CR-README-03 Pass 1: hierarchy only — trim, demote, filter, de-correlate - #261
Merged
Merged
Conversation
Public port of private PR #347 (merged 2026-09-19, merge commit 2959ce98). Both files are byte-identical to the merged private/master. SCOPED DELIBERATELY TO TWO FILES. private/master currently diverges from public by 196 files, including the whole DAG-2026 stack (PR-012b gate vocabulary + trace schema v3, PR-012c hygiene, PR-012d activation fail-visible) and several workflow files. None of that is ready for public release and none of it rides along here: this commit carries README.md and README_PYPI.md and nothing else. Implements the Research Team README Update Brief (2026-09-19) under the strategy-review §20 constraint: reorder truthful content, add no new claims. - §2.3 Correlated judges — accuracy fix, not cosmetics. "How it works" step 4 said "Multiple agents debate" unqualified. Agents sharing a model family, a prompt template or a source document are correlated, so their agreement is not independent confirmation. Now stated inline: debate is a reasoning mechanism, not a proof of truth. README twin of the CR-014 correlation-discount amendment. - §2.4 Regulated deployments trimmed 33 -> 15 lines. Every ADR-MARKETING-001 locked marker survives (aligned badge, the literal "Articles 6, 9, 12, 13, 14, 15, 25, 50" scope string, both NOT-claims); every link kept. The `graq compliance` CLI block moves to docs/compliance/eu-ai-act/, where the complete nine-article set already lives. No engineering deleted. - §2.5 Token economics -> "What you actually get back". Durable ROI leads; token cost follows as supporting evidence. The -88% year-two figure is dropped; the case-study link carries the full math. - §2.6 Recent releases filtered to 4 user-facing entries, 0.84.0 leading. The EU AI Act layer entry moves to CHANGELOG.md. - §3.1 PyPI: model/IDE compatibility lifted above the fold; the downstream section de-duplicated to the config example + `graq init`. HELD FOR PASS 2 (all ADD claims, which §20 forbids mixing with a hierarchy change): the model-dependent vs model-independent paragraph, the trusted-state chain, and the PyPI enterprise-trust paragraph. §2.7 (Evidence & benchmarks, Research roadmap) is blocked entirely — all seven roadmap docs are absent and the brief forbids linking before the file exists. VERIFICATION ON THIS PUBLIC TREE - Both files hash-identical to merged private/master. - README lock gate + eu_ai_act_docs_present: 62 passed. - Every graq command machine-diffed against `graq --help`: 0 invalid. - Forbidden-word scan: 0 violations in both files. - Stale counts (76+/74/14/Azure OpenAI): 0. - git diff --name-only: exactly README.md + README_PYPI.md. NO VERSION BUMP. pyproject.toml, __version__.py and CHANGELOG.md untouched. PyPI still serves 0.84.0 and its page is frozen at that version's metadata; this copy reaches PyPI at the next release, not on merge. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
🛡️ GraQle PR Guardian💥 Blast Radius: 2 modules affected
Total blast radius: 2 🏛️ Governance Verdict
|
| Metric | Value |
|---|---|
| Blast Radius | 2 |
| Files Analyzed | 2 |
| Blocked | 0 |
| SHACL Violations | 0 |
| Verdict | WARN |
🔬 Powered by GraQle PR Guardian v0.1.0 · Scan completed 2026-09-19T10:16:06.421653+00:00
harishquantamix
approved these changes
Sep 19, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Pass 1 of 2 — hierarchy only, no new claims
Public port of private PR #347 (merged 2026-09-19, merge commit
2959ce98). Both files are byte-identical to the mergedprivate/master.Implements the Research Team README Update Brief (2026-09-19) under the strategy-review §20 constraint: reorder truthful content, add no new claims.
private/mastercurrently diverges from public by 196 files, including the entire DAG-2026 stack (PR-012b gate vocabulary + trace schema v3, PR-012c hygiene, PR-012d activation fail-visible) and several workflow files. None of that is ready for public release and none of it rides along here.That is the whole change.
What changed
§2.3 Correlated judges — an accuracy fix, not cosmetics
"How it works" step 4 said "Multiple agents debate" with no qualification. Agents sharing a model family, a prompt template or a source document are correlated, so their agreement is not independent confirmation. Now stated inline:
README twin of the CR-014 correlation-discount amendment.
§2.4 Regulated deployments: 33 → 15 lines
Every ADR-MARKETING-001 locked marker survives — the aligned badge, the literal
Articles 6, 9, 12, 13, 14, 15, 25, 50scope string, both NOT-claims. Every link kept. Thegraq complianceCLI block moves todocs/compliance/eu-ai-act/, where the complete nine-article set already lives. No engineering deleted.§2.5 Token economics → "What you actually get back"
Durable ROI leads: less context reconstruction, fewer architectural mistakes, faster change validation, retained team knowledge, easier model switching, safer autonomy. Token cost follows as supporting evidence. The −88% year-two figure is dropped; the case-study link carries the full math.
§2.6 Recent releases filtered
Four user-facing entries, 0.84.0 leading. The EU AI Act layer entry moves to
CHANGELOG.md.§3.1 PyPI: compatibility above the fold
"Will it work with my stack" is now answerable on the first screen; the downstream section is de-duplicated to the config example +
graq init.Held for Pass 2
The model-dependent vs model-independent paragraph, the trusted-state chain, and the PyPI enterprise-trust paragraph — all add claims, which §20 forbids mixing with a hierarchy change.
§2.7 (Evidence & benchmarks, Research roadmap) is blocked entirely: all seven roadmap docs are absent, and the brief forbids linking before the file exists.
Verification on this public tree
private/mastereu_ai_act_docs_presentgraqcommand vsgraq --helpgit diff --name-onlyNo version bump
pyproject.toml,__version__.pyandCHANGELOG.mdare untouched. PyPI still serves 0.84.0, and that page is frozen at its version's metadata — this copy reaches PyPI at the next release, not on merge.🤖 Generated with Claude Code