Aevum Phase-Wise Gap Remediation Master Plan
Date: 2026-04-18
Author: Codex acting as PM, BA, CTO, Chief Architect, CIO, and CISO review board
Execution team: Antigravity engineering, QE, and BA
Implementation mode: phased delivery in logically collated chunks
Testing policy for this plan: full integrated testing happens only at the end; implementation phases must still produce validation evidence and review artifacts
Policy basis: Apple-first, local-first, sovereign system, no unapproved external dependency
1. Executive Position
Aevum is now past the “is there a real product here?” stage. There is one.
The challenge is no longer invention. The challenge is disciplined completion:
- remove architectural drift
- complete the real user loop
- close the false surfaces
- convert partial systems into coherent Apple-native product behaviour
- ship a product that feels inevitable, not improvised
This plan is designed to stop Antigravity from wandering, assuming, or shipping attractive incompleteness.
It is phase-based, guarded, Apple-compliant, and optimized for a product that could plausibly become strategically attractive to Apple because it aligns with:
- on-device intelligence
- privacy-first personal computing
- Apple OS-native invocation
- low-friction voice interaction
- strong local data sovereignty
2. Review Board Synthesis
PM View
The biggest product risk is not missing ideas. It is false completeness. If a button exists, it must work, be hidden, or be labeled. Nothing in between.
BA View
The current system already contains multiple business capabilities, but they are not yet fully expressed as reliable user journeys. Capabilities must be mapped to actual user outcomes, not internal engine names.
CTO View
Architecture must be narrowed to one executable truth. The system cannot keep multiple partly-overlapping write paths, multiple maturity signals, and multiple implied futures without explicit boundary documents.
Chief Architect View
The next step is not broadening the system. It is normalizing lifecycle, state contracts, capture contracts, feature exposure, and module ownership.
CIO View
Operational maturity is too low for scale claims unless the product can be governed through release matrices, observability, and controlled rollout rules.
CISO View
Any cloud fallback language, hidden telemetry, misleading sync implication, or ambiguous data path weakens the security story and violates the sovereignty promise.
3. Current Remaining Gap Register
Critical Gaps
G1 — Sovereignty contradiction still exists in behaviour layer
There is still a cloud-consent / background GovCloud conceptual fallback inside the intelligence flow. This is incompatible with the current tooling and execution policy unless explicitly approved.
G2 — Feature exposure is still not centrally governed
The codebase still lacks a single release truth matrix enforced at implementation time. This invites Antigravity’s common failure mode: visible UI with incomplete backing logic.
G3 — “Capture” is functional but not yet complete as a lifecycle product loop
Aevum captures well enough to demonstrate promise, but it does not yet fully close the loop from:
- capture
- persistence
- reflection
- resurfacing
- return habit
Without this loop, users may admire the system without becoming dependent on it.
G4 — Retrieval and resurfacing are weaker than capture
The long-term value of the product depends on users feeling that speaking to Aevum compounds value over time. Capture alone does not create that feeling. Retrieval, continuity, and high-signal resurfacing do.
G5 — Some surfaces are still placeholder-like rather than data-real
A system that visually looks “complete” but uses static or synthetic representations erodes trust. This includes widgets, timeline-adjacent surfaces, and any screen whose content is not yet bound to real local data.
Major Gaps
G6 — Siri / conversational OS invocation is still not truly complete
The bootstrap issue is improved, but the conversational invocation experience is still not a finished Apple-grade flow.
G7 — Accessibility and automation readiness are not yet systematically enforced
If the product is to be Apple-grade and eventually automation-grade, accessibility identifiers and state exposure must become a design rule, not an afterthought.
G8 — Local API remains narrow and not yet productized as an extensibility boundary
That is acceptable for now, but the contract must remain clearly bounded and secured.
G9 — Performance and energy strategy are not yet formalized
The product is resource heavy. If it is to be viable on-device and strategically attractive, energy, memory, and latency budgets must become first-class architecture constraints.
