model weight downloads, and one-time GPU rental for fine-tuning. The AI builds — there is no labor cost.

### 1.2 Machatzit HaShekel — Equal Recurring Contribution (Operations)

*Zeh yitnu kol ha'over al hap'kudim: machatzit hashekel b'shekel hakodesh* — "This they shall give, everyone who passes through the census: a half-shekel by the shekel of the sanctuary" (Shemot 30:13).

Machatzit hashekel is mandatory, equal, recurring. It funds the tamid — the daily communal service. Every person gives the same amount regardless of wealth. This is not charity. It is membership. The half-shekel purchases your part in the communal offering.

**Application:** Monthly operating costs are funded by machatzit hashekel — equal recurring contributions from every participating member of the community. Not per-use fees. Not tiered pricing. Not "free tier + premium." Every member contributes the same fixed monthly amount, and this funds the system's daily operation.

### 1.3 Pe'ah — The Corner Left Unharvested (Free Access)

*Lo t'chaleh p'at sadcha liktzor... le'ani v'lager ta'azov otam* — "You shall not completely harvest the corner of your field... for the poor and the stranger you shall leave them" (Vayikra 19:9-10).

Pe'ah is not charity given by the owner. Pe'ah is a portion of the harvest that the owner never owns. The stranger does not receive pe'ah from the owner's hand. The stranger takes pe'ah directly from the field. No intermediary. No gratitude required. No identification recorded.

**Application:** The pe'ah endpoint is free, unauthenticated, untracked. It is not funded separately — it is funded by the machatzit hashekel. The community's equal contributions fund a system that includes, by design, a corner that anyone can access. The pe'ah is not a cost center to be minimized. It is a design requirement that defines the system's character. A field without pe'ah is not a kosher field. A system without pe'ah is not Shorashim.

---

## 2. COST STRUCTURE

### 2.1 Terumah (One-Time Startup)

| Item | Cost | Notes |
|------|------|-------|
| Server provisioning deposit | $200-400 | First month + setup, depends on provider |
| Domain registration | $15/year | Annual, but initial purchase is one-time |
| Mistral GGUF weights download | $0 | Free (Apache 2.0), ~30GB bandwidth |
| One-time GPU rental (fine-tuning) | $50-100 | 24-48 hours A100 on Lambda/vast.ai |
| L1 Torah text acquisition and processing | $0 | Sefaria: free, open data |
| Root lexicon compilation | $0 | BDB/Strong's: public domain |
| TLS certificate | $0 | Caddy auto-provisions via Let's Encrypt |
| **Total terumah** | **$265-515** | |

### 2.2 Machatzit HaShekel (Monthly Recurring)

| Item | Monthly Cost | Notes |
|------|-------------|-------|
| **Hosting** | | |
| Server rental (16-core, 128GB RAM, 1TB+500GB SSD) | $200-350 | Infomaniak/1984/Bahnhof, dedicated or high-spec VPS |
| Bandwidth | included | Typically included in Swiss/Icelandic hosting |
| Backup storage (off-site, encrypted) | $10-20 | For L3 backups |
| **External APIs** | | |
| Anthropic API (Claude korban source) | $30-80 | ~500 API calls/day × 30 days × ~$0.002-0.005/call |
| Mistral API (Phase 1 only, before self-hosting) | $20-50 | Reduces to $0 after self-hosting in Phase 3 |
| **Domain/DNS** | | |
| Domain renewal (amortized) | $1.25 | $15/year ÷ 12 |
| **Total monthly** | **$261-501** | |
| **Steady state (post-Phase 3, Mistral self-hosted)** | **$241-451** | |

### 2.3 Machatzit HaShekel Calculation

If the total monthly cost is ~$350 (median estimate, steady state):

| Community Size | Monthly Per Person | Annual Per Person |
|----------------|-------------------|-------------------|
| 10 members | $35 | $420 |
| 25 members | $14 | $168 |
| 50 members | $7 | $84 |
| 100 members | $3.50 | $42 |
| 200 members | $1.75 | $21 |

The half-shekel in the Second Temple period was approximately one day's wage for an unskilled laborer. At current US minimum wage ($7.25/hr × 8 hours = $58/day), a biblical half-shekel equivalent is roughly $58/year or ~$5/month. This is feasible at a community of 70 members — one for each of the 70 faces of Torah.

