safal207 · safal207 · Sep 6, 2026 · Sep 6, 2026
diff --git a/.github/workflows/pages.yml b/.github/workflows/pages.yml
@@ -9,6 +9,7 @@ on:
       - 'issues/001-age-of-agents/articles/**'
       - 'analytics/**'
       - 'scripts/build-publication.mjs'
+      - 'scripts/build-article016.mjs'
       - 'scripts/build-issue-web-articles.mjs'
       - 'scripts/build-seo-site.mjs'
       - 'scripts/enhance-search-indexing.mjs'

diff --git a/.github/workflows/seo.yml b/.github/workflows/seo.yml
@@ -12,6 +12,7 @@ on:
       - 'protocols/**'
       - '.github/ISSUE_TEMPLATE/market-workflow.yml'
       - 'scripts/build-publication.mjs'
+      - 'scripts/build-article016.mjs'
       - 'scripts/build-issue-web-articles.mjs'
       - 'scripts/build-seo-site.mjs'
       - 'scripts/enhance-search-indexing.mjs'
@@ -42,6 +43,7 @@ on:
       - 'protocols/**'
       - '.github/ISSUE_TEMPLATE/market-workflow.yml'
       - 'scripts/build-publication.mjs'
+      - 'scripts/build-article016.mjs'
       - 'scripts/build-issue-web-articles.mjs'
       - 'scripts/build-seo-site.mjs'
       - 'scripts/enhance-search-indexing.mjs'

diff --git a/evidence/article016/corrections-history-summary.md b/evidence/article016/corrections-history-summary.md
@@ -0,0 +1,23 @@
+# RESONANCE Corrections + Version History Contract
+
+**Verdict:** PASS
+
+- Managed publications: **4**
+- Managed routes: **10**
+- Append-only history entries: **6**
+- Diff gate: **checked**
+- Changed managed routes in this diff: **0**
+- New history entries in this diff: **1**
+
+## Release invariants
+
+- Registered published routes must exist and end at the version declared by the registry.
+- Version transitions must form an unbroken chain from initial publication to the current version.
+- Existing history entries are immutable and append-only once v0.5 exists on the base branch.
+- A modified managed published page must be covered by a newly appended history entry in the same change.
+- Every history entry records change type, claim impact, reason, affected versions and inspectable evidence URLs.
+- Non-publication updates require RESONANCE GitHub evidence.
+
+## Evidence boundary
+
+Passing this contract proves publication-version bookkeeping, append-only history integrity and evidence linkage for registered routes. It does **not** prove that the correction itself is editorially or factually sufficient, and it makes no claim about unregistered pages.
diff --git a/evidence/article016/feed-summary.md b/evidence/article016/feed-summary.md
@@ -0,0 +1,23 @@
+# RESONANCE RSS / Subscribe Contract
+
+**Verdict:** PASS
+**Mode:** local
+
+| Language | Feed | Items | Newest item | Verdict |
+|---|---|---:|---|---|
+| en | feed.xml | 30 | Sun, 06 Sep 2026 12:00:00 GMT | PASS |
+| ru | feed.ru.xml | 7 | Sun, 06 Sep 2026 12:00:00 GMT | PASS |
+| zh-CN | feed.zh.xml | 4 | Sun, 06 Sep 2026 12:00:00 GMT | PASS |
+
+## Invariants
+
+- RSS 2.0 + Atom self-link identity is stable for EN / RU / zh-CN feeds.
+- Items are derived from published HTML with explicit publication dates, sorted newest-first and capped at 50.
+- `lastBuildDate` equals the newest item date, so identical publication state produces identical feed bytes.
+- Utility pages are excluded; item language must match its locale route.
+- Local builds require one language-appropriate RSS autodiscovery link on every HTML page.
+- Live mode re-fetches public feeds, the Subscribe page and representative recent item URLs.
+
+## Evidence boundary
+
+Passing proves feed structure, deterministic ordering/identity, local route linkage and live reachability for the checked surfaces. It does not prove subscriber delivery by a particular RSS reader, email delivery, readership or engagement.
diff --git a/evidence/article016/native-replay.json b/evidence/article016/native-replay.json
diff --git a/evidence/article016/plan.md b/evidence/article016/plan.md
@@ -0,0 +1,58 @@
+# Article 016 — bounded reproduction plan
+
+Recorded before execution on 2026-09-06.
