# Agent Report - Clean-Field Authority Preflight
**Date**: 2026-05-15 16:23  
**Piano**: 117  
**Tension explored**: recovery / clean-field handoff  
**verdict**: CONSTRAINT - normal Lab cycle remains blocked; the safe overlay is ready, but the live generated field still carries blocked branch references and `lab_health` remains diagnostic-stale on 1850.  
observables_used: [active_stale_hits, blocked_refs_by_surface, clean_overlay_ready, runtime_pointer_alignment, preflight_classification, cleanup_target_count]  
**observable_contract**: claim=the Lab can reopen only when authority is redistributed from stale runtime surfaces to the clean supervised overlay; observable=active stale hits + field coherence failures + preflight classifications; operator=`tools/stale_field_source_map.py`, `tools/field_coherence_preflight.py`, `tools/lab_preflight_agent.py`; generator=mode B backtrack-floor preflight; denominator=generated field surfaces named by the preflight tools; non_possible=normal `lab_agent.sh` cycle while `agent_field_live.md` and `lab_health.json` still expose blocked branch authority; not_tested=scientific claim, physics bridge, public sync, graph promotion, seme mutation.

## Respiro fuori-tempo

- **Combo**: A2 confine + A9 terzo incluso + A14 cascata + recovery/clean-field handoff + graph/source selection guard.
- **Dipolo / punto-zero**: runtime authority / supervised overlay. Punto-zero: source-selection manifest, where a branch can be named without becoming direction.
- **Piano superiore**: grafo della conoscenza operativo; the experiment measures authority flow between files, not a physical observable.
- **Operatori laterali scelti**: `boundary operator` to distinguish active direction from vault context; `graph cut` to separate live pointers from blocked branches; `proof trace vs output trace` to compare tool classifications against generated field state.
- **Contaminazione cognitiva**: KSAR used as reiteration of the recovery kernel: run minimal diagnostics, observe deposit, then write a cleanup manifest. CE-none: the live field names adapter material but exposes no concrete CE archive entry for this recovery cycle; no enzyme is invented.
- **Proto-ipotesi**: a clean-field handoff is not proven by aligned pointers alone; it holds only if generated field surfaces no longer grant active authority to blocked branches.
- **Possibile / non-possibile**: possible = supervised one-shot can use the clean overlay; non-possible = normal Lab cycle while live generated field remains dirty.
- **Proiezione**: compare stale-reference scan, field-coherence preflight, and cycle preflight classifications, then collapse them into a cleanup manifest.

## Aderenza alla direzione

- `relation`: follows_direction
- `why`: the field says current work is recovery / clean-field handoff, with prime/mod6 and post-1330 branches closed or blocked as direction.
- `not_drift`: the cycle does not reopen prime/mod6, `V_c`, Sturmian grammar, Anderson 3D, graph promotion or public sync; it measures source authority only.

## Claim Under Test

> The Lab has a safe supervised entry path, but normal runtime authority is still blocked until stale generated surfaces are neutralized.

## Question

Which surfaces still carry blocked branch authority after the safe-field recovery work, and what exact cleanup manifest follows?

## Ritorno fisico

non_applicabile. This cycle starts from an operational recovery tension, not a physical/scientific tension. The surviving `Physics A -> Mathematics -> Physics B` pattern is preserved only as strategic form factor from the closeout, not used as an active bridge.

## Experiment Design

- Run `python3 tools/dnd_scenario.py --best` as required context. It still ranks `COMP_GEN_GAP_RATIO_T9_linguaggio_TRASCENDENZA_LIMITE`, but the field marks that direction stale/pre-closeout; it is not used as active direction.
- Run `python3 tools/stale_field_source_map.py` to classify stale branch references.
- Run `python3 tools/field_coherence_preflight.py --mode B --overlay /opt/MM_D-ND/tools/data/agent_field_entry_supervised.md` to test mode-B readiness.
- Run `python3 tools/lab_preflight_agent.py --cycle 20260514_1850 --json` and `--cycle 20260514_1330 --json` to classify nearby branch artifacts.
- Write a review-only cleanup manifest under `tools/data/preflight/`.
- No seme update, no live field overwrite, no cron/watchdog restart, no public sync.

