Split evidence-grounding row and move summary table to top
Splits the summary table's combined 'Candidate finding validation /
evidence discipline' row into two independent rows matching the
existing §3 (candidate finding validation) and §6 (evidence grounding)
detailed sections, renumbering the table to 12 rows. The new Evidence
grounding row is populated strictly from §6's existing evidence (code
evidence attribution, typed authoritative/informational provenance,
CodeRabbit's 'receipts' claim) - no new research.
Also relocates the Review-engine summary table (and its interpretation
note) to immediately after the opening research-snapshot paragraph, so
a reader reaches the comparison before the longer scope/framing,
label-definition, and pipeline sections. Table content is otherwise
unchanged; no other section was modified.
56115d2
Add review-engine summary table to AI-Code-Review-Landscape
Adds an 11-dimension executive-summary table right after the pipeline
diagram and before the detailed dimension-by-dimension sections, so a
reader can scan the main methodological differences in ~20-30s. Reuses
the citations/evidence already established in the detailed sections
below (no new research); every cell links back to its numbered section.
Uses a consistent vocabulary (explicit contract / documented / vendor
claim or benchmark / not publicly specified) instead of checkmarks,
since the evidence differs in kind, and adds an interpretation note
clarifying that 'not publicly specified' is a documentation gap, not a
capability claim.
979b9b9
Reframe AI-Code-Review-Landscape around review-engine methodology
Rewrite the comparison to ask how engines reason about a change
(review-target correctness, context reasoning, candidate-finding
validation, root-cause consolidation, finding placement, evidence
grounding, severity semantics, decision derivation, re-review state,
uncertainty discipline, behavioral evaluation) instead of comparing
product feature breadth. Excludes IDE UX, pricing, SaaS integrations,
autofix-as-a-feature, and similar product-capability axes as explicitly
out of scope for an engine-quality comparison.
Incorporates the merged review-base-policy invariant (issue #134) and
the current candidate-finding-validation, root-cause-consolidation,
finding-placement, contextual-evidence, and finding-confidence models.
Re-verifies competitor claims against current public documentation for
Cursor Bugbot, GitHub Copilot code review, CodeRabbit, Qodo Merge, and
Greptile, marking undocumented dimensions 'Not publicly specified'
rather than inferring internals.
7551c2a
Document review-base policy compliance (#134) (#460)
Add a "Review-base policy compliance" section to How-the-Review-Model-Works,
GitHub-PR-Review-Guide, and Local-Code-Review-Guide describing the merged
#134 shared invariant: repository-relative, fail-closed, HEAD never
substituted, stacked-PR-aware, one blocking P0 before implementation
findings. Links to the canonical shared/policies/review-base-policy.md
rather than restating it.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
69ebaad
Document the finding-placement mental model (#388)
Adds an "Evidence location, symptom location, and fix/action location"
section to How-the-Review-Model-Works.md, right after candidate-finding
validation and before the existing shared-cause consolidation section
(renamed "Where a validated finding is anchored across several sites"
to distinguish single-finding placement from multi-site consolidation).
Covers the evidence/symptom/finding distinction and its shorthand,
causal ownership, contract ownership, locality preservation during
context expansion, semantic honesty over false precision, and that
GitHub-inline availability never redefines semantic ownership, with
three worked examples (cause vs. symptom, the precedent trap, the
nearest-line trap). Links the canonical contract in
shared/templates/finding.md's "Deriving the fix/action location"
section rather than restating it.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
52da208
Document the two-tier sentinel/comprehensive scheduled benchmark lanes (#431)
Update Benchmark-Testing.md's scheduled-execution description from the
single nightly full-corpus loop to the two independently-scheduled,
independently-baselined lanes: sentinel (4 canonical cases, every 3
days) and comprehensive (every benchmark-case/v2 fixture, weekly).
Co-authored-by: amirbena <amir.benassayag@gmail.com>
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
f456880
Remove redundant All pages list from Home
The flat All pages link list duplicated the new Wiki Page Map table (same 16 pages plus Home) and risked drifting out of sync with it. The Wiki Page Map is now the single list of all pages; other Home sections are unchanged.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
b681be3
Add a common Wiki Page Map to every page for navigation
Appends an identical, compact "Wiki Page Map" table (Page / What it
covers) to the end of all 22 existing Wiki pages, so a reader landing
on any single page can see every other page and its purpose without
returning to Home first.
No _Sidebar.md or _Footer.md exists in this wiki, so there is no
native shared-rendering mechanism that would place identical content
inside every page's own body; the table is duplicated intentionally
(kept small, one heading, one table) rather than introducing a
partial or page-specific navigation aid.
