docs(observatory): map the VLTIMA 5-primitive kernel to the engine KERNEL.md#965
docs(observatory): map the VLTIMA 5-primitive kernel to the engine KERNEL.md#9654444J99 wants to merge 1 commit into
Conversation
Pin each VLTIMA primitive (Member/Mandate/Standing/Standard/Governance) to its implementing module in cli/src/limen/observatory, so the organ's kernel map is stated both in prose (organs/observation/KERNEL.md) and at the code level.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: a60a282651
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| `observatory doctor` exits 0 ⟺ the engine is wired, deterministic (a re-run | ||
| reproduces byte-identical derived state — the idempotent fixed point asserted by | ||
| `ledger.write_latest`'s `sort_keys`), shape-complete (each brief carries both faces | ||
| + exactly 3 mechanisms, 3 confounders, 1 hero, 1 experiment, 1 measurement | ||
| contract, no score using stars), and **read-only against public surfaces** (default |
There was a problem hiding this comment.
In this documented module path, cli/src/limen/observatory/__main__.py:58-70 always returns exit code 0 for doctor, and cli/src/limen/observatory/doctor.py:103-112 only aggregates wiring/pipeline/paths/params/registries/online rungs; it does not rerun the beat or assert byte-identical derived state, brief shape, or star-free scoring. Calling this the done-predicate can let CI/operators mark Observatory complete while these listed invariants remain untested, so either narrow this invariant to the checks doctor actually performs or add those checks to doctor.run().
Useful? React with 👍 / 👎.
| The default beat (`python -m limen.observatory run`) is **dry**: it writes only | ||
| under `logs/observatory/`. `--apply` is the one arm — it homes a lever idempotently | ||
| and promotes a reversible-prep task via tabularius. Nothing here opens a public PR |
There was a problem hiding this comment.
When VVLTVS is present, the default executive.run_beat(apply=False) includes the reconcile stage; reconcile.run() invokes VVLTVS via assess() and then plan_text() (cli/src/limen/observatory/reconcile.py:151-165), and scripts/vvltvs-organ.py:868-869 calls write_stamp() before handling both --json and --apply, which writes logs/vvltvs-organ-state.json and logs/.voice/vvltvs (scripts/vvltvs-organ.py:751-755). This makes the documented dry-run footprint incomplete, so cleanup/sandboxing based on this boundary will miss real writes outside logs/observatory/; either document the VVLTVS stamp side effects or stop invoking the stamping CLI for dry runs.
Useful? React with 👍 / 👎.
| ``` | ||
| observatory/ | ||
| config.py params + ground-truth registries (value-repos / revenue-ladder / estate-ledger) | ||
| gh.py the ONLY shell-out boundary — GITVS's token cascade, fails OPEN |
There was a problem hiding this comment.
This module map is used as the engine inventory, but gh.py is not the only subprocess boundary: reconcile._run_vvltvs() shells out to python3 scripts/vvltvs-organ.py, and interpret._default_invoke() shells out to claude -p when OBSERVATORY_LLM=1. Anyone using this map for offline/security review will miss those non-GitHub subprocesses, so please either describe gh.py as only the GitHub boundary or list the other shell-out sites here.
Useful? React with 👍 / 👎.
| **Confounder-honesty rule (stars are a signal, never truth):** `cohort.confounders` | ||
| flags `existing_audience`, `corporate_brand`, and `star_manipulation_suspected`; | ||
| `confounder_discount` multiplies them into a factor in `(0, 1]`, so a repo that won | ||
| on inherited advantage yields low-priority mechanisms *by construction*. Star |
There was a problem hiding this comment.
On the normal collection path, collect.snapshot() always stores owner_followers: None, so cohort.confounders() cannot emit existing_audience for collected winners even when a solo maintainer has a large inherited audience; only corporate-brand and star-ratio signals are active. The documented honesty rule therefore overstates what the score discounts, which can make operators trust high-priority mechanisms that were actually driven by audience advantage. Either collect owner followers or narrow this claim to the confounders the pipeline can observe.
Useful? React with 👍 / 👎.
1 participant
ORG-observation-organ-kernel-0711
Author cli/src/limen/observatory/KERNEL.md: the engine-level twin of organs/observation/KERNEL.md, pinning each VLTIMA primitive to its implementing module.
Kernel map (prose -> code)
Verification
Convergence, not greenfield: reuses GITVS gh-helpers (gh.py), the JSONL keeper discipline (ledger.py), value-repos/revenue-ladder/estate-ledger as ground truth, and face-ownership.json/VVLTVS as the internal-legibility sensor. One PR, no lorem.