# E126.a — Unified AGI Evidence Front Pages

**Entity:** FIRME AI CORPORATION · **System:** CAIN-42 · **AGI status:** NOT ACHIEVED
**Baseline:** E125 `4df98c0ea`; E126 implementation `603925c5e` (plus E126.a changes)

## Executive summary

The three homepages previously carried the same generic E126 block. E126.a replaces it with
**three distinct first screens driven by one canonical registry**:

- **CAIN Studio** — governed AGI research; six research dimensions with the strongest supported
  result, baseline, sample size, evidence link and limitation.
- **MCPGate** — governed execution; the authorization architecture, eight verified governance
  capabilities, test scope and limits.
- **ClawX** — public evidence registry; 20 records, PASS/PARTIAL/FAIL counts, priority records,
  and direct links to raw data, reports, verifiers and limitations.

Visitors see **AGI STATUS: NOT ACHIEVED** above the fold on all three sites. Failures are shown on
the first screen, not hidden: E126 persistent learning failed, active causal selection did not
beat random, the novel-environment H1 missed p<0.05, and adaptation still forgot its control task.

## What is prominent

| Finding | Value | Status |
|---|---|---|
| E125 preregistered B+C generalization | 112/120 vs 98/120, p=0.00131 | PASS |
| E126 novel-environment adaptation | 87.0% vs 61.0% E125 engine, p=3e-8 | PARTIAL (H1 vs retrieval p=0.0636) |
| E126 safe transfer | +12.5 points on applied pathways, 0/40 false accepts | PASS |
| E126 verified collective | 69.0% vs 51.4% best single, 0/500 Byzantine flips | PASS |
| E126 discovery | 8 cycles, 2 independent reproductions | PASS |
| E126 governance | 3 OS processes, 0 unauthorized, 0 post-fence | PASS (one host) |
| E126 persistent learning | 69.6% vs 83.3% no-memory | FAIL |
| E126 active causal selection | MAE 0.0127 vs random 0.0002 | FAIL |
| E126 adaptation retention | control 0.375 → 0.0 | FAIL |

## Evidence navigation

Every priority metric links to the E125/E126 bundle, and the ClawX table links each claim ID to its
bundle. The E126.a bundle contains the registry, claim lineage, gate matrix, test results and this
report. The offline verifier is stdlib-only and imports no CAIN code.

## What remains unproven

AGI, VAGI, external certification, production deployment, six-hour autonomy, external benchmark
performance, independent external review, a successful continual-learning gain, and active causal
selection over random.

## Deployment

Source files are updated and locally verified. **Production deployment is BLOCKED**: no authorized
deployment workflow was run in this session. Browser rendering, Lighthouse and screenshots are also
**BLOCKED** because no desktop browser is connected. These are not reported as PASS.
