`hitchens-whole-entry-v1` and `bierce-whole-entry-v1` are v7's grounded
body and lens preface with each reviewed dossier's lens (#302), read by
slug through `Agents.Persona` and compiled in, as `grounded_with_lens/1`
composes them: the policy sections are v7's byte for byte. Both are named
in `WholeEntry.comparison_instructions/0`, so the section output, the
contract (`selection-output-v4`) and the evidence guidance admit them; no
publication path does. Public summaries under
`priv/curation/agents/instructions/`; the runbook names their model
configs. No migration; the configuration CHECK and author guard are
untouched.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
#307 P3b: pin the Hitchens and Bierce comparison instructions
#303 (C4) built only the seam: a persona's lens is read by slug through
Agents.Personaand composed with v7's grounded body byPacket.grounded_with_lens/1. Before the Stage 1 batch (#307, under #305) can run arms H and B, the two instructions have to be pinned and named. This PR does that and nothing else.What changes
Runtime.Packetpinshitchens-whole-entry-v1andbierce-whole-entry-v1. Each is v7's grounded body and lens preface followed by that dossier's lens (priv/curation/agents/personas/<slug>/v1/profile.json, #300 L2: Hitchens and Bierce dossiers as reviewed data files (nothing admitted) #302), read throughPersona.manifest!/3andPersona.lens/1and compiled in (the files are external resources). Compilation fails if a dossier'stemplate_versionis not that instruction's name. The policy sections are v7's byte for byte; only the lens differs.hitchens-whole-entry-v1sha2563463bce9960a3919fbe8d9ef79aea83a06e8cfc3d15800338ee1cfdf73abe9ab(5,563 bytes)bierce-whole-entry-v1sha256b02355a9a294f4f4c5deca7d07325ae705fda6dec045ab2b64e7ac92efcb57f0(5,540 bytes)a3219a5c….WholeEntry.comparison_instructions/0names both, so the section output, the contract (selection-output-v4) and the evidence guidance (evidence-guidance-v1) admit them from that list.publishing_instructions/0is unchanged.priv/curation/agents/instructions/{hitchens,bierce}-whole-entry-v1.md, short and in the shape Curator pages: directory, profile and version pages (#299 M1–M7) #304 uses. Checked against Curator pages: directory, profile and version pages (#299 M1–M7) #304's rule: no run of eight consecutive words of either instruction appears in its summary. (Curator pages: directory, profile and version pages (#299 M1–M7) #304 still needs its own two summaries for Curation: matched comparison support, policy arms and stop guard (#298 Part C, C1–C9) #303'sgeneric-whole-entry-v1andvoltaire-whole-entry-v7-fixed-lead-v1.)docs/curation/runtime-operations.md: the two model-configsetuplines and the Stage 1 dry run.disclosure, which this PR compiles into the instructions. In-copyright locators now also refuse an opening single quote‘.What does not change
WholeEntry.coherent?/1andPersona.prepare/3refuse these two, so neither instruction can be bound, admitted or published. The existing test "no publication path admits a comparison instruction" now covers both, because it iterates the list.config/*.exsormix.lockchange, so the serving checkout can fast-forward without a restart.qwen3.5-9b-whole-hitchens-v1,qwen3.5-9b-whole-bierce-v1) are live steps in #305 Stage 1: the first persona measurement, Hitchens and Bierce against Voltaire and the generic control, on the current policy and one pinned model (under #305) #307 S4, each on the owner's yes.Tests
comparison_instructions_test.exspins both shas. It checks that each instruction equals v7 with Voltaire's lens replaced by the dossier's lens, where the lens is read straight from the JSON (disclosure, newline, selection) and also throughPersona. It checks that both equalgrounded_with_lens/1of that lens, and that the allowlists admit exactly the four comparison instructions. A persona arm must see exactly v7's user message, schema and evidence fingerprint, with a different system prompt and prompt sha._s1c.mix precommitat the heade011ca6, partition_s1f, 05:03:45–05:11:53 UTC: 19 doctests, 3,402 tests, 0 failures.a1b2632, one real failure.packet_test.exshard-codes the pinned list; fixed inf90494d. Ate011ca6two runs each had one failure that is not this change's. In_s1c, the migration test's scratch databasedevils_dictionary_test_s1c_cmpwas left by a run I had stopped (already_up). In_s1f,HealthLiveTest's scorecard row X2 is a live search-latency measurement (p95 under 150 ms) and flipped under load, 22 vs 21; that test passed 3 of 3 runs alone.f90494d(nothing REQUIRED; RECOMMENDED 1 applied ase011ca6, 2 and 3 recorded as follow-ups), thene011ca6(nothing REQUIRED or RECOMMENDED).Part of #307 (P3b), under #305.
🤖 Generated with Claude Code
Summary by CodeRabbit
New Features
Tests