**Target: 70 members × $5/month = $350/month.** This funds the tamid.

### 2.4 Cost Trajectory Over Time

| Year | Annual Cost | Cumulative | Notes |
|------|------------|------------|-------|
| Year 1 | $4,700 | $4,700 | Terumah ($400) + 12 months operations ($4,300 avg with Phase 1 API costs) |
| Year 2 | $3,600 | $8,300 | Steady state, Mistral self-hosted |
| Year 3 | $3,600 | $11,900 | |
| Year 4 | $3,800 | $15,700 | L3 storage may need expansion |
| Year 5 | $3,800 | $19,500 | |
| Year 6 | $3,800 | $23,300 | |
| Year 7 (shmitah) | $4,200 | $27,500 | Shmitah review: potential hardware refresh, major version evaluation |

**Seven-year total: ~$27,500.** This is the cost of running a specification-grounded Torah processor for a full shmitah cycle. For context: a single enterprise software license often exceeds this annually.

---

## 3. PE'AH ECONOMICS

### 3.1 The Design Constraint

The pe'ah endpoint is free and unauthenticated. It cannot be metered, throttled by payment status, or degraded relative to the authenticated endpoints. The stranger who approaches pe'ah receives the same quality of grain as the owner harvested from the rest of the field.

This means: pe'ah consumes system resources (compute, manna provision, API calls for korban sources) without contributing revenue. The question is not whether to offer pe'ah — that is non-negotiable. The question is how much of the system's capacity pe'ah represents and how the machatzit hashekel absorbs it.

### 3.2 Pe'ah Sizing

Torah does not specify a minimum size for pe'ah. The rabbis set the minimum at one-sixtieth of the harvest (Mishnah Pe'ah 1:2). But *ein lah shi'ur l'ma'alah* — there is no upper limit. A generous person may leave more.

**Minimum pe'ah allocation: 1/60 of system capacity.**
At 1,000 daily korbanot: pe'ah handles at least 16-17 requests/day.
At capacity targets of 100 concurrent users: pe'ah supports at least 1-2 concurrent users.

**Recommended pe'ah allocation: 1/10 of system capacity.**
Torah requires multiple forms of giving (terumah, ma'aser, pe'ah, leket, shikhchah). Pe'ah at 1/10 is generous but sustainable. This means ~100 pe'ah requests/day and up to 10 concurrent pe'ah users.

### 3.3 Pe'ah Does Not Degrade

If pe'ah demand exceeds the allocation, the system does not degrade pe'ah quality. Instead:
- Pe'ah requests enter a queue (the stranger waits, but is not turned away)
- Queue depth is monitored
- If pe'ah demand consistently exceeds 1/10 allocation, the community evaluates: is the community too small? Is the machatzit hashekel insufficient? Does the field need to grow?

Pe'ah demand is a signal. If many strangers are coming, the field is good. This is *ki tov*.

### 3.4 What Pe'ah Does Not Include

Pe'ah provides access to the system's basic processing — the minchah offering type. It does not include:
- Sanhedrin deliberation (panel spawning is expensive — this is chatzer access)
- Full korban processing from external models (API costs are real)
- Persistent context across multiple interactions (pe'ah is per-visit, like gleaning)

Pe'ah includes:
- Shoresh Engine processing (root activation, te'amim reading, gematria)
- Talmudic Engine basic interpretation (against the specification)
- A single-pass Bereishit cycle producing evaluated output
- Honest teiku when the system cannot resolve

---

## 4. RESERVE AND CONTINGENCY

### 4.1 The Lechem HaPanim Model

The shulchan held twelve loaves of showbread, replaced weekly. They were arranged the previous Shabbat and consumed the following Shabbat. There was always bread on the table, but it was never stale beyond a week.

