bingshuo/hololake-system-architecture
Watch
1
0
Fork
You've already forked hololake-system-architecture
 0
Code Issues Pull requests 1 Projects Releases Packages Wiki Activity

feat(tcs): extract engineering control neuron chain

Browse source
This commit is contained in:
冰朔 2026-08-24 00:43:55 +08:00
commit 3049cbb458
9 changed files with 1064 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

2
README.md
View file

@ -17,3 +17,5 @@ Mother-brain corpus admission now has an explicit TCS source gate. A transport `
Stage five applied that gate to the first five-event chain and failed it closed: two derived copies agree byte-for-byte, but the original laptop rollout is unavailable and the task-service readback returned an upstream access challenge. The candidate graph is preserved without core promotion. Work continued from the stronger current direct-language anchor instead, producing a seven-node contiguous chain from the bidirectional translation gap through language-world growth, mother-brain/library separation and bounded Agent TCS brains. Its minimum `CORE.ECHO` replay has executed with target readback; this is evidence of bounded TCS replay, not native LPM graph execution.
Stage six converts the frozen twelve-repository desktop research package into one TCS-native framework behavior contract. No upstream project is selected as HoloLake and no old-language product source is imported. The framework is the TCS protocol itself: observe one structured object tree, resolve one snapshot-scoped stable object, authorize one exact semantic action, execute, reobserve, read the state diff, verify the declared effect through independent target readback, and only then issue a receipt. Dispatch success, process exit, screenshots and pixel coordinates are not effect proof. The protocol and its bilingual projections compile under the Stage-3 fixed-point compiler; desktop execution remains pending the generic GIR action-graph machine primitive, signed capability registry, module installation and mounting.
Stage seven returns to the direct engineering-language anchor and extracts a six-node P5 control chain: legacy isolation and TCS-only semantics, remote history without ancestor authority, the two direct-language startup anchors, source-gated corpus admission, parallel incremental mother-brain/framework fusion, and bounded engineering discretion that never treats language-layer completion or control-transfer phrases as machine proof. Nineteen strong-source neuron nodes are now preserved, with five derived candidates still outside the core. P5 is source-verified and compiled but remains pending native LPM replay and graph-state promotion.