From d9e16ba6ba58fb8b3e171f4e689239c22fe1ddd1 Mon Sep 17 00:00:00 2001 From: KimGLee <05_bolster_inkling@icloud.com> Date: Sat, 22 Aug 2026 14:19:15 +0800 Subject: [PATCH 1/2] Sync the Agent Systems Atlas profile example Replace the abbreviated example answers with the selected Atlas Profile snapshot at 15df10e, adapting only the example namespace and relative links. Record provenance and keep the package non-selectable.\n\nCloses #135 --- profiles/examples/agent-atlas/README.md | 35 +- .../examples/agent-atlas/escalation-policy.md | 41 +- .../examples/agent-atlas/expression-layer.md | 30 +- .../examples/agent-atlas/language-contract.md | 125 +++++- .../agent-atlas/metadata-contract.yaml | 55 ++- .../examples/agent-atlas/priority-rubric.md | 14 +- profiles/examples/agent-atlas/profile.md | 2 +- .../registries/audit-dimensions.md | 18 +- .../examples/agent-atlas/registries/roles.md | 10 +- .../scan-configs/interview-residuals.yaml | 31 +- .../agent-atlas/scope-and-architecture.md | 2 +- .../agent-atlas/structure-registry.yaml | 387 ++++++++++++++++-- .../agent-atlas/vocabulary-extensions.yaml | 15 + 13 files changed, 630 insertions(+), 135 deletions(-) diff --git a/profiles/examples/agent-atlas/README.md b/profiles/examples/agent-atlas/README.md index 545cf64..9d41646 100644 --- a/profiles/examples/agent-atlas/README.md +++ b/profiles/examples/agent-atlas/README.md @@ -2,17 +2,21 @@ ## What This Example Is -This directory is a non-normative, filled example of the Cambium profile interface for a real Chinese-first engineering knowledge corpus. It is not a template, default configuration, or adoption certificate. Because it lives under `profiles/examples/`, its manifest is intentionally not selectable in place. +This directory is a non-normative snapshot of the real, selected Agent Systems +Atlas Profile for a Chinese-first engineering knowledge corpus. It is not a +template, default configuration, or adoption certificate. Because it lives +under `profiles/examples/`, its manifest is intentionally not selectable in +place. The normative interface is [profiles/README.md](../../README.md), and cross-domain rules remain in the kernel. A real adopter copies [profiles/_template/](../../_template/profile.md) to `profiles//`, fills and validates that copy, and selects it through governance. -The example does not contain or synchronize an adopter corpus, runtime state, -or gate evidence. It remains a reference profile rather than an adoption -certificate or proof of corpus-wide acceptance. +The example does not contain the adopter corpus, runtime state, or gate +evidence. It remains a reference profile rather than an adoption certificate +or proof of corpus-wide acceptance. ## Directory Structure -The directory contains 14 published files: +The directory contains 17 published files: | Role | Count | Meaning | |---|---:|---| @@ -24,6 +28,27 @@ The directory contains 14 published files: The scan configuration is not another slot or standard. This example has no separate Interview standards directory, supplemental Profile Read Set, profile-owned executable, or plugin. +## Snapshot Provenance + +The 16 machine-active files in this directory were synchronized from the +selected Profile in Agent Systems Atlas at commit +`15df10eac89cafd381b145c48659c4a525f93f6d`. The source package was adopted +under Cambium Standards 3.12.0. + +Only location-preserving adaptations are allowed in this example copy: + +- the Profile root changes from `profiles/agent-atlas/` to + `profiles/examples/agent-atlas/`; +- relative links to the Profile interface and kernel execution-default + registry gain the additional `examples/` directory level; +- the manifest title and introduction identify the package as a non-selectable + example. + +No Atlas rule, enum, trigger, role, judgment requirement, corpus path, +structure declaration, scan predicate, or escalation threshold is abridged for +the example. The orientation README is Cambium-owned and is not part of the +selected Profile snapshot. + ## Atlas-specific Bindings K11 owns the universal expression-layer floor. This profile supplies only the Atlas-specific answers: what an Interview artifact is, how it binds to canonical knowledge, how bilingual answers and readiness work, and what becomes a residual-content candidate. diff --git a/profiles/examples/agent-atlas/escalation-policy.md b/profiles/examples/agent-atlas/escalation-policy.md index f2d73db..5cea31c 100644 --- a/profiles/examples/agent-atlas/escalation-policy.md +++ b/profiles/examples/agent-atlas/escalation-policy.md @@ -2,25 +2,25 @@ Interface: [Escalation Policy slot](../../README.md#escalation-policy-slot) -This corpus is built by long autonomous batch runs, so the executor reaches -situations no batch gate judges: the plan itself changes shape, a batch stops -moving, or an action cannot be undone. Each row below names one such -situation, what counts as it happening, and who decides. The kernel trigger — -explicit user authorization to modify the Standards or the selected profile — -is owned by +This corpus uses long autonomous batch runs, so the executor may reach +situations that require a person before work can continue: the plan changes +materially, a batch stops moving, or an action cannot be undone. The kernel +trigger for modifying the Standards or the selected Profile is owned by [[kernel/K13 Task Runtime and Execution Control/17 Escalation Policy|K13/17]] -and is deliberately not repeated here. - -The two magnitudes are this instance's numbers, recalculated at each -maintenance boundary or every twenty closed batches. `replan_magnitude` is -three times the largest change any ordinary replan has made to the Required -record count, floored at 5%, and falls back to 30% while fewer than ten -ordinary replans have run; an ordinary replan is one the user did not -pre-review item by item. `batch_age` is ten times the p99 batch duration over -a rolling twenty batches, clamped to between four and forty-eight hours, and -falls back to forty-eight hours while fewer than twenty batches have closed. -Both live in the Progress Ledger's run history, not in this file, because a -value recomputed from the ledger cannot be restated here without going stale. +and is not repeated here. + +Thresholds are derived from the Progress Ledger rather than copied into this +Profile. At each maintenance boundary, or after another twenty batches close: + +- `replan_magnitude` is three times the largest Required-record-count change + made by an ordinary replan, with a 5% floor. Until ten ordinary replans have + completed, the threshold is 30%. An ordinary replan is one the user did not + pre-review item by item. +- `batch_age` is ten times the rolling p99 duration of the latest twenty closed + batches, clamped to four through forty-eight hours. Until twenty batches + have closed, the threshold is forty-eight hours. + +The formulas are Profile policy; the current values remain runtime state. ## Escalation Triggers @@ -28,7 +28,6 @@ value recomputed from the ledger cannot be restated here without going stale. | Trigger ID | Condition that fires it | `machine-checkable` or `review-checkable` | Deciding Role ID reference | Resume condition | |---|---|---|---|---| -| `quota_overrun` | A priority-quota target is exceeded with no recorded exemption | `machine-checkable` | `stopper` | The user grants an exemption, or the quota is brought back within target | -| `replan_magnitude` | One replan changes the Required record count by more than the current `replan_magnitude` threshold | `machine-checkable` | `stopper` | The user approves the replanned scope | -| `batch_stalled` | An `open` or `merge-ready` batch has been in that state longer than the current `batch_age` threshold | `machine-checkable` | `stopper` | The user rules on whether to finish, split, or abandon the batch | +| `replan_magnitude` | One replan changes the Required record count by more than the ledger-derived `replan_magnitude` threshold. | `machine-checkable` | `stopper` | The user approves the replanned