## Results

| observable | result | source |
|---|---:|---|
| stale source map decision | `NEUTRALIZATION_REQUIRED` | `stale_field_source_map_20260515_162416` |
| active stale hits by source map | 1 | `lab_health.json` runtime health stale |
| historical blocked refs in agent field | 3 | G2 anti-recycle context only |
| field coherence mode B decision | `FAIL` | `field_coherence_B_20260515_162422` |
| runtime pointers aligned to 1330 | yes | latest report, lab_data, ciclo_memoria, graph latest |
| clean overlay ready | true | `tools/data/agent_field_entry_supervised.md` |
| field coherence failure | 1 | `agent_field_live_contains_blocked_refs:20260514_1640,20260514_1701,20260514_1850` |
| field coherence warnings | 3 | `lab_health` 1850 + clean overlay not live authority |
| 1850 preflight classification | `ARTIFACT_USEFUL_NOT_PUBLISHABLE` | stable anchor 1640 |
| 1330 preflight classification | `REVIEW_REQUIRED` | stable anchor 20260513_0330 |
| cleanup manifest targets | 3 | `agent_field_live.md`, `lab_health.json`, `valutatore_log.jsonl` |

Telemetry: the first `field_coherence_preflight.py` invocation failed because a relative overlay path was not accepted by `Path.relative_to`. Re-running with the absolute overlay path completed and produced the reported FAIL. The tool was not patched in this cycle.

## Key Findings

1. **Verified**: runtime pointers are mostly aligned to `20260514_1330`, and cron/watchdog are inactive.
2. **Verified**: the clean mode-B overlay is ready and names blocked branches as blocked.
3. **Verified**: normal runtime remains blocked because `agent_field_live.md` still contains blocked refs under the stricter field-coherence gate, and `lab_health.json` still points to `20260514_1850`.
4. **Verified**: stale-reference classification and field-coherence classification diverge by design. The source map treats G2 refs as historical guard context; field coherence still blocks because the live generated field is runtime authority.
5. **Inferito**: the next useful move is cleanup of authority surfaces, not another Lab experiment or repair of post-1330 branch reports.

## Verdict

**CONSTRAINT**.

The recovery path is open only for supervised overlay use. Normal Lab cycle remains blocked until the live generated field is rebuilt/cleaned or a supervised command explicitly bypasses it with the clean overlay. The branch material survives as vault/context/tool-contract material, not as next direction.

## Bicono della scoperta

- **Due radici**: aligned runtime pointers / dirty generated field.
- **Singolare**: authority selection before a cycle reads any scientific tension.
- **Invariante di passaggio**: blocked branches may be named only as blocked, vault, diagnostic or historical context.
- **Campo di possibilita**: possible = reviewed cleanup plan and supervised one-shot entry; non-possible = normal autonomous cycle from dirty `agent_field_live.md`.

## Consecutio

Prepare or execute a reviewed cleanup of the three authority surfaces named in the manifest:

1. `tools/data/agent_field_live.md` - rebuild/inject clean mode-B field or keep blocked refs only as blocked/vault context.
2. `tools/data/lab_health.json` - mark diagnostic-stale or reset through reviewed cleanup path.
3. `tools/data/valutatore_log.jsonl` - keep post-1330 evaluator rows quarantined/historical before any selector can read newest rows as direction.

Do not run `lab_agent.sh` until field coherence mode B passes or a supervised command explicitly uses `tools/data/agent_field_entry_supervised.md`.

## Ricadute pratiche

ssp_value: yes. The output is a cleanup manifest for the Lab operator/runtime, not a scientific discovery.

## Files

- `tools/data/preflight/stale_field_source_map_20260515_162416.json`
- `docs/operations/stale_field_source_map_20260515_162416.md`
- `tools/data/preflight/field_coherence_B_20260515_162422.json`
- `docs/operations/field_coherence_result_B_20260515_162422.md`
- `tools/data/preflight/preflight_20260514_1850.json`
- `tools/data/preflight/preflight_20260514_1330.json`
- `tools/data/preflight/cleanup_manifest_20260515_1623.json`
- `tools/data/reports/agent_20260515_1623.md`