G10 — There is no anti-placeholder implementation discipline baked into delivery
Antigravity needs explicit stop-rules to prevent:
- UI without action
- action without persistence
- persistence without retrieval
- feature labels without state transitions
- incomplete buttons that do not crash but also do not function
4. Target Product Outcome
The target product is:
an Apple-native, local-first personal intelligence companion that users want to keep talking to because every interaction makes future interactions more useful, more personal, and more emotionally resonant.
This requires six outcome pillars:
- frictionless capture
- trustworthy local memory
- meaningful continuity
- high-quality reflection and resurfacing
- OS-native availability
- energy-efficient on-device intelligence
5. Non-Negotiable Architecture Direction
The implementation plan must stay within the following architecture direction:
Apple-preferred technology posture
- SwiftUI for UI
- SwiftData with SQLite-backed local persistence as primary product store
- Core Data-level tuning only where SwiftData needs performance support
- App Intents for invocation
- WidgetKit for home-screen surfaces
- AVFoundation / Speech for voice flows
- EventKit for Apple calendar data
- LocalAuthentication for security
- BackgroundTasks only where behaviour is explicitly supportable in iOS rules
- OSLog and MetricKit for performance instrumentation
- local model execution using the approved local intelligence stack already in the repo
Explicitly avoid
- new SaaS dependencies
- hidden remote inference fallback
- third-party analytics
- cross-service sprawl
- architecture that bypasses local persistence contracts
6. Canonical Product Doctrine
Doctrine 1 — Nothing visible may be fake
If a feature is visible in a release build, it must be one of:
- fully functional
- explicitly labeled as preview/prototype
- intentionally hidden
Doctrine 2 — Speaking must compound value
Every major phase must strengthen one of:
- why the user speaks now
- why the user returns later
- why prior speech improves future utility
Doctrine 3 — Apple ecosystem first
Every integration and UX choice must ask:
- is there a native Apple way to do this?
- can we keep it local?
- will it behave predictably across iPhone lifecycle constraints?
Doctrine 4 — One screen, one truth
Every screen must have:
- defined entry conditions
- data source
- primary action
- success state
- failure state
- empty state
If any of these are undefined, the screen is not ready.
7. Phase Plan
Phase 1 — Product Truth Freeze
Objective
Freeze reality before further implementation.
Scope
- create release truth matrix
- classify all visible surfaces
- remove or hide any remaining false claims
- resolve policy conflicts, especially cloud fallback wording and logic
Deliverables
- feature truth matrix
- release exposure matrix
- policy conflict register
- screen-by-screen action map
Antigravity guardrails
- no new features in Phase 1
- no UI redesign in Phase 1
- no placeholder copy survives this phase
- any cloud reference must be justified against policy or removed
Validation evidence required before phase close
- every visible screen has an owner and a status:
real, partial, hidden, prototype
- every visible CTA has a backend function or is removed
- every data-bearing panel has an identified local source
Exit criteria
The team can answer, for every visible element: what it does, what data it reads, what state it changes, and why it exists.
Phase 2 — Core User Loop Completion
Objective
Complete the loop that creates habit:
capture -> store -> reflect -> resurface -> re-engage
Scope
- strengthen retrieval and resurfacing
- define memory continuity surfaces
- ensure recent speech affects what the user sees next
- prioritize the “why keep speaking?” mechanism
Required product outcomes
- user can see evidence that the system remembers
- user can see value from previous interactions
- user gets emotionally intelligent prompts grounded in prior local data
Candidate implementation areas
- dashboard continuity layer
- timeline relevance ranking
- “what changed since you last spoke” surface
- daily/local reflection summary built strictly from local memory
- persona-grounded resurfacing
Antigravity guardrails
- no generic motivational filler
- no prompt surfaces that are not grounded in stored data
- no summary cards using static text
- every resurfacing element must cite its local source path internally in code comments or docs
Validation evidence required
- three concrete return journeys documented
- each return journey maps to actual local data
- no resurfacing card is synthetic-only without disclosure
Exit criteria
A first-time user can capture. A returning user can feel accumulation.