scope. | +| `batch_stalled` | An `open` or `merge-ready` batch remains in that state longer than the ledger-derived `batch_age` threshold. | `machine-checkable` | `stopper` | The user decides whether to finish, split, or abandon the batch. | | `irreversible_operation` | The next action physically deletes a page, or rewrites history by any means other than a registered Amendment | `review-checkable` | `stopper` | The user authorizes that specific operation | diff --git a/profiles/examples/agent-atlas/expression-layer.md b/profiles/examples/agent-atlas/expression-layer.md index 4eead4e..8fc9d97 100644 --- a/profiles/examples/agent-atlas/expression-layer.md +++ b/profiles/examples/agent-atlas/expression-layer.md @@ -28,13 +28,16 @@ Interface: [Expression Layer Entry slot](../../README.md#expression-layer-entry- | Artifact role/type | `roadmap` | | Reader-facing display label | Interview Roadmap | | Entry point (target-corpus relative) | `Interview Preparation/Roadmaps/` | -| Current pre-adoption target-corpus paths | `Agent Knowledge/Agent Interview Roadmap.md`; `Deep Learning Knowledge/DL Interview Roadmap.md`; `LLM Knowledge/LLM Interview Roadmap.md`; `Machine Learning Knowledge/ML Interview Roadmap.md` | | Profile-rule owner | `profiles/examples/agent-atlas/expression-layer.md#Interview Roadmap` | | Existing canonical dependency-map path | `Interview Preparation/Interview Overview.md` | | Testable regeneration/invalidation predicate | Re-evaluate an affected roadmap when its ordered topic set, prerequisite graph, or linked Interview Card inventory changes. | | Vocabulary Extensions readiness-field ID | `None` | | Routing/Gate Registry supplemental Gate ID | `None` | +## Terminology Extraction Extension + +For a proper noun, satisfying the profile's [[profiles/examples/agent-atlas/expression-layer#Card Granularity|independent-Card predicate]] is an additional reason to consider a standalone Term Note under [[kernel/K05 Terminology/01 Terminology Extraction#Extraction Criteria|K05/01 Terminology Extraction]]. The Card still does not own the term's definition, and this trigger does not bypass the kernel's Do Not Extract or terminology-acceptance rules. + ### Artifact: Interview Cheat Sheet | Property | Value | @@ -43,17 +46,12 @@ Interface: [Expression Layer Entry slot](../../README.md#expression-layer-entry- | Artifact role/type | `cheat-sheet` | | Reader-facing display label | Interview Cheat Sheet | | Entry point (target-corpus relative) | `Interview Preparation/Cheat Sheets/` | -| Current pre-adoption target-corpus paths | `Agent Knowledge/Agent Basics Cheat Sheet.md`; `LLM Knowledge/LLM Basics Cheat Sheet.md`; `Machine Learning Knowledge/Algorithms/ML Algorithms Checklist.md` | | Profile-rule owner | `profiles/examples/agent-atlas/expression-layer.md#Interview Cheat Sheet` | | Existing canonical dependency-map path | `Interview Preparation/Interview Overview.md` | | Testable regeneration/invalidation predicate | Re-evaluate an affected cheat sheet when a summarized canonical owner, linked card, or high-priority retrieval route changes. | | Vocabulary Extensions readiness-field ID | `None` | | Routing/Gate Registry supplemental Gate ID | `None` | -## Terminology Extraction Extension - -For a proper noun, satisfying the profile's [[profiles/examples/agent-atlas/expression-layer#Card Granularity|independent-Card predicate]] is an additional reason to consider a standalone Term Note under [[kernel/K05 Terminology/01 Terminology Extraction#Extraction Criteria|K05/01 Terminology Extraction]]. The Card still does not own the term's definition, and this trigger does not bypass the kernel's Do Not Extract or terminology-acceptance rules. - ## Interview Card Contract This is the stable profile-owned rule entry for Agent Systems Atlas Interview Cards. It owns the concrete Card behavior and organization described below. The cross-profile separation, evidence, status, linking, and migration floor remains owned by [[kernel/K11 Expression Layer Standard|K11 Expression Layer Standard]]; corpus placement remains owned by the profile's [[profiles/examples/agent-atlas/scope-and-architecture|Profile Scope]], and bilingual answer parity remains owned by its [[profiles/examples/agent-atlas/language-contract#Bilingual Answer Contract|Language Contract]]. @@ -192,16 +190,32 @@ A source note or public case cannot be presented as personal project evidence. Q ## Interview Roadmap -An Interview Roadmap orders preparation by prerequisites, role relevance, priority, and verification milestones. The canonical entry for new Roadmaps is `Interview Preparation/Roadmaps/`; the pre-adoption paths listed in the artifact registration above remain migration inputs until moved. Roadmaps link to both canonical knowledge and applicable Interview Cards. They may name a practice or review checkpoint, but they do not contain the complete answers owned by Cards. +An Interview Roadmap orders preparation by prerequisites, role relevance, +priority, and verification milestones. Its canonical entry is +`Interview Preparation/Roadmaps/`. Roadmaps link to both canonical knowledge +and applicable Interview Cards. They may name a practice or review checkpoint, +but they do not contain the complete answers owned by Cards. A roadmap checkbox reports the learner's progress on that roadmap only. It does not promote `authoring_status`, `interview_status`, evidence maturity, or corpus completion; the governing progress semantics are [[kernel/K11 Expression Layer/06 Sequence and Progress Semantics|K11/06 Sequence and Progress Semantics]]. ## Interview Cheat Sheet -An Interview Cheat Sheet is a compact retrieval aid whose canonical entry for new files is `Interview Preparation/Cheat Sheets/`; the pre-adoption paths listed in the artifact registration above remain migration inputs until moved. It may contain concise distinctions, decision cues, failure cues, formula reminders, and links to the canonical page and Interview Card. It does not own definitions, mechanisms, evidence, complete 30/90-second answers, or readiness decisions. +An Interview Cheat Sheet is a compact retrieval aid whose canonical entry is +`Interview Preparation/Cheat Sheets/`. It may contain concise distinctions, +decision cues, failure cues, formula reminders, and links to the canonical +page and Interview Card. It does not own definitions, mechanisms, evidence, +complete 30/90-second answers, or readiness decisions. When a Cheat Sheet conflicts with a canonical owner or an accepted Card, it is stale and must be corrected from those owners; it is never used as the authority for resolving the conflict. +`Interview Preparation/Cheat Sheets/ML Algorithms Cheat Sheet.md` derives from +the canonical `Machine Learning Knowledge/Algorithms/ML Algorithms Checklist.md` +and records that relation in its `canonical_bindings` frontmatter. The derived +page keeps only interview retrieval cues, key distinctions, and links to +canonical owners. When the canonical page's selection or acceptance criteria +change, the derived page enters revalidation; the derived page's continued +existence never proves synchronization. + ## Related - [[kernel/Read Sets/R05 Expression Layer Read Set|R05 Expression Layer Read Set]] diff --git a/profiles/examples/agent-atlas/language-contract.md b/profiles/examples/agent-atlas/language-contract.md index e24cb85..f6c7fd3 100644 --- a/profiles/examples/agent-atlas/language-contract.md +++ b/profiles/examples/agent-atlas/language-contract.md @@ -8,15 +8,45 @@ Interface: [Language Contract slot](../../README.md#language-contract-slot) |---|---| | Body language (language name or tag) | Simplified Chinese (`zh-Hans`) for explanations, mechanisms, causality, comparisons, limitations, and conclusions. | | Secondary language and canonical form on disagreement (`None — monolingual` allowed) | English preserves official identities, source wording, machine interfaces, and designated English Interview answers. When translations disagree, the exact external identity or source-language claim remains fixed and the Chinese