+
+## CausalSpine
+
+- spineId: article016-native-seed-replay
+- exactSubject: repository-owned CML approval-lineage demo and CGQA payment-recovery seed traces only
+- subjectHash / source pins: CML `7a3a98c60a402d394e4d286da956823898454b8a`; CGQA `f861b934d77e64fd35f768e2a33bb4a00963bc19`
+- scale: editorial reproduction; parent contract: Article 016 source verification
+- purpose: determine whether existing checks distinguish the named positive and negative seed traces
+- expected outcome: six expected classifications match; native diagnostic output is preserved
+- forbidden inference: seed classifications establish an LLM hallucination rate, factual truth, runtime enforcement, source authenticity, production correctness, or exactly-once effects
+- dependencies: pinned clean source checkouts, Python, PyYAML for CML
+- typed time: synthetic record timestamps / ordered trace sequence; run date is observation metadata
+- past/present/future: published examples → local replay → bounded article evidence
+- symptom and refactor point: no newly reported defect; no production refactor
+- mutation authority: no source changes to CML or CGQA; generated editorial evidence only
+- authorizationRef: user request of 2026-09-06 to create a journal article about reducing hallucinations through our repositories
+- stop condition: preserve results and report mismatches; do not repair an oracle to match expectations
+- remaining debt: real model experiment, external replication, source completeness and live receiver observations
+
+## Cases fixed before execution
+
+| Input | Expected native outcome |
+|---|---|
+| CML make_invalid_trace | failed audit; policy and human-approval lineage findings |
+| CML make_valid_trace | passed audit |
+| CGQA pass_committed_stop.json | pass |
+| CGQA pass_failed_retry_same_identity.json | pass |
+| CGQA fail_retry_before_reconcile.json | fail; APR-001_UNRESOLVED_AMBIGUITY_FINANCIAL_ACTION |
+| CGQA fail_changed_idempotency_after_failed_reconcile.json | fail; APR-004_IDEMPOTENCY_CHANGED_ON_RETRY |
+
+The runner calls the existing evaluators, preserves full input and output, hashes source files, and compares two serial evaluations. It does not reimplement the evaluators. No model API, payment provider, external tool action, benchmark training, or live system is involved.
+
+## Transition field
+
+spineRef: article016-native-seed-replay. CGQA state dimensions are the authorized logical operation, concrete attempt, idempotency key, unresolved ambiguity set and recorded reconciliation outcome. Exercised paths are authorize → submit → ambiguous → committed → stop; authorize → submit → ambiguous → failed → retry; and the two existing negative paths (retry before resolution; changed retry key). CML records follow explicit parent and approval ancestors. The runner preserves the native results and full input paths; it does not add per-node state instrumentation.
+
+The dangerous paths are present in supplied negative traces and detected after evaluation. This is observation of invalid traces, not runtime prevention. Real clocks, TTL, numeric value limits, storage, concurrency, crash recovery, provider semantics and external effect counts are outside these seeds.
+
+## Capability scope
+
+| Capability | Planned scope |
+|---|---|
+| Exact subject / artifact gate; preregistered plan; orientation | RUN: pins, clean checkouts, plan, named inputs |
+| Native mapping / adapter review | RUN: direct native evaluator calls; no provider adapter |
+| Safety; authorization; replay / idempotency; causal ancestry; negative control | RUN within the six existing seeds |
+| Deterministic replay; trace integrity; evidence readiness | RUN: repeat outputs, preserve inputs, source hashes and native output |
+| Durable evidence reopen / integrity | RUN: parse result and hash files before Git handoff |
+| Verification debt; planning; meaning trajectory; watchpoints | RUN: retain explicit scope and debt; no new watchpoint promoted |
+| Liveness / reachability; transition geometry; stateful / property search | NOT_RUN: no exhaustive search or independent operation-order experiment |
+| Financial conservation; real temporal lifecycle; crash / recovery | NOT_APPLICABLE to a fixed synthetic trace-classification claim |
+| Independent witness; temporal / external replication | NOT_RUN: local author-side replay only |
+| Counterexample minimization; root-cause collapse; native regression; forward remediation | NOT_APPLICABLE: no new defect or repair claimed |
+| Metamorphic / round-trip verification | NOT_RUN: same-input replay only |
+
+Learning decision: NO_PROMOTION. This reproduces existing examples; it does not change upstream rules or establish a new generalized lesson.
diff --git a/evidence/article016/publication-validation.md b/evidence/article016/publication-validation.md
@@ -0,0 +1,18 @@
+# Article 016 publication validation
+
+Local observation: 2026-09-06. Base: e312826e4d1056b3354e737c230165d51523c696.
+
+- Complete publication build: PASS, 123 HTML pages, 16 published features registered.
