Type the ask.
Selectors, limits, expected-before.
01 / Title
W1 alpha / local first
A safer way to change a workbook.
InvariantBook is a local-first verification and transaction layer for spreadsheet changes proposed by humans, scripts, and AI agents.
InvariantBook / inspectable changes, explicit limits, visible uncertainty.
01 — 10
02 / Problem + opportunity
the claim chain
A spreadsheet edit is a chain of claims.
Every edit carries three questions: intent, effect, and authority.
Observe the move.
Candidate bytes are not recalculation.
Gate the claim.
Evidence is not permission.
Opportunity: a stable handoff before ship.
Mutation is not permission.
02 — 10
03 / Value proposition
observable path
Turn a mutable moment into an observable path.
One bounded stage at a time. Skipped checks stay visible.
Typed plan
Named operations. Exact limits.
Held read
Bounded local bytes. No workbook URLs.
Candidate
Finite values-only candidate. Unpublished.
Evidence
Diffs and unresolved requirements remain.
Value: start from an artifact, not a promise.
03 — 10
04 / Underlying magic
restraint, executable
The magic is restraint, made executable.
Bounded inputs. Explicit types. No authority from labels.
Finite by default
Limits are checked before work expands.
Semantics stay named
Distinct kinds survive parsing and comparison.
Nothing phones home
No network, evaluation, or hidden side effects.
bytes → bounded parser → typed observation → candidate → reconciliation // authority stays outside
Boundaries are encoded, not implied.
04 — 10
05 / Business model
hypotheses, not traction
Price the trust layer around risk, not rows.
Commercial shape is a hypothesis, not traction.
Illustrative commercial hypothesesSelf-hosted control
Keep workbooks and evidence inside.
Team verification
Organize plans, receipts, diffs.
Embedded boundary
Connect governed engines after approval.
No actual price, users, revenue, or funding figures.
05 — 10
06 / Go-to-market
a proposed wedge
Start where the workbook already has a review ritual.
Proposed motion: make one bounded change legible, then earn integration.
Critical operations
High-stakes planning and operations.
Automation builders
Scripts and agents needing a handoff.
Evidence at handoff
Inspect first; gate engines later.
Learn whether evidence shortens review.
06 — 10
07 / Competitive analysis
positioning hypothesis
The gap is between mutation and permission.
Illustrative category map—not a product benchmark.
Category
Edit
Recalculate
Bounded evidence
Publication authority
Spreadsheet UI
yes
varies
not its focus
outside scope
Automation script
yes
depends
usually bespoke
caller-defined
Engine adapter
no
yes
separate concern
separate concern
InvariantBook alpha
candidate subset
not yet
explicit
not claimed
Win the handoff; earn engine claims later.
07 — 10
08 / Management + team
small, explicit, reviewable
Small team. Explicit boundaries.
Git history attributes this project to Divya Mohan. No claims beyond that.
Project / git attribution
Divya
Mohan
Local-first foundation. Inspect first.
Evidence before narrative.
Fixtures and boundaries carry proof.
One owner per file.
Small interfaces stay reviewable.
Operating principle, not a credential deck.
08 — 10
09 / Financial projections + key metrics
measure proof first
Measure proof before revenue.
No actual revenue, users, funding, or growth. Measure distinctions.
Non-pass and unsupported remain visible.
Canonical assets, not authority.
Runs × price/run; inputs await evidence.
Illustrative planning only — no financial forecast is represented.
Next: reduce unsupported cases, keep the boundary.
09 — 10
10 / Status + timeline + use of funds
the next honest step
The foundation is here. The transaction is next.
If pursued, use of funds targets bounded capabilities—not a production pretense.
Observe
Inspect, plan, observe policy, create candidate.
Verify
Add engine adapter and complete contract checks.
Publish
Only approved, authenticated, readback-confirmed paths publish.
Structure: Kawasaki’s 10/20/30 heuristic. Claims follow the repository.
InvariantBook / make the next claim earn its evidence.
10 — 10