explanation is corrected to the same bounded meaning. | -| General proper-name display (localized, original, or bilingual; include occurrence scope) | At first meaningful use, show `English identity（中文解释）`; later uses may keep the stable English identity or an unambiguous Chinese form. | +| General proper-name display (localized, original, or bilingual; include occurrence scope) | At first meaningful use, show `English identity（中文解释）`; later uses follow the Term Band Rule: identifier-band names keep the stable English identity, and all other terms continue in Chinese. | | Official external-name display around preserved identity | Preserve the organization's, product's, model's, protocol's, framework's, library's, or algorithm's official English name and explain its role in Chinese nearby. | | Machine-identifier display around the exact token | Preserve code identifiers, fields, enum values, commands, paths, formulas, and configuration literals exactly in code formatting; explain their semantics in adjacent Chinese prose. | +## Term Band Rule + +The default language of reader-facing body prose is Chinese. An English word +or phrase in body prose must belong to one of the following bands. A general +computing term with an established Chinese rendering belongs to neither band +and is written in Chinese. + +1. **Identifier band — preserve English.** Source, configuration, and + documented interface identities such as `Store`, `namespace`, + `checkpointer`, and `thread_id`, plus official product, organization, and + protocol names such as LangGraph, MCP, and PostgreSQL, retain their exact + spelling. Translating these strings would break the evidence path to their + source. Machine identifiers remain in code formatting. +2. **Term-anchor band — Chinese after one English anchor.** A domain term whose + Chinese rendering is unsettled, or whose Chinese form is insufficient for + source retrieval, appears at first meaningful use as + `中文译名（English term）`. Later uses on the same page use the Chinese form. + +Review applies this test to each English phrase independently. A compound +phrase is split by band: in `namespace authorization`, `namespace` remains an +identifier and `authorization` is rendered in Chinese. + ## Bilingual Answer Contract -For an Interview Card, the 30-second and 90-second answers are complete in both English and Chinese. Follow-up prompts use bilingual labels, and every follow-up intended for spoken delivery includes an English answer or a usable English answer skeleton. The two language versions preserve the same claim, limitation, uncertainty, and metric meaning. +For an Interview Card, the 30-second and 90-second answers are complete in +both English and Chinese. Follow-up prompts use bilingual labels, and every +follow-up intended for spoken delivery includes an English answer or a usable +English answer skeleton. The two language versions preserve the same claim, +limitation, uncertainty, and metric meaning. -This section owns only Interview Card answer parity. General body language, naming, display order, protected identifiers, and formatting remain owned by the other sections of this Language Contract. Interview Card structure and evidence requirements remain owned by [[profiles/examples/agent-atlas/expression-layer#Interview Card Contract|Expression Layer]]. +This section owns only Interview Card answer parity. General body language, +naming, display order, protected identifiers, and formatting remain owned by +the other sections of this Language Contract. Interview Card structure and +evidence requirements remain owned by +[[profiles/examples/agent-atlas/expression-layer#Interview Card Contract|Expression Layer]]. ## Canonical Naming @@ -32,11 +62,90 @@ This section owns only Interview Card answer parity. General body language, nami | Concern | Profile rule | |---|---| | Alias forms (included name/language forms) | Include the English full form, accepted acronym, common Chinese name, and a materially distinct established synonym; do not create aliases for spelling noise. | -| Headings and labels (display pattern) | Reader-facing structure uses Chinese or `English Title（中文解释）`. Exact external titles and machine identifiers may remain unchanged where translating them would alter identity. | -| Abbreviation first use (form and page/vault scope) | On each page, introduce an abbreviation as `English Full Name（中文解释，ABBR）`; later uses may use `ABBR`. | -| Reader-facing display order (ordered components) | Exact English identity, then full-width Chinese explanation, then the Chinese sentence; never reverse the identity as `中文（English）`. | +| Headings and labels (display pattern) | Markdown section headings are stable navigation labels. Pure English and `English Title（中文解释）` both conform; a bilingual heading must not invert the order as `中文（English）`. Table headers and other reader-facing explanatory labels stay Chinese-first under the Term Band Rule. The Section Role Display registration below separately owns `Sources` and `Related`. Changing an existing heading triggers the `ATLAS-LANG-MIG-001` anchor and inbound-link check. | +| Abbreviation first use (form and page/vault scope) | On each page, introduce an abbreviation as `English Full Name（中文解释，ABBR）`; later uses may use `ABBR` for identifier-band names and inside tables, labels, and code. Non-identifier terms in body prose continue in Chinese under the Term Band Rule. | +| Reader-facing display order (ordered components) | Identifier-band names use exact English identity first and a full-width Chinese explanation second. Term-anchor-band terms use `中文译名（English term）` at first use in body prose only; headings remain identity-first. | | File-name annotation boundary (allowed annotations or `None`) | None — do not append translations, statuses, bracketed explanations, or convenience labels to canonical file names. | +## Section Role Display + +This contract owns the reader-facing display forms of the kernel section roles +defined by K07/02 (`sources`) and K09/04 (`related`). The machine projection in +`metadata-contract.yaml` carries the same binding and never replaces this +prose owner. + +| Role | Display titles | Migration-period aliases | +|---|---|---| +| `sources` | `Sources（来源）`; `来源` | `Sources` | +| `related` | `Related（相关）`; `相关` | `Related`; `Related（相关内容）` | + +Aliases are accepted only for existing content during migration. New and +rewritten pages use a registered display title. One page carries at most one +sources-role section. + +## Content Form Review + +Classify each reviewed content block by its natural form. The deciding +question is whether a list is a closed contract or a compressed causal chain. + +1. **Contract enumeration.** Responsibility boundaries (`owns`, `excludes`, + goals, non-goals) and external-reference sets belong in the frontmatter + `boundary` block. `boundary-contract` validates that block, and + `render_boundary_projection.py` produces its reader view. Body prose may + refer to the projection but must not duplicate the enumeration. The author + chooses where the projection markers appear. +2. **Natural table, diagram, or code.** Comparison tables, sequence diagrams, + architecture diagrams, structure definitions, pseudocode, and formulas + keep their native form. Marker comments distinguish authored content from + generated projections. +3. **Compressed narrative.** Parallel bullets that carry causality, + mechanisms, or failure analysis are rewritten as connected prose. The + prose makes causal and trade-off relations explicit and is judged by + whether it explains the mechanism, not by enumeration completeness. +4. **Natural prose.