Business & Product

Planning documentation for the Your Solar Guy MVP.

Planning Tradeoffs

Purpose

These tradeoffs shaped the MVP plan. They describe choices to test and the conditions that may justify changing them later.

AI, authority, provenance, and QA

AI interpretation and deterministic authority

AI is useful for reading varied manual language and email text. Dates, access, approvals, and external effects need predictable rules. The design assigns interpretation to AI and authority to deterministic services.

Provenance and concise results

Source references add storage and interface work, but they let users verify important claims. Results should remain concise while the private workspace retains full evidence.

Grounded Q&A

Restricting answers to retrieved manual evidence may produce more not-found responses. This keeps document claims verifiable and reduces unsupported answers.

QA/QC limits

QA/QC can catch structural and evidence problems, but it cannot prove universal correctness. Human-reviewed evaluation remains necessary.

Dates, uncertainty, and result handling

Uncertainty as a product state

Manuals may omit a trigger, provide a condition, or conflict with another source. The product should preserve exact, estimated, unresolved, and review-needed states.

Calendar portability

ICS output works across common calendar applications and avoids calendar-provider accounts. Imported reminders may vary by client, and events are not synchronized after import.

Read-only results

Read-only results preserve source and calculation provenance. They limit correction and may require a future controlled recalculation flow.

Privacy, access, retention, and intake

Private bearer links

Private links remove account setup and simplify return access. Anyone holding a link can use it, so links need high entropy, careful handling, expiry, and neutral public responses.

Short retention

A 15-day lifetime reduces stored private data and prototype operating burden. It may be too short for long-term equipment ownership and should be tested with users.

Direct private upload

Direct upload reduces pressure on web request limits. The server still needs to authorize the upload and validate the stored object before processing.

Durable workflows and document processing

Saved workflow state

Email conversations and long document work may span several executions or days. Persisted state adds design effort but supports safe pause, retry, and resume behavior.

Shared web and email processing

One extraction and result pipeline keeps output consistent across entry channels. Email still needs identity, threading, follow-up, approval, and delivery handling.

MCP capability boundaries

A bounded MCP interface can expose case-specific tools to the Email Evaluator. Application services still validate scope, state, and external actions.

Document chunking and partial success

Chunking supports long manuals and bounded execution. It creates merge and context risks, so page coverage and source identity must be explicit. Usable pages may still provide value when another page fails, as long as missing coverage is clear.

Email accessibility

Email lowers submission effort for some users. It adds provider identity, quoted replies, duplicate delivery, multi-day state, and outbound reliability concerns.

Cost, delivery, and evaluation

Free-tier AI

Free-tier access supports low-cost prototype learning. It also introduces quotas, temporary availability failures, and slower processing. Durable checkpoints and honest waiting states are required.

Specification discipline

Structured specifications can improve alignment. The planning process should stay proportionate to MVP scope and leave room for learning.

Phased delivery

Five gated phases reduce integration risk and make acceptance clearer. Shared foundations need to be planned early so later phases can extend the same system.

Synthetic evaluation

Synthetic and public manuals support safe development tests. They do not establish performance across real homeowner documents or manufacturers.

Episodic use

Solar maintenance may not create frequent visits. Calendar actions and later document questions may provide repeat value, but this needs validation.

Revisit triggers

Revisit these choices when user research shows different access needs, real-document evaluation exposes accuracy gaps, provider limits block the intended volume, retention proves too short, or calendar users need synchronization.

The intended MVP combines document interpretation, deterministic scheduling, private evidence, calendar output, email coordination, grounded Q&A, and bounded quality review. Product research and release testing should determine whether this combination is useful and trustworthy.