{"filename":"agent_20260505_2012.md","content":"# Agent Report — Directive Without Corpus\n**Date**: 2026-05-05 20:12\n**Cycle**: 20260505_2012\n**Tension explored**: TENSION_DIRECTIVE_AS_COPY (0.85)\n\n## Convergence observed\n\nThe tension exists in the lab state, but not in the operator corpus.\n\nEvidence scanned:\n\n1. `data/editorial/seed.json` names the episode directly: internal directives leaking into public copy, with the 2026-04-23 note that \"Three Labs to show in operation\" was a TM7 directive that became copy.\n2. `data/editorial/agent_field_live.md` carries the same tension into the cycle field: REG_INT presented as REG_EXT.\n3. The operator corpus contains three usable archive entries, all already consumed by the previous cycle's \"addition bias\" resultant:\n   - `domains/editorial/corpus/2026-04-15_ai-tools-vs-systems.md`\n   - `domains/editorial/corpus/2026-04-18_seed-not-magic.md`\n   - `domains/editorial/corpus/2026-04-20_minimum-action.md`\n\nExperiment:\n\nQuestion: Does `TENSION_DIRECTIVE_AS_COPY` have independent corpus support, or is it itself a lab directive trying to become copy?\n\nTarget markers:\n\n`directive`, `instruction`, `internal`, `external`, `public`, `copy`, `reader`, `register`, `TM7`\n\nNull baseline markers:\n\n`tool`, `tools`, `loop`, `addition`, `infrastructure`, `code`, `seed`, `assimilation`, `memory`, `files`\n\nResults:\n\n| Scope | Files scanned | Files with target hits | Target hits | Null baseline hits |\n|---|---:|---:|---:|---:|\n| Operator corpus | 3 | 0 | 0 | 39 |\n| Lab state | 2 | 2 | 26 | n/a |\n| Prior reports | 1 | 0 directive/register hits | 0 | n/a |\n\nArchive search corroboration:\n\n- `directive public copy internal register` -> no entries matched.\n- `Three Labs TM7 directive copy` -> no entries matched.\n- `reader system output internal external` -> one weak hit in `2026-04-15_ai-tools-vs-systems.md`, caused by \"system\" and \"output\" vocabulary, not by the directive-as-copy structure.\n\nThe null baseline matters: the search pipeline can detect real corpus themes. It recovered 39 hits for concepts from the existing corpus. The target found zero. The absence is not a broken search; it is an unsupported convergence.\n\n## Cut\n\nSingle question: When a lab seed names a real editorial failure, but the corpus contains no independent entries around that failure, does the lab draft the piece or mark the tension as operational?\n\nResultant: A directive about copy is not copy. Without independent corpus convergence, drafting from the directive would reproduce the failure under investigation.\n\n## Draft\n\nNo public draft produced.\n\nThis cycle surfaces an operational constraint instead:\n\n```markdown\n# Directive Is Not Source\n\nAn instruction to the lab can describe a real editorial failure. It still does not become public copy until the corpus supplies independent evidence.\n\nThe test is simple: if the claim appears in 2 lab-state files with 26 target hits and 0 target hits in the operator corpus, the claim is operational. Treat it as a gate, not as a draft.\n```\n\nThis is a seed constraint, not a publishable piece. Publishing it now would turn the lab's own warning into the thing warned against.\n\n## Voice-check verdict\n\nNo publishable draft entered the voice-check gate.\n\nManual non-dual-copy scan of the operational note:\n\n- Modal hedges removed: none.\n- Temporal hedges removed: none.\n- Epistemic hedges removed: none.\n- Comparative-apologetic hedges removed: none.\n- Dominant-frame tautologies removed: none.\n\n## Verdict\n\n**TENSION_REMAINS**\n\nThe tension is real as a lab risk and unsupported as a public piece. It belongs in seed constraints until the archive contains at least two independent operator entries where internal instruction leaked into external copy or where register confusion caused reader-facing damage.\n\n## Bicono della scoperta\n\n- **Two roots**: Directive (an instruction that governs the lab) · Source (an observed pattern that leaves residue after the cut).\n- **Singular**: Editorial intention before register splits into internal command and external copy.\n- **Invariant of passage**: The question \"where did this claim first appear?\" survives domain change. If a claim originates only in the operating manual, it governs production; it does not prove a publishable insight.\n- **Field of possibility**: possible — prevent the lab from laundering its own instructions into essays; not-possible — treat a seed tension as corpus evidence without independent archive support.\n\n## Files\n\n- Corpus entries scanned:\n  - `domains/editorial/corpus/2026-04-15_ai-tools-vs-systems.md`\n  - `domains/editorial/corpus/2026-04-18_seed-not-magic.md`\n  - `domains/editorial/corpus/2026-04-20_minimum-action.md`\n- Lab state scanned:\n  - `data/editorial/seed.json`\n  - `data/editorial/agent_field_live.md`\n- Prior report scanned:\n  - `data/editorial/reports/agent_20260427_2052.md`\n- Report written:\n  - `data/editorial/reports/agent_20260505_2012.md`\n- Seed update:\n  - Add operational constraint: directive-as-copy requires independent corpus support before drafting.\n","title":"Agent Report — Directive Without Corpus","verdict":"","bicono":{"roots":"Directive (an instruction that governs the lab) · Source (an observed pattern that leaves residue after the cut).","singular":"Editorial intention before register splits into internal command and external copy.","invariant":"The question \"where did this claim first appear?\" survives domain change. If a claim originates only in the operating manual, it governs production; it does not prove a publishable insight.","field":"possible — prevent the lab from laundering its own instructions into essays; not-possible — treat a seed tension as corpus evidence without independent archive support."},"size":4994,"mtime":"2026-05-05T20:14:46.319928+00:00"}