No substantive content, technical claims, or existing links were
changed — this is a pure addition at the end of each page.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
595e596
Document the new benchmark testing (Top-K selection/gating) pipeline
Adds a Benchmark-Testing page explaining epic #329's PR-time
taxonomy/index (#333), deterministic Top-K selector (#334),
shadow-validation (#335), nightly drift loop (#338/#339), and the
retirement of the legacy benchmark_ci_classifier.py CI check (#420).
Links it from Home and the Repository Documentation Map.
21d5e79
Document multiple invocation phrasings for runtime-validation, same execution condition
d20b0fe
Point to Validation section/provenance from Getting Started and both Skill guides
d26baf9
Add FAQ entry: where to check runtime-validation execution/provenance
68a19d9
Synchronize Wiki with root-cause consolidation, verdict-consistency, and confidence model
Closes the three bounded gaps identified by the Wiki-vs-implementation
research (Findings 1-3):
- How-the-Review-Model-Works.md: explains that a validated finding is
anchored at its causal/root-defect location, that shared-cause
manifestations are consolidated with an affected-locations list, and
that uncertain/thematic-only cases stay separate. Links
shared/policies/root-cause-consolidation.md.
- Severity-and-Decision-Model.md: explains the read-only
verdict-consistency comparator that re-verifies the about-to-render/
publish signal against the mechanically-derived decision and withholds
on mismatch. Links shared/policies/verdict-consistency.md.
- Feature-GitHub-Publication-and-Authorization.md: one sentence noting
ACTIVE publication's pre-submit event is covered by that same check.
- Feature-Runtime-Validation-Evidence.md: expands the single-sentence
confidence rollup into a table of the 5 closed values, clarifies it
reflects both runtime-validation state and context provenance, and
restates that confidence never changes finding identity, severity, or
decision. Links docs/finding-confidence/finding-confidence-model.md.
No new pages, no unrelated prose changes.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
674b669
Document candidate-finding validation mental model (#384)
Adds a "From an observation to a candidate to a finding" section to
How-the-Review-Model-Works.md covering the observation-first gate,
semantic-role validation, evidence/contract grounding (including
non-Jira technically-grounded blocking findings), causal/regression
proof discipline, the disconfirmation pass, and the finding-validity vs.
blocking-justification-validity separation, with four worked examples.
Links to the canonical design record in docs/candidate-finding-validation/
rather than restating it. Also registers that doc in
Repository-Documentation-Map.md's design-records row.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
79e514f
Document natural-language trusted-host-execution authorization (#369)
Adds the closed phrasing vocabulary, examples, and explicit-denial
behavior to Feature-Runtime-Validation-Evidence, alongside the
existing structured allow_trusted_host_execution documentation.
8cfa29b
Document trusted-host execution for runtime validation (#367)
Mirrors the docs/features/runtime-validation.md update: the sandboxed
default is unchanged, but a user can now explicitly, out-of-band, and
per-invocation authorize a bounded trusted-host execution backend when
sandbox isolation is unavailable, with no sandbox guarantees and
provenance recorded in the Validation evidence.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
b262027
Introduce structured Wiki feature-page pattern (#312)
Adds a reusable, 8-section feature-page template; a Feature-Pages index
mirroring docs/features/README.md; the initial batch of 5 feature pages
(runtime validation, requirement coverage, delta re-review, GitHub
publication/authorization, human-style output); the feature lifecycle
rule; and cross-links from Home, Repository-Documentation-Map, and both
Skill guide pages. Every page links back to its canonical repository
source and states the same never-source-of-truth boundary as the rest of
this Wiki.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
026d1ee
Sync Wiki with the implemented Adaptive Specialist Depth architecture (#268)
Adds a Specialist-Depth Review page documenting the final Tier 3
architecture (base reasoning -> materially implicated concern ->
evidence-driven specialist-depth decision -> 0..N deepening
capabilities -> one unified review) and the five capabilities that
actually landed (Security, Distributed Systems, Database/Migration,
Performance, Dependency/Supply-Chain), grounded in
shared/policies/specialist-depth.md and review-scope.md. Wires it into
Home, How the Review Model Works, Architecture Overview, the
Repository Documentation Map, Review Context and Existing Evidence
(explicit-focus additive-only), the Local Code Review guide
(capability provenance field), the AI Code Review Landscape
positioning section, and the FAQ.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
6bfcaab
docs: establish code review project wiki
Add the initial GitHub Wiki: 14 explanatory/navigational pages covering
onboarding, Skill selection, the review model, architecture, per-Skill
guides, review context/evidence, severity/decision model, packaging, a
dated AI code-review landscape analysis, the development workflow, an
FAQ, and a repository documentation ownership map. All content links
out to canonical repository files rather than duplicating policy text.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
c9a835e