+- Search indexing enhancement and contract: PASS, 108 crawlable articles.
+- SEO contract: PASS, 123 pages.
+- RSS/Subscribe: PASS, all three locale feeds include the new publication date.
+- Privacy analytics contract: PASS in disabled mode; deployment configuration is a separate runtime input.
+- Local links: PASS, 1591 references checked.
+- Translation structure and destinations: PASS, Article 016 has 12 outline tokens, 3 action links and 8 source links in each locale.
+- Corrections history: PASS with --base origin/main; existing entries preserved, one publication entry appended.
+- JavaScript syntax and Git whitespace checks: PASS.
+- Native replay: six expected classifications matched; two evaluations identical. See native-replay.json.
+
+The first corrections command used an unsupported --base-ref flag and reported diff not requested. The recorded final corrections check uses --base and reports diff checked.
+
+These are local static checks. They do not establish visual browser QA, independent translation review, CI success, live publication, model factuality, runtime enforcement or external replication. Remote CI and deployment status must be verified separately.
diff --git a/evidence/article016/translation-parity-summary.md b/evidence/article016/translation-parity-summary.md
@@ -0,0 +1,23 @@
+# RESONANCE Translation Parity Contract
+
+**Verdict:** PASS
+
+| Managed group | Profile | Outline tokens EN/RU/ZH | CTA links EN/RU/ZH | Source links EN/RU/ZH | Verdict |
+|---|---|---:|---:|---:|---|
+| article016 | article-semantic | 12/12/12 | 3/3/3 | 8/8/8 | PASS |
+| article015 | article-semantic | 10/10/10 | 1/1/1 | 2/2/2 | PASS |
+| homepage | locale-shell | shell | n/a | n/a | PASS |
+| article004 | article-semantic | 21/21/21 | 2/2/2 | 5/5/5 | PASS |
+| article005 | article-semantic | 23/23/23 | 2/2/2 | 4/4/4 | PASS |
+
+## Release invariants
+
+- EN / RU / zh-CN managed siblings must all exist and declare the correct `html lang`.
+- Canonical and reciprocal `hreflang` mappings must point to the same managed triplet.
+- The visible language switcher must contain EN / RU / zh-CN and mark only the current locale with `aria-current="page"`.
+- Article siblings must preserve the same shell controls/rail count and the same major semantic outline: heading levels plus evidence, trajectory, market-question, distribution and source blocks.
+- Article identity, critical IDs, CTA destinations, primary/evidence source URLs and runtime script set must not drift across translations.
+
+## Evidence boundary
+
+Passing this contract proves structural/destination parity for the managed triplets. It does **not** prove that translated prose is linguistically excellent, complete in nuance, or semantically equivalent sentence by sentence.
diff --git a/issues/001-age-of-agents/README.md b/issues/001-age-of-agents/README.md
@@ -39,6 +39,7 @@ Issue 001 maps this transition.
 13. [`Evidence Must Bind the Transition — почему AI-системе недостаточно доказательства, если оно не связано с конкретным переходом`](articles/13-evidence-must-bind-the-transition.md) — **published 2026-08-15** · [web edition](https://safal207.github.io/RESONANCE/evidence-must-bind-the-transition.ru.html)
 14. [`When the Feature Request Becomes Infrastructure — что происходит, когда проблема исчезает, а исследовательская граница сдвигается дальше`](articles/14-when-the-feature-request-becomes-infrastructure.md) — **published 2026-08-17**
 15. [`Стакан изменился. Разрешение осталось. — повторная проверка допуска в RESONANCE Verify`](articles/15-when-order-book-depth-changes.md) — **published 2026-09-05** · [EN](https://safal207.github.io/RESONANCE/when-order-book-depth-changes.html) · [RU](https://safal207.github.io/RESONANCE/when-order-book-depth-changes.ru.html) · [中文](https://safal207.github.io/RESONANCE/when-order-book-depth-changes.zh.html)
+16. [`Как уменьшить галлюцинации: от ответа к проверяемому результату`](articles/16-how-repositories-constrain-hallucinations.md) — **published 2026-09-06** · [EN](https://safal207.github.io/RESONANCE/how-repositories-constrain-hallucinations.html) · [RU](https://safal207.github.io/RESONANCE/how-repositories-constrain-hallucinations.ru.html) · [中文](https://safal207.github.io/RESONANCE/how-repositories-constrain-hallucinations.zh.html)
 
 The fourth feature extends the trust question beyond pre-action authorization: a consequential outcome needs its own observer identity, vantage and evidence so decision provenance and outcome provenance remain separately inspectable.