Light Table: Experimental Public Testing Release
The unsigned Windows x64 experimental testing release v0.9.0-revival.1 and our restoration source are now public. Runtime workflows require Node 24+ and PowerShell 7+; Java, Python and Ollama are optional for their respective features. Clean-machine certification, hosted CI and a fresh full regression remain pending. This is not a supported or full-parity release.
Private receipts were read without running or changing the restoration. Raw test transcripts and program source files are not published here. The project screenshot preserves the actual interface and work-in-progress status; private paths and conversation are covered with solid boxes. The tables are a reviewed summary, not externally reproducible public logs.
| Checkpoint | Recorded evidence |
|---|---|
| Save preservation | Final verifier 04:28 UTC. 41 original-editor checks; complete 31-entry regression; Vite 33/33; 61 receipts bound to 222 files. |
| Live state and local languages | Final verifier 13:44 UTC. 7 transformation and 10 native state checks; complete 37-entry regression; Vite 33/33; demo 5/5; 68 receipts bound to 246 files. |
| External language transports | Final verifier 14:49 UTC. 25 backend, 19 editor and 8 protocol checks; complete 39-entry regression; 71 receipts bound to 261 files. |
| Memory allowances | Focused receipt 15:08 UTC. 134 integration checks and 5 gate checks; local families 1 GiB, Clojure heap 1 GiB/family 2 GiB. |
| Assistant V1 baseline | Final verifier 18:44 UTC. 41 controller / 17 native-assistant checks; complete 41-entry regression; fresh Vite 33/33; demo 5/5; 74 receipts bound to 277 files. Separate real-model web 8/8 and Python 3/3 receipts. |
The assistant baseline verifier completed at and recorded no owned proof processes/jobs/helpers remaining. Later context/Stop/read-loop refinements pass 50 focused controller checks, 20 native checks and the demo 5/5. They are recorded separately and are not labeled as another completed full regression. The newer context-picker follow-up passes 50 controller and 22 native checks, including an actual custom 24,576-token request. These focused receipts bind their relevant implementation files and do not replace the older frozen full baseline. An earlier save regression failure was retained and superseded by the completed save checkpoint; the old pending-save website draft is not the current status.
Release-specific checks for v0.9.0-revival.1: fresh ClojureScript rebuild; 50 controller, 27 native original-editor, 6 disposable JavaScript worker, 5 provider-registration and 3 extracted-package smoke checks. The packaged editor verified an actual 42-to-43 source-versioned edit. These focused checks do not represent a fresh full regression or clean-machine certification.
Still open: clean Windows installation, hosted CI, distribution hardening, full legacy/live-coding parity, arbitrary module state migration, a full React workflow, remaining linked-view/reload cases, language services and later assistant integrations. An unsigned experimental binary is available; no supported release or upstream contribution is claimed.
The current project photo is an actual in-progress editor, live-preview and Ollama session captured on 5 October 2026. It shows that session, not a complete regression result. The early synchronous proof is preserved as historical evidence. The Ollama integration article, restoration story and live-editing article explain the difference. Original authorship and Light Table's MIT license remain credited.
Retro Visits: Public Source, Deployed Tool
The current workshop GitHub source includes the persistent Python/SQLite counter and its client. Counter activation is recorded in the public activation commit. GitHub Pages hosts the site; the dedicated backend retains the aggregate. Its route allowlist accepts homepage loads only.
The counter documentation describes ten backend test groups and ten browser fixture states. The backend tests and browser tests are public. The new blog pages do not increment the counter or change its backend configuration.
Page views are approximate accepted loads, not unique people or verified human traffic. See the counter story for retry data, privacy boundaries and failure behavior.
Read the Summary as Data
Download the public evidence summary as JSON. It contains the reviewed labels and counts above, with no raw private logs or machine paths.