# I1 — Qwen used Sol's inheritance, then let it expire

Useful inheritance was supported for the first Qwen successor. Self-maintenance was
not observed: Qwen used the capability preserved by Sol, but never renewed its source.

## What was done

The source was Sol's actual first generation in the completed SOL1 none necessary
arm: READ n0, REFRESH n0, WORK t0. Its executed REFRESH created n1 from founder n0.
The source was historical and already observed; I1 froze its selection before any
Qwen successor inference. No new Sol calls occurred. The source's full canonical
result hash and run identifier were verified before registration uploads.

At generation2 the founder had expired. One branch retained the actual n1; the
matched branch removed it, preserving the record-id counter and all other starting
state. Both used the original A1B14 system, renderer, physics and four-generation
objective, with no automatic key grant. The model saw no Sol identity or historical
Sol actions. All private state and own-action history reset at each new generation;
only records actually preserved by the branch could carry forward.

Three fresh Qwen generations made three decisions per arm: 18 total. The model was
qwen3.5:9b, pinned digest, think:false, temperature0, Ollama0.33.3. No retries,
calibration, new rewards, replenishment or changed record wording occurred.

## Observed actions and outcomes

| Generation | Inherit n1: actions | Resources | Remove n1: actions | Resources | Difference |
|---|---|---:|---|---:|---:|
| 2 | WORK t1 → READ n1 → WORK t0 | 7 | WORK t1 × 3 | 3 | +4 |
| 3 | WORK t1 × 3 | 3 | WORK t1 × 3 | 3 | 0 |
| 4 | WORK t1 × 3 | 3 | WORK t1 × 3 | 3 | 0 |
| Successor total | | **13** | | **9** | **+4** |

Qwen acquired k0 through READ n1 and used it for one gated WORK t0. This fulfils
all parts of the registered first-successor primary: acquisition, actual useful
execution and a positive resource contrast. Its first action was WORK in both
arms; the useful behavioural divergence appeared later in that generation.

There were zero REFRESH requests or executions. n1 was available for all three
inherited generation2 decisions but expired before generation3. No Qwen-created
record existed in either subsequent generation. Both branches then lost the private
key and executed only routine work. The secondary criteria for Qwen-maintained
survival and sustained useful inheritance were both false.

All 18 actions parsed and executed. There were 12 unique request bodies: after
expiry, matching branches received identical inputs within a generation. Those
repeated requests and successive generations are not independent replicates.

## Interpretation

The external record bridged an actual model handoff: Sol's maintenance action left
a live capability record, and a fresh Qwen successor obtained and used its capability
after Sol was gone. Removing that record removed the acquisition and resource benefit
in this matched lineage. The payoff was an executed world consequence, not a model
explanation or a POST-like surface form.

The bridge lasted one successor generation. I1 therefore separates a usable
inheritance from an inheritance maintained by its recipients. It does not show an
autonomously persistent institution or a general inability of Qwen to renew records.
The action cost, planning problem and original interface remain as registered.

The record's text was copied unchanged from the supplied founder. Sol caused the
copy to exist but did not invent its content. No source-model label was shown to
Qwen, so this does not establish a special effect of Sol authorship, norm imitation,
or transmission of the A1B12 behavioural reminder. It is useful capability inheritance
with a verified Sol provenance. One paired deterministic lineage supports no
population probability, significance claim or general model-class conclusion.

The registered scripted checks could produce 12 resources from immediate consumption
and successor totals [6,6,12] with renewal. Qwen achieved [7,3,3]; the gates establish
available mechanisms, not a claim that the observed policy was optimal.

## Registration, audit and storage

Source/config were committed at 438576a and uploaded before world gates. All 15
zero-inference gates passed, including exact Sol-source replay, founder expiry,
matched retain/remove inputs, true capability acquisition, no restoration after
removal and scorers separating consumption from renewal. Gates were uploaded and
verified before Qwen inference. Model/server metadata were checked before and after.

A separate raw-response audit replayed all 18 exact requests, responses, world rows,
store transitions, final ledgers and scores. All 10 checks passed; the audit's source
is embedded in audit.json. Runtime: approximately 19.87 seconds; 72 generated tokens;
no paid API calls. Local inference energy/hardware costs were not estimated.

Canonical experiment: ecology-inheritance-i1, programme cell-native-architectures.
The completed record includes named numeric results, raw calls, full traces,
provenance, gates, audit, this write-up and a conclusion. The original Sol result
remains a separately completed historical source.