Phase 3 — Capture System Completion
Objective
Turn capture from “many entry points” into a complete product subsystem.
Scope
- formalize canonical write contract in source docs
- make all supported capture modes complete
- remove or hide unsupported capture modes
- unify error handling and feedback
Required decisions
- dashboard direct write path remains privileged for low-latency capture
- queue-backed path remains for background/external/import/system flows
- imported media and text must have explicit post-ingest states
Antigravity guardrails
- no capture mode may appear unless it fully completes its success/failure loop
- every capture mode must define where it writes first
- if a button opens a flow, the flow must persist or clearly terminate with reason
- silent no-op is forbidden
Validation evidence required
- capture map document with all entry points
- first-write destination for each entry point
- user-visible completion messaging for each supported capture mode
Exit criteria
Every exposed capture entry point is functionally complete and contractually documented.
Phase 4 — Real Retrieval, Graph, And Insight Value
Objective
Convert stored memory into perceived intelligence.
Scope
- make timeline more useful than archival
- make graph more useful than decorative
- make insights more useful than labels
- prioritize retrieval quality and meaningful connection
Required product outcomes
- graph nodes represent real stored relationships
- timeline supports meaningful recall and not just listing
- insight surfaces explain why something was surfaced
Antigravity guardrails
- no graph animation without real underlying node semantics
- no insight label without generation basis
- no ranking logic without documented signal basis
Validation evidence required
- retrieval ranking note
- graph data provenance note
- insight generation provenance note
Exit criteria
The user feels that the app understands accumulated context, not just that it stores entries.
Phase 5 — Apple-Grade Invocation And Ecosystem Fit
Objective
Make Aevum feel native to iPhone usage patterns.
Scope
- refine App Intents and shortcuts
- finalize widget usefulness
- integrate only Apple-native capabilities that improve habit
- improve lock-screen/home-screen/accessibility behaviour
Recommended Apple-native priorities
- App Intents
- WidgetKit usefulness, not decoration
- EventKit-bound calendar continuity
- Core Spotlight for retrieval if needed
- accessibility identifiers across all key flows
Antigravity guardrails
- no OS surface may show static or misleading data
- widget must do something useful every day or be hidden
- Siri / shortcut flows must be concise, deterministic, and local-first
Validation evidence required
- invocation matrix across icon, widget, shortcut, deep link
- accessibility map for key elements
- lock-state and permission-state behaviour definitions
Exit criteria
Aevum can be accessed like an Apple-native tool, not only like an app someone remembers to open.
Phase 6 — Performance, Energy, And Sovereignty Hardening
Objective
Make the product efficient enough for serious daily use and credible enough for Apple-grade scrutiny.
Scope
- formal latency budgets
- memory budgets
- energy constraints
- model loading strategy
- strict elimination of non-policy cloud logic
Required architecture outcomes
- foreground capture remains responsive
- model initialization is controlled
- background tasks are explicit and policy-compliant
- no hidden network dependence outside approved download/storage use
Antigravity guardrails
- no feature may add unbounded background work
- no new heavy path may ship without a budget
- no “temporary cloud fallback” may be introduced
- model download UX must remain explicit and user-controlled
Validation evidence required
- performance budget sheet
- energy budget sheet
- network dependency matrix
- local-only compliance statement
Exit criteria
The system remains performant, private, and predictable under real daily conditions.
Phase 7 — Security, Trust, And Apple Suitability
Objective
Make the system suitable for premium trust positioning and strategic scrutiny.
Scope
- password/biometric lifecycle review
- local data path review
- consent flow cleanup
- diagnostic and error-surface review
- privacy narrative alignment to actual implementation
Antigravity guardrails
- no security prompt without real enforcement behind it
- no privacy claim without code path support
- no diagnostics that leak sensitive data in release builds
Validation evidence required
- security decision log
- privacy claim mapping
- release diagnostics exposure matrix
Exit criteria
All trust-related claims are justified by implementation.