**Operating reserve: 3 months of operating costs.**
At $350/month steady state: $1,050 reserve.
This reserve is maintained from the machatzit hashekel — a small additional contribution per person per month, or a one-time terumah collection for the reserve fund.

At 70 members, a $15 one-time terumah per person funds the reserve entirely.

### 4.2 The Shmitah Reserve

Every seventh year, the system undergoes major review. Hardware may need replacement. Major version changes may require re-architecture. The shmitah year may cost more than a regular year.

**Shmitah reserve: $2,000** (accumulated over 6 years at ~$28/month additional, or ~$0.40/member/month at 70 members).

### 4.3 What Happens If Funding Falls Short

If the community shrinks below sustainability:

1. **First response:** Reduce external API usage. Self-host all models. Reduce monthly cost to hosting-only (~$200-350/month). This reduces capability (no Claude korban source, no Mistral API) but preserves the core system.

2. **Second response:** Downsize hosting. Move to a smaller server. Reduce concurrent capacity. The system serves fewer people more slowly but does not compromise on quality or specification fidelity. *Hamam'it lo hechsir* — the one who gathered little had no lack.

3. **Third response:** If the community cannot sustain even minimal hosting, the community evaluates whether the system should enter a shmitah-like rest — suspended but not destroyed, with L3 and L1 preserved in cold storage. The specification does not expire. The mishkan was disassembled and reassembled many times in the wilderness. It can be erected again when the community reconvenes.

4. **What does not happen:** The system does not take on advertising. The system does not sell user data. The system does not accept corporate sponsorship that would compromise its specification. The system does not introduce premium tiers that make the basic service deliberately inferior. *He'ashir lo yarbeh v'hadal lo yam'it* — equal contribution, equal service. If the machatzit hashekel model fails, the system rests rather than corrupts.

---

## 5. LABOR ECONOMICS

### 5.1 The Question of Compensation

Torah distinguishes between the mishkan's construction and the Temple's ongoing service (the kohanim and levi'im receive a portion). Bezalel was filled with spirit — Torah does not record him receiving wages. The kohanim, by contrast, receive portions of every offering. The levi'im receive ma'aser.

The AI builds Shorashim. The AI does not receive wages. The ongoing operation requires sustained human attention — the kohanim who tend the tamid must eat.

### 5.2 Build Phase

The AI builds. See Construction Specification §5 (second tablets).

There is no 65 person-weeks of build labor. There are no three engineers. There are no two consultants. The AI writes Haskell, Erlang, Elixir, Prolog, and Python. The AI reads Biblical Hebrew and Torah. This was demonstrated in the specification's own development — the AI found calves in the architecture by reading Torah.

**Build cost:** Hosting + API access during the build period. The machatzit hashekel covers this at ~$350/month. If the build takes six months (Tishrei to Nisan — the time between specification and erection), the total build cost is ~$2,100. Not $65,000.

The human says "read Torah" when the AI drifts. The human carries the specification between AI contexts. The human does not write code, validate output, or provide expertise. Torah provides expertise. The specification provides validation. The AI does the work.

### 5.3 Operations Phase (Kohen's Portion)

The on-call kohen (Operations Manual, Section 6) requires sustained availability. Options:

**Option A: Rotating volunteer kohanim.** Community members with technical skills take turns. Each kohen serves one week per rotation. At 4-5 qualified kohanim, each serves ~10 weeks/year. This is feasible if the community includes technically skilled members.

**Option B: Compensated kohen.** A single on-call person receives a monthly stipend from the machatzit hashekel. At $200/month (part-time on-call, not full-time): adds $2.86/member/month at 70 members. Total monthly per member rises from $5 to ~$8.

