# Runtime-Only 50k Evidence Gate: Split-Control Reference Lane

## Scope

This note records the evidence gate for the current runtime-only 50k reference lane from:

- run id: `guarded-segmented-runtime-only-50k-validation-20260313-002`
- harness: `analysis/runtime_only_probe/run_guarded_segmented_medium_validation.py`
- mode order: `split_control` then `guarded_segmented`
- stage totals: `20000`, `50000`
- compile cadence: every `5000` accepted events plus stage-end

This is a control-lane gate note only.

It does **not** claim a full runtime-only 50k pass across storage paths.

## Executive Summary

- Preflight passed `15/15`.
- The current reference lane, `split_control`, passed the exact 50k lane on the current patch.
- The four key control-lane checkpoints all returned `http_status: 200`:
  - `stage_end_020000`
  - `restart_020000`
  - `stage_end_050000`
  - `restart_050000`
- Control-lane restart stayed healthy and split-unit cold reconstruction remained disabled at restart:
  - `startup_authority_split_task_list_units_ms = 0.0`
  - `startup_authority_split_task_read_units_ms = 0.0`
- Control-lane retained bytes stayed `0 / 0` through the validated 50k lane.
- The challenger lane, `guarded_segmented`, did **not** clear this gate and remains outside the acceptable storage/restart envelope based on observed 20k and post-20k evidence.

## What Was Run

The actual proving run for this note is:

```text
py -3 analysis/runtime_only_probe/run_guarded_segmented_medium_validation.py --run-id guarded-segmented-runtime-only-50k-validation-20260313-002 --output-basename 20260313_runtime_only_50k_validation --stage-totals 20000 50000 --compile-every 5000 --progress-every 1000 --http-timeout-ms 120000 --event-retries 2 --event-retry-backoff-ms 50 --task-id probe_task_guarded_segmented_runtime_only_50k_validation --step-prefix guarded_segmented_runtime_only_50k_validation --sleep-ms 0 --total-tokens-max 5000 --known-facts-max 400 --segmented-pathological-persist-p95-ms 750 --segmented-pathological-scan-p95-ms 250 --segmented-pathological-event-p95-ms 1500 --segmented-max-file-ratio-vs-split 2 --material-improvement-pct 20 --build-binary --run-preflight-tests
```

Important scope rule for interpreting this run:

- `split_control` is the current control/reference lane.
- `guarded_segmented` is a challenger lane.
- Only the control lane is cleared by this note.

## Proven

### Control-lane checkpoints

| Checkpoint | HTTP | Client elapsed ms | Daemon compile/open ms | Key supporting metrics |
|---|---:|---:|---:|---|
| `stage_end_020000` | `200` | `1301` | compile `1294.580` | event `224.553`, persist `15.480`, current bytes `20,718,267`, retained `0 / 0` |
| `restart_020000` | `200` | `130` | open `4912.709`, compile `109.141` | startup healthy, split list/read `0.0 / 0.0` |
| `stage_end_050000` | `200` | `4300` | compile `4251.124` | event `516.978`, persist `18.150`, current bytes `51,828,267`, retained `0 / 0` |
| `restart_050000` | `200` | `397` | open `23272.689`, compile `379.370` | startup healthy, split list/read `0.0 / 0.0` |

### Control-lane overall envelope from the proved 50k lane

- event handle p50 / p95 / p99: `341 / 758 / 1509 ms`
- persist p50 / p95 / p99: `16 / 43 / 58 ms`
- compile p50 / p95 / p99: `1521 / 5338 / 5338 ms`
- retained total bytes at validated 50k stage end: `0`
- retained obsolete bytes at validated 50k stage end: `0`

### Control-lane restart/open behavior

The control lane stayed restart-healthy at both stage boundaries:

- `daemon_stdio.log` shows post-boundary `startup_healthy` after the 20k restart
- `daemon_stdio.log` shows post-boundary `startup_healthy` after the 50k restart
- `daemon.open.instrumentation` shows split-unit cold reconstruction remained disabled on the control lane at both restart boundaries:
  - `startup_authority_split_task_list_units_ms = 0.0`
  - `startup_authority_split_task_read_units_ms = 0.0`

## Not Proven

- This run does **not** prove a full runtime-only 50k pass across all storage paths.
- This run does **not** clear `guarded_segmented`.
- This run does **not** justify promoting `guarded_segmented` to the reference lane.
- This run does **not** erase the need for separate segmented-lane storage/restart validation.

## Bounded Challenger Statement

The challenger lane, `guarded_segmented`, did not clear the runtime-only 50k gate.

Observed evidence supporting that bounded statement:

- at the 20k stage-end event, retained bytes were already materially elevated:
  - current bytes `17,887,512`
  - retained total bytes `583,595,098`
  - retained obsolete bytes `565,707,586`
- the 20k restart-open on the challenger lane was materially outside the control-lane envelope:
  - `open_total_ms = 43,805.386`
  - `load_startup_state_ms = 37,903.555`
- post-20k runtime rows remained materially degraded:
  - sample step `020879` event ms `2883.300`
  - sample step `020879` persist ms `1093.460`
  - sample step `020879` retained total / obsolete bytes `609,426,258 / 590,751,378`

This is sufficient to state that `guarded_segmented` remains an evaluated challenger and is not cleared.

It is **not** used here to make a stronger claim about the final terminal status of that challenger run beyond "not accepted."

## Evidence Files

- `analysis/runtime_only_probe/analysis_runs/guarded-segmented-runtime-only-50k-validation-20260313-002/preflight.json`
- `analysis/runtime_only_probe/analysis_runs/guarded-segmented-runtime-only-50k-validation-20260313-002/split_control/compile_samples.jsonl`
- `analysis/runtime_only_probe/analysis_runs/guarded-segmented-runtime-only-50k-validation-20260313-002/split_control/daemon_stdio.log`
- `analysis/runtime_only_probe/analysis_runs/guarded-segmented-runtime-only-50k-validation-20260313-002/split_control/summary.json`
- `analysis/runtime_only_probe/tmp_root_analysis/guarded-segmented-runtime-only-50k-validation-20260313-002/split_control/logs/runtime.log.jsonl`
- `analysis/runtime_only_probe/analysis_runs/guarded-segmented-runtime-only-50k-validation-20260313-002/guarded_segmented/daemon_stdio.log`
- `analysis/runtime_only_probe/tmp_root_analysis/guarded-segmented-runtime-only-50k-validation-20260313-002/guarded_segmented/logs/runtime.log.jsonl`

## Evidence-Gate Wording Proposal

Recommended conservative wording:

> The runtime-only 50k evidence gate is cleared for the current reference lane, `split_control`, on run `guarded-segmented-runtime-only-50k-validation-20260313-002`. The exact control-lane checkpoints `stage_end_020000`, `restart_020000`, `stage_end_050000`, and `restart_050000` all returned `http_status: 200`. Control-lane restart remained healthy, split-unit cold reconstruction stayed disabled at restart, and retained bytes remained `0 / 0`. This is not a claim that all runtime storage paths passed 50k. The challenger lane, `guarded_segmented`, did not clear and remains outside the acceptable storage/restart envelope from the observed evidence in this run.

## Relationship To Earlier Notes

This note supersedes the earlier no-go interpretation in:

- `analysis/runtime_only_probe/20260312_runtime_only_50k_validation_note.md`

It does so narrowly:

- the earlier note remains historically accurate for the prior run it described
- this note records the current proving run for the control/reference lane
- no broader all-lanes pass should be inferred from that supersession