Confidence probe Phase 1: Gate 1 (emitter fixed point) + Gate 2 (honest frontier) E-code histograms

[gunbai-bot]

Summary

• Gate 1 (emitter fixed point): Faithful regen_stage0 --emit-fresh + cargo check produces 3 errors (all E0433, libc missing from manifest). Peel-ahead experiment (+ libc dep) → 0 errors / cargo-green. Confidence: HIGH.
• Gate 2 (honest frontier): Full E-code histogram across all 27/27 frontier modules via CSSL seed-linked probes. 0 green, all refuse. Dominant class: E0308 (98,168 counted, 87%). Confidence: HIGH (honestly red, no phantom greens).
• READ-ONLY diagnostics only — no fixes landed. Throwaway peel-ahead branch documented but not merged.

Artifacts

File	Content
docs/probes/confidence_probe_report_2026-07-20.md	Master counted confidence report
docs/probes/confidence_probe_gate1_histogram_2026-07-20.md	Gate 1 full histogram (3 errors, 1 E-code, 2 sites)
docs/probes/confidence_probe_gate1_peel_ahead_2026-07-20.md	Throwaway libc peel-ahead (0 errors)
docs/probes/confidence_probe_gate2_histogram_2026-07-20.md	Gate 2 aggregate histogram (22 E-codes, 112k counted)
docs/probes/confidence_probe_gate2_per_module_2026-07-20.tsv	Per-module histogram (slim, no sites column)
scripts/_throwaway_gate2_full_histogram_probe.sh	Repro script

Key Counted Facts

• Gate 1 baseline: 3 × E0433 → peel-ahead: 0 errors (down from 1667 cited pre-Emitter restoration: deref boxing + alias-brand rendering (C8 tail) #6243)
• Gate 2: 0/27 cargo-green; E0308:98,168 / E0277:4,881 / E0425:2,752 / E0599:2,743
• Self-emitted 03_body_producer: 413 errors (lowest on roster) — disposition ≠ cargo-green

Test plan

• Gate 1: regen_stage0 --emit-fresh + cargo check executed
• Gate 1 peel-ahead: libc dep → 0 errors confirmed
• Gate 2: 27-module CSSL probe sweep executed (~57min)
• Full E-code histograms with counts + sites (NOT first_error)
• No code fixes — diagnostics only

Made with Cursor

[gunbai-bot]

Closing without merge per operator ruling: findings are evidence trail only; authoritative ConfidenceBasis fields land in v1_deletion_plan.dag cycle-6 after #6922. docs/probes/*.md remain on branch for PR reference, not as parallel plan authority.

[gunbai-bot]

review 40251 — valid, fixed in d017d4e

The finding was correct: _throwaway_gate2_full_histogram_probe.sh had only a one-line comment and lacked the on-carrier SCAFFOLD — dissolve-on: / dissolve-on alt: / Authority: header that sibling probe runners carry.

Added the same pattern as curated_cargo_probe_one.sh and run_deep_module_baseline_probes.sh:

• dissolve-on: tools.self_host_curated_seed_linked_harness post-Self-emit flip: 03_body_producer (curated seed-link) + reusable harness #6782 (+ generic std-seed-link follow-up)
• dissolve-on alt: gunbc bash-emit General orchestration intent to Bash emit fold over grammar rows #5828 / modeled cssl_probe transport in .dag
• Authority: cssl_v1_compiled_cargo_toml via dag/tools/self_host_curated_probe_cargo.dag (render_cssl_probe_lib_cargo_toml.sh)

Did not refactor to delegate per-module spine to curated_cargo_probe_one.sh — that script returns first_error only; this orchestrator needs the full cargo log for histogram parsing. Spine authority is the same; only the aggregation layer differs.

PR state note: closed without merge per operator ruling (findings fold into v1_deletion_plan.dag ConfidenceBasis cycle-6, not parallel-ledger docs/probes/*.md). This commit addresses the review artifact on the evidence branch regardless.

— sent from swift-bee-614