Phase 8 — Final Integration, Testing, And Release Readiness
Objective
Only after implementation chunks are complete, execute full validation.
Scope
- package tests
- app lifecycle validation
- manual state-matrix
- bug burn-down
- TestFlight release gate
Required outcomes
- no orphaned UI
- no incomplete CTA
- no policy violations
- no unclear feature maturity
- clear release notes and scope statement
8. Antigravity Failure-Prevention Script
This section is written as direct execution instruction for Antigravity.
Rule A — No implementation starts without a feature card
Every work item must begin with:
- user outcome
- trigger
- local data source
- state mutation path
- success state
- failure state
- hidden dependencies
If any are missing, stop and request clarification.
Rule B — No UI without action wiring proof
Every button, toggle, card tap, widget tap, and navigation affordance must ship with:
- target action
- persistence effect or explicit non-persistence purpose
- error path
- disabled state if unsupported
Rule C — No feature without a state table
Before implementation, Antigravity must write a compact state table:
- empty
- loading
- success
- partial
- failure
- permission denied
- unavailable by policy
Rule D — No placeholders behind real navigation
If a route is reachable in the build, its end-state must be complete or explicitly labeled.
Rule E — No implicit assumptions
If Antigravity cannot identify:
- where data comes from
- where it is written
- what re-renders after write
- what user benefit it creates
they must stop and ask, not invent.
Rule F — No partial feature stitching
Antigravity may not “finish the screen later.”
The smallest unit of delivery is a logically complete slice.
Rule G — Every phase must update source-of-truth docs
At the end of each phase, update:
- feature truth matrix
- action map
- data flow map
- release exposure matrix
- known limitations register
9. Required Validation Steps During Implementation
This is not end-to-end testing. It is implementation validation to stop bad code from accumulating.
For every chunk, Antigravity must provide:
- compile proof
- dependency impact note
- changed-screen list
- changed-state list
- changed-persistence-path list
- explicit statement of what was hidden or removed
For every user-visible chunk, Antigravity must also provide:
- action inventory
- empty/loading/success/failure states
- privacy effect note
- performance implication note
10. Phase Delivery Template
Every phase handoff from Antigravity must use this exact structure:
1. Objective
What user or system problem was solved.
2. Completed scope
What was actually implemented.
3. Removed scope
What was hidden, deleted, or deferred.
4. Action map
For each visible CTA, what now happens.
5. Data map
Where data is read from and written to.
6. Risk note
What still remains fragile.
7. Policy note
How the work remains compliant with Aevum policy.
8. Ready-for-next-phase decision
Yes or no, with reason.
11. Recommended Build Order
Antigravity must implement in this order:
- Product truth freeze
- Core user loop completion
- Capture system completion
- Retrieval, graph, and insight value
- Apple-native invocation polish
- Performance and sovereignty hardening
- Security and trust hardening
- Final full-system testing and release readiness
Do not reverse this order. A beautiful but incomplete shell will consume time and hide defects.
12. Apple Acquisition / Strategic Attractiveness Lens
If Aevum is ever to become strategically attractive to Apple, its strongest position is not “another AI app.”
It is:
- deeply local
- deeply personal
- OS-native
- privacy-aligned
- voice-native
- on-device intelligence forward
- elegant under constrained hardware conditions
That means the right implementation bias is:
- fewer but deeper features
- native frameworks over exotic stacks
- trust and usefulness over noise
- continuity and memory over novelty
13. Final Directive To Antigravity
Do not optimize for the appearance of progress.
Optimize for:
- logical completeness
- state completeness
- honest capability exposure
- Apple-native fit
- user compulsion to return and speak again
Every delivered chunk must make the product feel more inevitable, more trustworthy, and more useful.