** Existing prose is edited for clarity without changing + its form. + +A responsibility block belongs only to a page that asserts who owns a +concern. Overview, module-entry, and system-design pages commonly qualify; +source notes, cases, and Interview Cards ordinarily express evidence or +narrative scope instead. Concern slugs use a layer prefix and remain an open +kebab-case vocabulary until a later Profile revision explicitly closes the +set. Kernel projection labels remain in effect because this Profile declares +no `boundary_projection` override. + +An `excludes` owner may be a kernel or Profile page. A missing boundary on an +out-of-manifest owner remains an advisory gap; it does not excuse a qualifying +manifest page from carrying its own boundary. When a batch introduces its +first new boundary block, Batch Review records the slug design for integrator +review. + +Before `merge-ready`, Batch Review applies the four-way classification to +every manifest page and records an explicit result in that page's +content-correctness evidence. A conclusion that no migration is needed is a +valid result; an omitted result is not a completed review. Category 3 and 4 +edits also apply the Term Band Rule. The classification is not a separate +checkpoint artifact. Pages not reached by ordinary batches are handled by a +finishing batch before Queue exhaustion can be claimed. + +## Registered Term Displays + +The following sense-aware table is the current word-level display authority. +An unregistered term follows the Term Band Rule, and a provisional mapping +needed during a batch is recorded as a gap. Adding or changing a stable entry +is a governed Profile revision, not a batch-close side effect. A new entry +does not authorize a bulk rewrite; affected pages align when they next enter +scope unless an explicit migration is approved. + +| Term | Sense | Band | First display literal | Subsequent literal | Accepted variants | Canonical owner | +|---|---|---|---|---|---|---| +| `Harness` | Agent execution runtime shell | identifier | `Harness（执行外壳）` | `Harness` | `Agent Harness` | `Agent Knowledge/Harness/Agent Harness.md` | +| `span` | One timed unit in a distributed trace | term-anchor | `追踪跨度（span）` | `追踪跨度` | `span_id` | `Agent Knowledge/Harness/Tracing.md` | +| `trace` | Causal tree of spans as an object | term-anchor | `链路追踪（trace）` | `链路追踪` | `trace_id` | `Agent Knowledge/Harness/Tracing.md` | +| `trace` | Act of following causality | term-anchor | `追踪` | `追踪` | None | `Agent Knowledge/Harness/Tracing.md` | +| `artifact` | General produced and versioned output | term-anchor | `产物（artifact）` | `产物` | None | `Agent Knowledge/Harness/Artifact Lifecycle.md` | +| `artifact` | Build or release output | term-anchor | `构建制品（artifact）` | `构建制品` | None | None | +| `artifact` | Audit evidence object | term-anchor | `证据产物（artifact）` | `证据产物` | None | None | + ## Content Length Unit - Unit (`words` or `characters`): characters @@ -47,7 +156,7 @@ This section owns only Interview Card answer parity. General body language, nami | Kernel anti-pattern ID or profile extension ID | Scoped exception or added predicate | Predicate-owner path | |---|---|---| -| `ATLAS-LANG-001` | In reader-facing Atlas content, an English-only explanatory heading or table label, or bilingual display ordered as `中文（English）`, is a review failure unless the text is an exact external title, machine identifier, or inside a designated English Interview-answer block. | `profiles/examples/agent-atlas/language-contract.md#K10/04 Scoped Anti-pattern Extensions` | +| `ATLAS-LANG-001` | In reader-facing Atlas content, an English-only explanatory table label or figure caption, or a bilingual display inverted as `中文（English）`, is a review failure unless the text is an exact external title, a machine identifier, or part of a designated English Interview-answer block. Markdown section headings are exempt. In body prose, an English term that is neither an identifier-band name nor a first-use term-anchor gloss is also a review failure. | `profiles/examples/agent-atlas/language-contract.md#K10/04 Scoped Anti-pattern Extensions` | ## Additional Formatting Migration Invalidations @@ -55,4 +164,4 @@ This section owns only Interview Card answer parity. General body language, nami | K10/04 change-kind ID or profile trigger | Additional invalidated dimensions | Scoped exception | Rule-owner path | |---|---|---|---| -| `ATLAS-LANG-MIG-001`: add, remove, or reorder Chinese annotation in a heading; rename a canonical page, folder, or term | Heading-link resolution, canonical naming, aliases, and reader-facing language conformance for the changed target and all incoming references | Exact external and machine identities remain unchanged, but their surrounding links and explanations are still checked. | `profiles/examples/agent-atlas/language-contract.md#Additional Formatting Migration Invalidations` | +| `ATLAS-LANG-MIG-001`: add, remove, or reorder a Chinese annotation in a heading; rename a canonical page, folder, or term | Heading-link resolution, canonical naming, aliases, and reader-facing language conformance for the changed target and all incoming references | Exact external and machine identities remain unchanged, but their surrounding links and explanations are still checked. | `profiles/examples/agent-atlas/language-contract.md#Additional Formatting Migration Invalidations` | diff --git a/profiles/examples/agent-atlas/metadata-contract.yaml b/profiles/examples/agent-atlas/metadata-contract.yaml index e02c6b9..4caff9e 100644 --- a/profiles/examples/agent-atlas/metadata-contract.yaml +++ b/profiles/examples/agent-atlas/metadata-contract.yaml @@ -1,33 +1,29 @@ -# Agent Atlas metadata-contract slot: the configured branch. +# Agent Systems Atlas metadata-contract slot: the configured branch. # Interface: profiles/README.md#Metadata-Contract-Slot # Kernel defaults stay in kernel/K08 Metadata and Status/applicability-base.yaml # and relationship-base.yaml; only Atlas differences and extensions appear -# here. The two Interview fields are the profile's Expression Layer bindings -# (K08/07: `interview_card` needs a resolvable reciprocal binding once the -# page is mapped; readiness itself is granted by the Expression gate, never -# inferred). `canonical_bindings` is the expression-to-canonical relation and -# never enters `evidence_sources`. +# here. `interview_card` and `interview_status` +# are the Expression Layer bindings under K08/07 authority: a mapped-class +# value needs a resolvable reciprocal binding, and interview-ready is granted +# only by the registered expression gate. `canonical_bindings` is the +# expression-to-canonical relation and never enters `evidence_sources`. +# `source_class` is the L-SOURCES grouped-taxonomy page field of the +# Structure Registry; `case_class` is the L-CASES evidence-form class. The +# bilingual section-role titles project the Language Contract's +# display binding; the prose owner stays language-contract.md. schema_version: 1 applicability: state: configured -applicability_differences: - - field: last_verified - mode: conditional - condition: - all: - - field: volatility - in: - - fast - note: "Fast-decaying pages must carry their last external verification date; kernel default is optional." +applicability_differences: [] extension_fields: - field: interview_status mode: optional shape: nonempty-string - owner: "profiles/agent-atlas/expression-layer.md" + owner: "profiles/examples/agent-atlas/expression-layer.md" - field: interview_card mode: conditional shape: path @@ -37,7 +33,25 @@ extension_fields: in: - mapped - interview-ready - owner: "profiles/agent-atlas/expression-layer.md" + owner: "profiles/examples/agent-atlas/expression-layer.md" + - field: case_class + mode: conditional + shape: nonempty-string + condition: + all: + - field: type + in: + - case-study + owner: "profiles/examples/agent-atlas/structure-registry.yaml" + - field: source_class + mode: conditional + shape: nonempty-string + condition: + all: + - field: type + in: + - source-note + owner: "profiles/examples/agent-atlas/structure-registry.yaml" relationship_extensions: - field: canonical_bindings @@ -45,7 +59,7 @@ relationship_extensions: direction: expression-to-canonical target: page shape: list-of-paths - owner: "profiles/agent-atlas/expression-layer.md" + owner: "profiles/examples/agent-atlas/expression-layer.md" section_roles: - role: sources @@ -54,11 +68,12 @@ section_roles: - "来源" aliases: - "Sources" - owner: "profiles/agent-atlas/language-contract.md" + owner: "profiles/examples/agent-atlas/language-contract.md" - role: related titles: - "Related（相关）" - "相关" aliases: - "Related" - owner: "profiles/agent-atlas/language-contract.md" + - "Related（相关内容）" + owner: "profiles/examples/agent-atlas/language-contract.md" diff --git a/profiles/examples/agent-atlas/priority-rubric.md b/profiles/examples/agent-atlas/priority-rubric.md index 8819525..31aa354 100644 --- a/profiles/examples/agent-atlas/priority-rubric.md +++ b/profiles/examples/agent-atlas/priority-rubric.md @@ -11,4 +11,16 @@ Interface: [Priority Rubric slot](../../README.md#priority-rubric-slot) ## Priority Quota -- Registration: None +The standing quota targets this corpus holds its P0/P1 shares to. K00/07 owns +the quota model; this registration replaces the retired +`priority_quota.