Human + platform policy
Fixed, preferred, forbidden, consent, quality floor, and zero-spend constraints.
Workspace and operatorBlueprint · careervector.blueprint-graph/v5
Graph explorer
Twenty composable top-level diagrams plus recursively nested 21×21 semantic layers. Nodes descend; boundary arrows cross to neighbours.
Nested layer 1 · runtime · privacy · control · feedback
Private adaptive LLM routing
Human policy gates the feasible routes; additive addressed preferences rank them; private bounded feedback updates the same path without retaining prompts or keys.
- Human + platform policy to Additive option field: gate + shape
- Semantic address path to Additive option field: select prefixes
- Additive option field to Route + explanation: rank
- Feasible executors to Route + explanation: eligible set
- Reservation authority to Route + explanation: reserve
- Route + explanation to Bounded outcome: execute
- Bounded outcome to Path projection: normalize
- Path projection to Additive option field: learn
Semantic address path
root → workspace → capability → stage → consumer → instance?
Open layer ↘Additive option field
Every prefix adds signed preference, belief, uncertainty, and aging to each route option.
SporewrightFeasible executors
Local, peer, BYOK, and included routes pass availability, privacy, and entitlement gates.
CareerVector runtimeReservation authority
Tokens or currency micros are reserved before curiosity can spend them.
CareerVector D1Route + explanation
Ranked primary and fallbacks include a causal receipt and short user-facing reason.
CareerVector routerBounded outcome
Success, quality loss, latency, tokens, exhaustion, and failure class become numeric evidence.
CareerVector D1Path projection
Leaf evidence updates bounded ancestors and decays exactly on read.
Sporewright01 · policy · as-built Human + platform policy Fixed, preferred, forbidden, consent, quality floor, and zero-spend constraints.
Workspace and operator- 02 · process · as-built Semantic address path
root → workspace → capability → stage → consumer → instance?
Open deeper layer ↘ 03 · process · as-built Additive option field Every prefix adds signed preference, belief, uncertainty, and aging to each route option.
Sporewright04 · boundary · mixed Feasible executors Local, peer, BYOK, and included routes pass availability, privacy, and entitlement gates.
CareerVector runtime05 · store · as-built Reservation authority Tokens or currency micros are reserved before curiosity can spend them.
CareerVector D106 · process · as-built Route + explanation Ranked primary and fallbacks include a causal receipt and short user-facing reason.
CareerVector router07 · store · as-built Bounded outcome Success, quality loss, latency, tokens, exhaustion, and failure class become numeric evidence.
CareerVector D108 · process · as-built Path projection Leaf evidence updates bounded ancestors and decays exactly on read.
Sporewright
Flows
- Human + platform policy → Additive option field gate + shape
- Semantic address path → Additive option field select prefixes
- Additive option field → Route + explanation rank
- Feasible executors → Route + explanation eligible set
- Reservation authority → Route + explanation reserve
- Route + explanation → Bounded outcome execute
- Bounded outcome → Path projection normalize
- Path projection → Additive option field learn
Canonical architecture note
Open note ↗CareerVector
Product outcome
Help a person understand the market, manage opportunities, improve their materials, decide where to invest effort, and collaborate with trusted people and agents. Documents and AI are means; the private career-intelligence workspace is the product.
Authorities
- D1/Yjs owns workspace state, collaboration, CV/CL trees, notes, jobs, preferences, and application activity.
- R2 owns private workspace files and large artifacts.
- JobCache Core supplies shared public role/ad knowledge through a strict seam.
- Smart Routing selects among Included AI, consented Local AI, and user-provided AI within declared preferences, feasibility, quality, and zero-operator-cost constraints.
Invariants
- CareerVector is permanently free to users.
- Private candidate and workspace activity is not monetized.
- One workspace serves one person while supporting trusted collaboration.
- Product-local identity and private state remain below the privacy seam.
- Explicit user AI preferences are gates or policy, not suggestions for the router to erase.
- No AI success or quality is fabricated where no honest outcome signal exists.
Current state
The collaborative workspace, shared app/API/MCP foundations, document engine, job tracking, adaptive LLM routing, BYOK, and corpus seam exist in substantial form. Product experience, Local AI adoption, outcome learning, and several radar workflows continue to evolve.
Completion evidence
Human and agent journeys over the same workspace truth, offline/realtime convergence, privacy and erasure tests, honest AI-route explanation, and no-paywall product behavior.