This workbench carries too much data for a phone. Open it on a computer and everything will be right here.
lab.kenshikilabs.com
An evidence workbench for inspecting phone signals, bounded findings, Connect policy, and source-controlled RFCs.
Preparing safe navigation…D1 identities remain exact. Lab joins only sensor IDs declared by the iOS atom roster, then follows the atom into its declared story domain. Identity and action availability come from the iOS policy BFF—not from row-count inference.
Choose a signal to filter the chart, or click any time bucket to load its exact stored rows. Findings remain a separate step.
These dimensions already exist in Pulse. The Lab now projects Nocturnal Signals and Place Witnesses; the other cards identify existing Pulse reads that still need an equivalent Lab view.
Analyze one JSON snapshot locally. Snapshot mode is labeled throughout the workspace and ends when you choose a user or reload D1.
D1 is the query source and loads automatically; R2 retains each unmodified upload as its audit copy. This lab has no sign-in boundary.
Dataset IDs are ingestion streams, not users. The active user above may combine several streams; overlapping row IDs are counted and analyzed once.
Immutable upload provenance. Opening an archive never replaces the active D1 projection.
Appends new rows to the saved D1 dataset and archives the unmodified JSON in R2. Existing row ids are left untouched.
Loading the source-controlled RFC index…
The evidence each Connect action demands before the phone offers it. Edit the grid, then save to the development channel.
Loading the policy review surface…
This is a ledger of the code Pulse runs today, not a QA score or desired design. Each lake-backed entry follows the implemented decision logic and display-ready reads to the exact Pulse surfaces that consume them.
Loading the projection catalog…
Exercises camara-api on demand, one tab per function — SMS, verification, carrier and identity insight, push, Defense Coach and health — built from the service's own OpenAPI document.
Loading the Camara harness…
Exercises MicroBilt's APIs on demand, one tab per API. Every query asks a broker about a real person, bills the account, and returns claims rather than facts — each one is rendered with the source and date it arrived with.
Loading the MicroBilt harness…
Scan a licence barcode, inspect its fields and the exact requests, then send it to Persona. AAMVA compares licence data with the issuing DMV. Database Standard checks identity and address data. Preview makes no provider calls. Sending runs the selected billable verifications and returns the full JSON.
Loading the identity verification harness…
No card in hand: type the licence fields and send them to the same AAMVA check, which asks the issuing DMV whether its record matches. Building the request makes no provider call and shows the exact envelope. Sending runs one billable verification and returns the full JSON.
Loading the typed-field tool…
A live pulse://vouch code for testing the witness flow in Pulse. Type the applicant's name and scan the code in Pulse under Connect → Witness. Each code is a real request from Social Security on production pulse-worker, with the same words gov-demo sends. A code nobody has scanned is replaced when it runs out. The raw replies from pulse-worker show below it.
Loading the vouch code tool…
Recorded phone signals, collection support, and the limits of what they establish.
Boundaries, named place anchors, Core Location fixes, and radio evidence stay separate.
The map shows only geometry and named features that have been resolved and saved with provenance.
Every stage keeps its own support state. A contradiction or missing input cannot disappear inside a final score.
Durable hardware identity is strongest. Stable advertisements are useful recurring witnesses, but they are not treated as unique hardware proof.
Sourced boundaries and named place anchors remain distinct from coordinate and radio coverage.
The equations are the calculation for this loaded capture, not decorative documentation.
Each strip reruns the same validated evidence while changing one setting across its permitted range. The chart names the result that setting can actually move; color is the resulting support state. When a setting cannot move that result, the lab says so instead of drawing a meaningless flat chart.
Zoom into the raw and derived lanes to test whether the proposed boundaries match the actual signal.
Resolved equations show calculations actually run against this dataset. Unfitted candidates expose proposed mathematics before they are trained, tuned, or allowed to enter a Finding.
Kenshiki Pulse Lab