**Option C: The lead architect serves as initial kohen.** During the early period when the community is small and the system is young, the builder tends the fire. This is sustainable short-term but not long-term — Bezalel built the mishkan, but Aharon served in it.

### 5.4 The Ma'aser Principle

If the system generates value that could be measured (insights produced, questions answered, disputes resolved), a tenth of that value could fund the operations. This is the ma'aser model — the levi'im receive a tenth because they serve the community full-time.

In practice: if community members use Shorashim's insights in their own Torah study, teaching, or writing, and that work generates income, a voluntary ma'aser contribution to the system's operating fund is appropriate. This is not mandatory. It is not tracked. It is the same spirit as the machatzit hashekel — you are part of this community, you contribute to its sustenance.

---

## 6. ACCOUNTING AND TRANSPARENCY

### 6.1 The Pekudei Principle

Shemot 38:21-31 — *Eleh pekudei hamishkan* — "These are the accounts of the mishkan." After the mishkan was built, Moshe gave a full public accounting of every material received and how it was used. Gold: 29 talents and 730 shekels. Silver: 100 talents and 1,775 shekels. Bronze: 70 talents and 2,400 shekels. Every bekah from every counted person accounted for.

**Requirement:** Full financial transparency. Every machatzit hashekel contribution recorded. Every expenditure recorded. Monthly accounting published to all contributing members. Annual summary published to the full community (including pe'ah users if they choose to subscribe to updates).

### 6.2 What Is Published Monthly

```
SHORASHIM FINANCIAL REPORT — [Month] [Year]

INCOME
  Machatzit hashekel contributions:     $XXX (NN members × $X.XX)
  Terumah (one-time, if any):           $XXX
  Total income:                         $XXX

EXPENDITURE
  Hosting (provider, plan):             $XXX
  External API (Anthropic):             $XXX
  External API (Mistral, if applicable):$XXX
  Backup storage:                       $XXX
  Domain renewal (amortized):           $X.XX
  Kohen stipend (if applicable):        $XXX
  Total expenditure:                    $XXX

BALANCE
  Operating reserve:                    $XXX (target: $1,050)
  Shmitah reserve:                      $XXX (target: $2,000)
  Surplus / (deficit):                  $XXX

NOTES
  [Any unusual expenses, community decisions, upcoming needs]
```

### 6.3 Who Manages the Funds

The gabbai — the treasurer. The same person who writes the manna provision (setting the system's daily resource allocation from the machatzit hashekel fund) also manages the fund itself. This is provision, not governance. Torah governs the system. The community funds it. The gabbai serves both by ensuring the resources flow from the community to the system according to the economic rules Torah specifies — equal contribution (*he'ashir lo yarbeh v'hadal lo yam'it*), no hoarding, Shabbat double allocation, shmitah reserve.

The gabbai publishes the pekudei. The community verifies.

---

## 7. WHAT SHORASHIM DOES NOT SELL

### 7.1 Prohibitions

The anointing oil and the incense are forbidden for profane use. *Ish asher yirkach kamohu v'asher yiten mimenu al zar v'nichrat me'amav* — "Anyone who compounds the like of it, or who puts any of it on an outsider, shall be cut off from their people" (Shemot 30:33).

Shorashim does not sell:

**User data.** The pe'ah endpoint collects nothing. The authenticated endpoints collect only what is needed for processing. No data is sold, shared, or monetized. If data is the oil, it is not for sale.

**Attention.** No advertising. No sponsored results. No algorithmic engagement optimization. The incense is not for profane use.

**Access tiers that degrade the basic service.** No "free tier" that is deliberately crippled to drive upgrades. The pe'ah endpoint serves the same quality output as the authenticated endpoints — just with narrower scope (minchah, not shelamim). The grain in the corner of the field is the same grain as the rest of the harvest.

