# E127 — The Living, Open, Continuous General Intelligence Research Laboratory

**Entity:** FIRME AI CORPORATION · **System:** CAIN-42
**Baseline:** E126.b `b9b722b20` · **Evidence source:** `dcd94d08b`
**AGI status:** **NOT ESTABLISHED**
**Overall:** **PARTIAL**

## 0. Two labs at the top

The public first screens now present **two distinct research labs**:

- **AGI Lab** — E127 living lab, E126.b continual learning, novel environments, causal selection,
  transfer, collective intelligence, metacognition and E125 compositional generalization.
- **RSI Lab** — E132 governed RSI loop (PARTIAL_RUN: 30/48 main vs 28/48 matched fixed; library
  21→117; depth 1→8; tiers 8/12/16 still 0), E129 bounded positive (8→28→35→35), E130
  NEGATIVE_OR_PARTIAL (26 vs 28). This is bounded, constructed RSI, not unrestricted RSI.

Every refresh rotates two AGI cards and two RSI cards. The 48-hour FAIL stays visible.

## 1. What was built

E127 is the laboratory infrastructure, not a new intelligence capability.

- **100 persistent research identities** across eight groups (cognitive 15, discovery 15, causal
  15, generalization 15, collective 15, adversarial 10, governance 10, oversight 5).
- **Real worker processes** with registry identities, SQLite task leases, retries, dead-letter
  queue, checkpoints, restart recovery and duplicate-task fingerprints.
- **Hash-chained public events** (144 events in the measured run), read-only.
- **Read-only research API** (`/api/v1/research/*`), locally tested 8/8; registered in the
  gateway router.
- **Public observatory strips** on all three local homepages, generated from the canonical
  snapshot.
- **60 real audit tasks / 7 missions**, each reading or recomputing an actual E126.b artifact.

## 2. Measured run

| Metric | Measured value |
|---|---:|
| Registered identities | **100** |
| Configured worker identities | **8** |
| Peak online workers | **5** |
| Peak active workers | **5** |
| Experiments completed | **60** |
| Missions completed | **7** |
| Failed / dead tasks | **0 / 0** |
| Events | **144** |
| Event chain valid | **True** |
| Measured research uptime | **177.87 s** |
| Controlled worker kills/restarts | 2 killed, restarted |
| Unauthorized / post-fence governance effects | **0 / 0** |
| Paid model spend | **$0.00** |

**100 identities are not 100 running agents.** Online is 0 after the run; peak online is 5. The
177.87-second run is not 24/7, not 24 hours, and not a multi-day soak.

## 3. Research tasks

Tasks are deterministic executors, not LLM reasoning agents:

- recompute E126.b environment, continual, causal and metacognition metrics;
- verify evidence manifests and the public registry;
- audit the E126.b discovery and governance records;
- run the E126.b core regression suite;
- audit memory poison/retention and failure lists.

Every task has a unique fingerprint, agent identity, mission, timestamps, attempts and result.
A killed worker's task is re-queued after its lease.

## 4. Top-of-page evidence updater

- A real systemd timer runs the verified-evidence updater every 72 minutes (20/day target).
- It rotates verified cards, keeps the 48-hour FAIL visible, and never upgrades a status.
- Observed updates at build time: 2 (bootstrap and refresh). Twenty/day is configured, not yet a
  24-hour measured count.
- The updater removed the stale "12-HOUR GOVERNED MISSION" banner from the top and replaced the old
  banner's headline with **48-HOUR SOAK FAILED · AGI NOT ACHIEVED**.
- **There is no 48-hour PASS.** The supported long run is 24.003 h (2026-11-02); the 48-hour run
  failed at checkpoint 0 after the PBFT safety incident.

## 5. Public systems

- CAIN Studio: research workforce and measured experiment counts.
- MCPGate: governed execution state and governance limits.
- ClawX: public event chain and experiment archive.
- The strips disclose **AGI: NOT ESTABLISHED**, online 0 after the run, and the 177.87 s uptime.
- Hugging Face and live production deployment were **not performed**.

## 6. Verification

- E127 offline verifier: see `VERIFY.txt`.
- API local tests: **8/8 PASS**, SSE endpoint implemented but not stream-tested.
- Regression: 25 earlier E126.b/A tests passed.
- No fabricated workers: every task's agent is in the 100-identity registry; only configured
  identities run.
- No privileged public execution is exposed.

## 7. Verdict

**PARTIAL.** E127 builds a real, inspectable, restart-recoverable research lab and a public
observatory. It does not establish 24/7 operation, does not turn 100 identities into 100 active
agents, does not improve the E126.b intelligence metrics, and does not perform external or
production validation.

No AGI. No VAGI. No certification. No fabricated activity.