*` execution-default override rows (same values, now the +instrument K00/07 names for long-lived shares). A temporary excess is not +registered here -- it is a bounded contract policy exception via +`apply_contract_amendment.py`, and it dies with its task. + +- Registration: Configured + +| Class | Maximum corpus share | Rationale | +|---|---|---| +| `P0` | `17%` | The P0 predicate binds this class to sole canonical contracts for the end-to-end Agent/Harness control path and to Capability Matrix P0 rows. That path spans goal and plan, context and state, tool effects, permission and policy, completion and recovery, evaluation evidence, and production operation, each with its own canonical owner. The 17% ceiling accommodates that registered owner set rather than padding the class. | +| `P1` | `37%` | P1 carries direct prerequisites of P0 owners and the interview-preparation target topics registered in the Capability Matrix. The 37% ceiling accommodates that prerequisite-and-target band while reserving the remaining 46% for P2 content. | diff --git a/profiles/examples/agent-atlas/profile.md b/profiles/examples/agent-atlas/profile.md index f158b6d..8a0c762 100644 --- a/profiles/examples/agent-atlas/profile.md +++ b/profiles/examples/agent-atlas/profile.md @@ -20,8 +20,8 @@ Normative interface: [Profile interface](../../README.md). Example namespace rul - `Role Registry`: `registries/roles.md` - `Audit Dimension Registry`: `registries/audit-dimensions.md` - `Registered Scan Registry`: `registries/registered-scans.md` -- `Routing And Gate Registry`: `registries/routing-and-gates.md` - `Escalation Policy`: `escalation-policy.md` +- `Routing And Gate Registry`: `registries/routing-and-gates.md` ## Execution Default Overrides diff --git a/profiles/examples/agent-atlas/registries/audit-dimensions.md b/profiles/examples/agent-atlas/registries/audit-dimensions.md index 09f788d..6e04f79 100644 --- a/profiles/examples/agent-atlas/registries/audit-dimensions.md +++ b/profiles/examples/agent-atlas/registries/audit-dimensions.md @@ -19,7 +19,7 @@ Interface: [Audit Dimension Registry slot](../../../README.md#audit-dimension-re | `agent-atlas-interview-readiness-acceptance` | `interview` | `Single Note Review` | One Interview Card and its bound canonical topics presented for `interview-ready` promotion satisfy the registered readiness predicate. | `emits` | `profiles/examples/agent-atlas/registries/audit-dimensions.md#Interview Readiness Acceptance` | | `agent-atlas-profile-wide-interview-acceptance` | `interview` | `Specialized Audit` | The complete in-scope Agent Systems Atlas snapshot satisfies the registered profile-wide Interview-layer acceptance predicate. | `emits` | `profiles/examples/agent-atlas/registries/audit-dimensions.md#Profile-wide Interview Acceptance` | | `agent-atlas-interview-residual-disposition` | `coverage_and_integration` | `Batch Review` | The merged in-scope snapshot's Interview-answer residual candidates outside `Interview Preparation/` all have an accepted disposition. | `emits` | `profiles/examples/agent-atlas/registries/audit-dimensions.md#Residual-content Disposition` | -| `agent-atlas-content-form-classification` | `content_and_depth` | `Batch Review` | Every page on one batch's frozen manifest carries an explicit four-class content-form classification judged against the registered criteria. | `emits` | `profiles/examples/agent-atlas/registries/audit-dimensions.md#Content Form Classification` | +| `agent-atlas-content-form-classification` | `content_and_depth` | `Batch Review` | Every page on one batch's frozen manifest carries an explicit four-class content-form classification judged against the Content Form Review criteria. | `emits` | `profiles/examples/agent-atlas/language-contract.md#Content Form Review` | ## Interview Card Review @@ -67,19 +67,3 @@ The Agent Systems Atlas interview layer is profile-wide complete only when: - the Interview Overview reflects the actual artifact set without becoming a second answer owner. This profile-wide verdict consumes the applicable Card, link, residual-scan, source, and coverage evidence; it does not replace their individual judgments. -## Content Form Classification - -Judge one manifest page against the four-class content-form criteria: - -- class 1: contract enumerations (goals, non-goals, ownership assertions) sink - into the frontmatter `boundary` block and the body cites its projection; -- class 2: natural tables, diagrams, code, formulas, and closed query-style - enumerations keep their form; -- class 3: flattened causal or layered narratives — entries a reader must - understand in order — are rewritten as causal prose; -- class 4: natural prose is only smoothed, applying the Term Band Rule per - word. - -An explicit "judged natural record, no migration needed" verdict is a pass; -an absent verdict is not. The judgment covers every page of the frozen -manifest exactly once. diff --git a/profiles/examples/agent-atlas/registries/roles.md b/profiles/examples/agent-atlas/registries/roles.md index a8bab39..d2dde99 100644 --- a/profiles/examples/agent-atlas/registries/roles.md +++ b/profiles/examples/agent-atlas/registries/roles.md @@ -6,16 +6,16 @@ Interface: [Role Registry slot](../../../README.md#role-registry-slot) | Kernel role | Bound actor or system ID/name | |---|---| -| `proposer` | Model proposal | -| `gatekeeper` | Harness control | -| `executor` | External executor | +| `proposer` | Model-proposal actor | +| `gatekeeper` | Harness-control system | +| `executor` | External-executor actor | | `stopper` | Human authority | ## Knowledge Host | Kernel role | Binding | |---|---| -| `knowledge-host` | Obsidian vault | +| `knowledge-host` | Agent Systems Atlas Obsidian vault | | `knowledge-host UI` | Obsidian desktop UI | ## Metric Traceability Roles @@ -40,4 +40,4 @@ These bindings apply when Atlas reports evaluated metrics, whether the measureme | Role ID | Bound actor or system ID/name | Responsibility | |---|---|---| | `interview-reviewer` | Human authority or explicitly delegated qualified reviewer | Judge Interview Card acceptance and authorize promotion to `interview-ready`. | -| `content-reviewer` | Human authority or explicitly delegated qualified reviewer | Judge the per-page content-form classification a batch must complete before `merge-ready`. | +| `content-reviewer` | Human authority or explicitly delegated qualified reviewer | Judge the per-page four-class content-form classification a batch must complete before `merge-ready`. | diff --git a/profiles/examples/agent-atlas/scan-configs/interview-residuals.yaml b/profiles/examples/agent-atlas/scan-configs/interview-residuals.yaml index 8029985..6a231d1 100644 --- a/profiles/examples/agent-atlas/scan-configs/interview-residuals.yaml +++ b/profiles/examples/agent-atlas/scan-configs/interview-residuals.yaml @@ -23,11 +23,16 @@ frontmatter_match: # The generic matcher strips Markdown heading syntax only: numbering, # parentheticals, and bilingual suffixes are NOT removed. `Required Card -# Structure` in ../expression-layer.md mandates the compound -# `English（中文）` form, so each mandated form is registered verbatim below -# alongside the bare and CJK variants that legacy pages still carry. Dropping a -# mandated form here does not weaken the scan quietly: `mandated_headings` -# makes it a hard configuration error. +# Structure` in ../expression-layer.md mandates the compound `English（中文）` +# form, so each mandated form is registered verbatim below alongside the bare +# and CJK-only variants that legacy and partially migrated pages still carry. +# Dropping a mandated form here does not weaken the scan quietly: +# `mandated_headings` makes it a hard configuration error. +# +# Placement follows the same rule the bare forms already used: a heading that +# alone identifies an Interview Card goes in `any`; a heading that also occurs +# in ordinary knowledge pages goes in `combination`, where two distinct hits +# are needed. heading_match: any: - Interview Card @@ -81,11 +86,17 @@ heading_match: - 自测问题 minimum_distinct: 2 -# The section headings `Required Card Structure` in ../expression-layer.md -# mandates for every Interview Card, transcribed verbatim and in order. The -# tool proves every one of them is matchable above and that a page carrying -# exactly this set classifies as a candidate; it does not judge whether this -# list is the right one, which stays this profile's call. +# The Card sections that `Required Card Structure` in ../expression-layer.md +# mandates for every Interview Card, transcribed verbatim and in that order. +# The two indented rows under the 30-/90-second sections (`English（英文回答）` +# and `中文`) are bilingual sub-headings inside those sections rather than Card +# sections of their own, and are generic enough that registering them as +# matchers would fire on most bilingual pages; the bilingual parity they carry +# is owned by ../language-contract.md, not by this scan. +# +# The tool proves every entry below is matchable in heading_match and that a +# page carrying exactly this set classifies as a candidate; it does not judge +# whether this list is the right one, which stays this profile's call. mandated_headings: - Scope（范围） - Knowledge Prerequisites（知识前置） diff --git a/profiles/examples/agent-atlas/scope-and-architecture.md b/profiles/examples/agent-atlas/scope-and-architecture.md index 63649fd..d7c8324 100644 --- a/profiles/examples/agent-atlas/scope-and-architecture.md +++ b/profiles/examples/agent-atlas/scope-and-architecture.md @@ -71,7 +71,7 @@ Interview Preparation/ | `Shared Foundation Layer` | Layer ID | `L-FOUNDATION` | | `Production Systems Layer` | Layer ID | `L-PRODUCTION` | | `Cross-domain Concepts Layer` | Layer ID | `None` — fallback `L-FOUNDATION`; place the concept with the closest reusable mechanism owner. | -| `Expression Layer Predicate` | Predicate | The page has `type: interview-card`, `roadmap`, or `cheat-sheet` and either lives under `Interview Preparation/` or is one of the pre-adoption Roadmap/Cheat Sheet paths registered in `profiles/examples/agent-atlas/expression-layer.md`. | +| `Expression Layer Predicate` | Predicate | The page has `type: interview-card`, `roadmap`, or `cheat-sheet` and lives under `Interview Preparation/`. | | `Case Study Layer` | Layer ID | `L-CASES` | | `Source Note Layer` | Layer ID | `L-SOURCES` | | `Research Synthesis Layer` | Layer ID | `L-SYNTHESIS` | diff --git a/profiles/examples/agent-atlas/structure-registry.yaml b/profiles/examples/agent-atlas/structure-registry.yaml index adf3f3a..bfd1afa 100644 --- a/profiles/examples/agent-atlas/structure-registry.yaml +++ b/profiles/examples/agent-atlas/structure-registry.yaml @@ -1,12 +1,15 @@ -# Agent Atlas structure-registry slot: the configured branch. +# Agent Atlas Structure Registry — the selected profile's structural units +# (K01/05) and support layers (K01/06). # Interface: profiles/README.md#Structure-Registry-Slot -# Semantics owners: kernel/K01 Scope and Architecture/05 Structural Unit -# Interface.md and 06 Support Layer Structural Interfaces.md. Layer IDs come -# from this profile's Profile Scope; paths are the adopting vault's own and, -# as with corpus-planning.yaml, do not resolve inside the Cambium repository. -# This example registers a representative subset: two domains, one admitted -# module, and the four support layers. A real adoption registers every domain -# and every admitted module. +# The registry declares the current target state, and +# Tools/check_structure.py is the structure-registry gate that proves the +# adopting repository agrees with it. +# The eleven Global Map submodule entries satisfy the K01/05 admission test: +# each has a unique registered +# responsibility, a canonical entry, one parent, Coverage projectability, +# and independent maintenance history, so all eleven register as modules. +# No other first-level directory passes the test; unregistered directories +# stay ordinary groupings covered by their parent's entry. schema_version: 1 @@ -30,16 +33,91 @@ units: heading: "Prerequisite Map" coverage: mode: derived - generator: "Tools/render_coverage_projection.py" + generator: "Tools/render_structure_projection.py" inputs_owner: "Corpus Planning/capability_matrix.yaml" path: "Modeling Fundamentals/Modeling Fundamentals Overview.md" heading: "Coverage Reader View" quick_reference: mode: not-applicable - reason: "No general-audience quick reference exists; the interview cheat sheet is an expression artifact, not this role." + reason: "No general-audience quick reference is maintained; the interview cheat sheet is an expression artifact, not this role." expression: mode: standalone path: "Interview Preparation/Roadmaps/Modeling Fundamentals Roadmap.md" + - id: U-ML + kind: domain + parent: null + root: "Machine Learning Knowledge" + entry: + path: "Machine Learning Knowledge/Machine Learning Overview.md" + expected_type: overview + global_map_entry: E-ML + roles: + sequence: + mode: embedded + path: "Machine Learning Knowledge/Machine Learning Overview.md" + heading: "Prerequisite Map（前置依赖图）" + coverage: + mode: derived + generator: "Tools/render_structure_projection.py" + inputs_owner: "Corpus Planning/capability_matrix.yaml" + path: "Machine Learning Knowledge/Machine Learning Overview.md" + heading: "Coverage Reader View（Coverage 读者视图）" + quick_reference: + mode: standalone + path: "Machine Learning Knowledge/Algorithms/ML Algorithms Checklist.md" + expression: + mode: standalone + path: "Interview Preparation/Roadmaps/ML Interview Roadmap.md" + - id: U-DL + kind: domain + parent: null + root: "Deep Learning Knowledge" + entry: + path: "Deep Learning Knowledge/Deep Learning Overview.md" + expected_type: overview + global_map_entry: E-DL + roles: + sequence: + mode: embedded + path: "Deep Learning Knowledge/Deep Learning Overview.md" + heading: "Prerequisite Map（前置依赖图）" + coverage: + mode: derived + generator: "Tools/render_structure_projection.py" + inputs_owner: "Corpus Planning/capability_matrix.yaml" + path: "Deep Learning Knowledge/Deep Learning Overview.md" + heading: "Coverage Reader View（Coverage 读者视图）" + quick_reference: + mode: not-applicable + reason: "No general-audience quick reference is maintained for this domain." + expression: + mode: standalone + path: "Interview Preparation/Roadmaps/DL Interview Roadmap.md" + - id: U-LLM + kind: domain + parent: null + root: "LLM Knowledge" + entry: + path: "LLM Knowledge/LLM Overview.md" + expected_type: overview + global_map_entry: E-LLM + roles: + sequence: + mode: embedded + path: "LLM Knowledge/LLM Overview.md" + heading: "Prerequisite And Lifecycle Flow（前置与生命周期）" + coverage: + mode: derived + generator: "Tools/render_structure_projection.py" + inputs_owner: "Corpus Planning/capability_matrix.yaml" + path: "LLM Knowledge/LLM Overview.md" + heading: "Coverage Reader View（覆盖状态读者视图）" + quick_reference: + mode: not-applicable + reason: "No general-audience quick reference is maintained for this domain." + expression: + mode: standalone + path: "Interview Preparation/Roadmaps/LLM Interview Roadmap.md" - id: U-AGENT kind: domain parent: null @@ -52,19 +130,65 @@ units: sequence: mode: embedded path: "Agent Knowledge/Agent Overview.md" - heading: "Reader Route" + heading: "Reader Route（读者路线）" coverage: mode: derived - generator: "Tools/render_coverage_projection.py" + generator: "Tools/render_structure_projection.py" inputs_owner: "Corpus Planning/capability_matrix.yaml" path: "Agent Knowledge/Agent Overview.md" - heading: "Coverage Reader View" + heading: "Coverage Reader View（Coverage 读者视图）" quick_reference: mode: not-applicable - reason: "No general-audience quick reference is maintained for this domain." + reason: "No general-audience quick reference is maintained; the Agent Basics Cheat Sheet is an expression artifact." expression: mode: standalone path: "Interview Preparation/Roadmaps/Agent Interview Roadmap.md" + - id: U-AISE + kind: domain + parent: null + root: "AI Systems Engineering" + entry: + path: "AI Systems Engineering/AI Systems Engineering Overview.md" + expected_type: overview + global_map_entry: E-AI-SYSTEMS + roles: + sequence: + mode: embedded + path: "AI Systems Engineering/AI Systems Engineering Overview.md" + heading: "Study Order" + coverage: + mode: derived + generator: "Tools/render_structure_projection.py" + inputs_owner: "Corpus Planning/capability_matrix.yaml" + path: "AI Systems Engineering/AI Systems Engineering Overview.md" + heading: "Coverage Reader View" + quick_reference: + mode: not-applicable + reason: "No general-audience quick reference is maintained for this domain." + expression: + mode: not-applicable + reason: "The profile allows but does not require an Interview Roadmap per domain; none is maintained here." + - id: U-LLM-RAG + kind: module + parent: U-LLM + root: "LLM Knowledge/Retrieval" + entry: + path: "LLM Knowledge/Retrieval/RAG Pipeline.md" + expected_type: system-design + global_map_entry: E-RAG + roles: + sequence: + mode: not-applicable + reason: "The parent domain's lifecycle flow covers reading order." + coverage: + mode: not-applicable + reason: "Coverage stores this module as the finest owner; the parent domain's derived view presents it." + quick_reference: + mode: not-applicable + reason: "The parent entry routes here." + expression: + mode: not-applicable + reason: "Expression binds per page through interview_card, not per module." - id: U-AGENT-HARNESS kind: module parent: U-AGENT @@ -75,19 +199,206 @@ units: global_map_entry: E-HARNESS roles: sequence: - mode: embedded - path: "Agent Knowledge/Harness/Agent Harness.md" - heading: "Execution Path" + mode: not-applicable + reason: "The parent domain's System Spine covers execution order." coverage: - mode: derived - generator: "Tools/render_coverage_projection.py" - inputs_owner: "Corpus Planning/capability_matrix.yaml" + mode: not-applicable + reason: "Coverage stores this module as the finest owner; the parent domain's derived view presents it." quick_reference: mode: not-applicable - reason: "The parent domain entry covers routing to this module." + reason: "The parent entry routes here." expression: - mode: standalone - path: "Interview Preparation/Topic Cards/Agent/Agent Harness Interview Card.md" + mode: not-applicable + reason: "Expression binds per page through interview_card, not per module." + - id: U-AGENT-MEMORY + kind: module + parent: U-AGENT + root: "Agent Knowledge/Memory" + entry: + path: "Agent Knowledge/Memory/Memory Overview.md" + expected_type: roadmap + global_map_entry: E-MEMORY + roles: + sequence: + mode: not-applicable + reason: "The parent domain's System Spine covers execution order." + coverage: + mode: not-applicable + reason: "Coverage stores this module as the finest owner; the parent domain's derived view presents it." + quick_reference: + mode: not-applicable + reason: "The parent entry routes here." + expression: + mode: not-applicable + reason: "Expression binds per page through interview_card, not per module." + - id: U-AGENT-LONG-HORIZON + kind: module + parent: U-AGENT + root: "Agent Knowledge/Reliability" + entry: + path: "Agent Knowledge/Reliability/Long-Horizon Agent Reliability Overview.md" + expected_type: overview + global_map_entry: E-LONG-HORIZON + roles: + sequence: + mode: not-applicable + reason: "The parent domain's System Spine covers execution order." + coverage: + mode: not-applicable + reason: "Coverage stores this module as the finest owner; the parent domain's derived view presents it." + quick_reference: + mode: not-applicable + reason: "The parent entry routes here." + expression: + mode: not-applicable + reason: "Expression binds per page through interview_card, not per module." + - id: U-AISE-OPERATIONS + kind: module + parent: U-AISE + root: "AI Systems Engineering/Operations" + entry: + path: "AI Systems Engineering/Operations/Agent Production Operations Overview.md" + expected_type: overview + global_map_entry: E-OPERATIONS + roles: + sequence: + mode: not-applicable + reason: "The parent domain's Study Order covers reading order." + coverage: + mode: not-applicable + reason: "Coverage stores this module as the finest owner; the parent domain's derived view presents it." + quick_reference: + mode: not-applicable + reason: "The parent entry routes here." + expression: + mode: not-applicable + reason: "Expression binds per page through interview_card, not per module." + - id: U-AISE-CACHE + kind: module + parent: U-AISE + root: "AI Systems Engineering/Caching" + entry: + path: "AI Systems Engineering/Caching/Cache Management Overview.md" + expected_type: overview + global_map_entry: E-CACHE + roles: + sequence: + mode: not-applicable + reason: "The parent domain's Study Order covers reading order." + coverage: + mode: not-applicable + reason: "Coverage stores this module as the finest owner; the parent domain's derived view presents it." + quick_reference: + mode: not-applicable + reason: "The parent entry routes here." + expression: + mode: not-applicable + reason: "Expression binds per page through interview_card, not per module." + - id: U-AISE-RAG + kind: module + parent: U-AISE + root: "AI Systems Engineering/RAG" + entry: + path: "AI Systems Engineering/RAG/RAG Production System Overview.md" + expected_type: overview + global_map_entry: E-PRODUCTION-RAG + roles: + sequence: + mode: not-applicable + reason: "The parent domain's Study Order covers reading order." + coverage: + mode: not-applicable + reason: "Coverage stores this module as the finest owner; the parent domain's derived view presents it." + quick_reference: + mode: not-applicable + reason: "The parent entry routes here." + expression: + mode: not-applicable + reason: "Expression binds per page through interview_card, not per module." + - id: U-AISE-COST-CAPACITY + kind: module + parent: U-AISE + root: "AI Systems Engineering/Cost and Capacity" + entry: + path: "AI Systems Engineering/Cost and Capacity/AI Cost and Capacity Engineering Overview.md" + expected_type: overview + global_map_entry: E-COST-CAPACITY + roles: + sequence: + mode: not-applicable + reason: "The parent domain's Study Order covers reading order." + coverage: + mode: not-applicable + reason: "Coverage stores this module as the finest owner; the parent domain's derived view presents it." + quick_reference: + mode: not-applicable + reason: "The parent entry routes here." + expression: + mode: not-applicable + reason: "Expression binds per page through interview_card, not per module." + - id: U-AISE-DATA-FEEDBACK + kind: module + parent: U-AISE + root: "AI Systems Engineering/Data and Feedback" + entry: + path: "AI Systems Engineering/Data and Feedback/Agent Data Feedback and Learning System Overview.md" + expected_type: overview + global_map_entry: E-DATA-FEEDBACK + roles: + sequence: + mode: not-applicable + reason: "The parent domain's Study Order covers reading order." + coverage: + mode: not-applicable + reason: "Coverage stores this module as the finest owner; the parent domain's derived view presents it." + quick_reference: + mode: not-applicable + reason: "The parent entry routes here." + expression: + mode: not-applicable + reason: "Expression binds per page through interview_card, not per module." + - id: U-AISE-TOOLS-MCP + kind: module + parent: U-AISE + root: "AI Systems Engineering/Tools and MCP" + entry: + path: "AI Systems Engineering/Tools and MCP/Tool and MCP Trust Boundary Overview.md" + expected_type: system-design + global_map_entry: E-TOOLS-MCP + roles: + sequence: + mode: not-applicable + reason: "The parent domain's Study Order covers reading order." + coverage: + mode: not-applicable + reason: "Coverage stores this module as the finest owner; the parent domain's derived view presents it." + quick_reference: + mode: not-applicable + reason: "The parent entry routes here." + expression: + mode: not-applicable + reason: "Expression binds per page through interview_card, not per module." + - id: U-AISE-EVALUATION + kind: module + parent: U-AISE + root: "AI Systems Engineering/Evaluation" + entry: + path: "AI Systems Engineering/Evaluation/Evaluation Overview.md" + expected_type: overview + global_map_entry: E-EVALUATION + roles: + sequence: + mode: not-applicable + reason: "The parent domain's Study Order covers reading order." + coverage: + mode: not-applicable + reason: "Coverage stores this module as the finest owner; the parent domain's derived view presents it." + quick_reference: + mode: not-applicable + reason: "The parent entry routes here." + expression: + mode: not-applicable + reason: "Expression binds per page through interview_card, not per module." support_layers: - layer_id: L-CASES @@ -110,10 +421,10 @@ support_layers: directory: "Industry Cases/Composite Reconstructions" coverage: mode: derived - generator: "Tools/render_coverage_projection.py" + generator: "Tools/render_structure_projection.py" inputs_owner: "Corpus Planning/capability_matrix.yaml" bindings: - evidence_binding_owner: "kernel/K08 Metadata and Status/04 Evidence and Relationship Metadata.md" + evidence_binding_owner: "kernel/K08 Metadata and Status/08 Relationship Metadata Contract.md" - layer_id: L-SOURCES role: sources root: "Knowledge Sources" @@ -124,7 +435,7 @@ support_layers: layout: grouped taxonomy: axis: source-authority - page_field: source_type + page_field: source_class classes: - class: official-engineering directory: "Knowledge Sources/Official Engineering" @@ -140,12 +451,12 @@ support_layers: directory: "Knowledge Sources/Community Engineering" coverage: mode: derived - generator: "Tools/render_coverage_projection.py" + generator: "Tools/render_structure_projection.py" inputs_owner: "Corpus Planning/capability_matrix.yaml" bindings: - authority_taxonomy_ref: "profiles/agent-atlas/source-policy.md#Source Authority" - intake_policy_ref: "profiles/agent-atlas/source-policy.md#Verification Entry Points" - freshness_policy_ref: "profiles/agent-atlas/vocabulary-extensions.yaml" + authority_taxonomy_ref: "profiles/examples/agent-atlas/source-policy.md" + intake_policy_ref: "profiles/examples/agent-atlas/source-policy.md" + freshness_policy_ref: "profiles/examples/agent-atlas/vocabulary-extensions.yaml" index_mode: derived - layer_id: L-SYNTHESIS role: synthesis @@ -158,7 +469,7 @@ support_layers: taxonomy: null coverage: mode: derived - generator: "Tools/render_coverage_projection.py" + generator: "Tools/render_structure_projection.py" inputs_owner: "Corpus Planning/capability_matrix.yaml" bindings: question_identity_field: "claim_scope" @@ -168,27 +479,27 @@ support_layers: root: "Interview Preparation" entry: path: "Interview Preparation/Interview Overview.md" - expected_type: overview + expected_type: roadmap global_map_entry: E-INTERVIEW layout: grouped taxonomy: axis: artifact-kind page_field: type classes: - - class: interview-roadmap + - class: roadmap directory: "Interview Preparation/Roadmaps" - class: interview-card directory: "Interview Preparation/Topic Cards" - - class: interview-cheat-sheet + - class: cheat-sheet directory: "Interview Preparation/Cheat Sheets" coverage: mode: derived - generator: "Tools/render_coverage_projection.py" + generator: "Tools/render_structure_projection.py" inputs_owner: "Corpus Planning/capability_matrix.yaml" bindings: - artifact_registry_ref: "profiles/agent-atlas/expression-layer.md" - preparation_route_ref: "Interview Preparation/Interview Overview.md#Preparation Routes" + artifact_registry_ref: "profiles/examples/agent-atlas/expression-layer.md" + preparation_route_ref: "Interview Preparation/Interview Overview.md" readiness_projection: mode: derived - generator: "Tools/render_coverage_projection.py" + generator: "Tools/render_structure_projection.py" inputs_owner: ".cambium/state/coverage_ledger.yaml" diff --git a/profiles/examples/agent-atlas/vocabulary-extensions.yaml b/profiles/examples/agent-atlas/vocabulary-extensions.yaml index 914b248..92a8aab 100644 --- a/profiles/examples/agent-atlas/vocabulary-extensions.yaml +++ b/profiles/examples/agent-atlas/vocabulary-extensions.yaml @@ -23,6 +23,21 @@ fields: - missing - mapped - interview-ready + case_class: + owner: "profiles/examples/agent-atlas/structure-registry.yaml" + values: + - reported-system + - controlled-study + - composite-reconstruction + source_class: + owner: "profiles/examples/agent-atlas/structure-registry.yaml" + values: + - official-engineering + - paper + - standard + - book + - ecosystem-implementation + - community-engineering volatility_defaults: modeling-fundamentals: stable From 834f247e182ca5e8d543b1777d40012d45c8321f Mon Sep 17 00:00:00 2001 From: KimGLee <05_bolster_inkling@icloud.com> Date: Sat, 22 Aug 2026 16:07:45 +0800 Subject: [PATCH 2/2] Restore the two pinned founding rules the README rewrite dropped MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The rewrite compressed the empty-corpus paragraph into "adopt a Profile revision that configures Corpus Planning", which lost two load-bearing statements that LifecycleTextPinTests pins on purpose. The founding sequence must name the second R09 revision. Without it the configured Corpus Planning slot has no documented path from not-applicable, and the README would disagree with K02/03, R13, and the interview's closing review — the three owners of that statement. The adoption section must state that founding pages are never merged only to save files. Without the merge rule, "one page may serve as both owner and witness" reads as an invitation to collapse founding into one artificial page. Both restorations keep the new concise register and are made to both languages, so the two READMEs keep teaching the same founding sequence. test_profile_flow_e2e now passes whole (28 tests), matching the two CI failures on the n-r shard. Co-Authored-By: Claude Fable 5 --- README.md | 8 +++++--- README.zh-CN.md | 3 ++- 2 files changed, 7 insertions(+), 4 deletions(-) diff --git a/README.md b/README.md index b337ebf..e2c8d62 100644 --- a/README.md +++ b/README.md @@ -167,9 +167,11 @@ generated contracts, Runtime Cards, and adoption receipts. It restores the previous control plane if any step fails. An empty corpus follows the same adoption contract. First perform bounded -founding work to create real canonical owners and the residual-scan witness; -then adopt a Profile revision that configures Corpus Planning before starting -large-scale work. The exact sequence is documented in +founding work to create real canonical owners and the residual-scan witness — +one page may serve as both owner and witness when that is semantically +natural, but pages are never merged only to save files. Then a second R09 +revision configures the Corpus Planning slot before large-scale work begins. +The exact sequence is documented in [profiles/README.md](profiles/README.md#adoption-flow). ## Start Or Resume A Task diff --git a/README.zh-CN.md b/README.zh-CN.md index 2611393..23b3907 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -161,7 +161,8 @@ python3 Tools/apply_profile_adoption.py . --plan .yaml --apply 采用 receipt 绑定在一起。任何一步失败，工具都会恢复之前的控制面。 空语料库也使用同一份采用合同。先进行有界的 founding 工作，创建真实的 -canonical owner 和残留扫描见证；然后再次修订 Profile，配置 Corpus Planning， +canonical owner 和残留扫描见证——语义自然时一页可以同时充当 owner 和见证， +但绝不为了少建文件而强行合并；随后由第二次 R09 修订配置 Corpus Planning， 之后才能开始大规模工作。完整顺序见 [profiles/README.md](profiles/README.md#adoption-flow)。