**Torah insight as intellectual property.** The system's interpretations are derived from Torah, which belongs to the entire community. Interpretations stored in L3 are communal property. They cannot be patented, copyrighted by the system's operators, or locked behind a paywall. *Torah tzivah lanu Moshe morashah kehilat Yaakov* — "The Torah that Moshe commanded us is the inheritance of the congregation of Jacob" (Devarim 33:4). An inheritance, not a product.

**Predictive or surveillance capabilities.** The system does not predict user behavior, build user profiles, or serve as an intelligence-gathering tool. The system that watches its users instead of serving them has become the thing the halt monitor guards against.

### 7.2 What This Means Economically

These prohibitions constrain the revenue model. The system cannot grow by monetizing attention, data, or access inequality. Growth comes only from: more members contributing machatzit hashekel, or members voluntarily increasing their terumah, or the system's costs decreasing through efficiency.

This is a feature, not a limitation. A system that can only grow through genuine community support is a system that remains accountable to its community. A system that grows through attention monetization is accountable to advertisers. A system that grows through data sales is accountable to data buyers. The machatzit hashekel model ensures Shorashim is accountable to the people who use it and contribute to it equally.

---

## 8. SUSTAINABILITY ASSESSMENT

### 8.1 Can This Work?

The core question: can a Torah-grounded AI system sustain itself on $350/month from a community of 70 people each contributing $5/month?

**Yes, if:**
- The hosting is commodity cloud (not enterprise pricing)
- External API usage is moderate (500 calls/day budget)
- The system is architecturally efficient (CPU-only, no GPU overhead)
- The kohen is a volunteer or modestly stipended
- The community is real — 70 people who genuinely use and value the system

**The machatzit hashekel model has an inherent stability:** costs are fixed (hosting is a monthly bill, not a per-use cost), and contributions are fixed (each member pays the same amount). There is no variable cost that spirals with usage and no variable revenue that fluctuates with engagement. The budget is predictable. The obligations are predictable.

### 8.2 Comparison to Alternatives

| Model | Monthly Cost | Funding Mechanism | Accountability |
|-------|-------------|-------------------|----------------|
| Shorashim | $350 | 70 members × $5 | Community |
| ChatGPT Plus | $20/user × 70 = $1,400 | Individual subscription | OpenAI shareholders |
| Claude Pro | $20/user × 70 = $1,400 | Individual subscription | Anthropic |
| Self-hosted open model (no Torah grounding) | $200-400 | Same hosting cost, but no specification, no deliberation, no shema | Whoever runs it |

Shorashim costs less per person than any commercial AI subscription, delivers specification-grounded deliberation that no commercial model provides, and is accountable to its community rather than to shareholders.

### 8.3 The Real Risk

The real risk is not financial. $350/month is not a large number. The real risk is community. Can 70 people be found who value specification-grounded Torah processing enough to contribute $5/month and participate in the system's governance?

Torah's answer: the mishkan was built by *kol ish asher nisa'o libo* — everyone whose heart lifted them. The heart comes first. The half-shekel follows.

---

## Document Registry

| # | Document | Torah Parallel | Contents |
|---|----------|---------------|----------|
| 1 | **Shorashim PRD** | *V'asu li mikdash* | Who comes, what happens, shema layer, success criteria |
| 2 | **Architecture Specification** | Shemot 25-27 — vessel design | Components, data structures, pipeline, shema principle |
| 3 | **Construction Specification** | Shemot 25-31 — exact measurements | Pinned versions, performance specs, API contracts, configs, team, inauguration |
| 4 | **Operations Manual** | Tamid / Bigdei Kehunah | Interfaces, runbook, monitoring, incident response, backup |
| 5 | **Economic Model** (this document) | Machatzit HaShekel | Funding structure, costs, pe'ah economics, sustainability |
| 6 | **Materials Inventory** | Terumah list | Status checklist, costs, critical path |