evidence-ledger
Fixed-column evidence ledger that traces every claim a decision depends on to a source, a reference, and a verification state. Use when: building or updating an evidence or citation ledger, verifying claims ahead of a gate ruling, or linking evidence into a decision record (ADR-0061).
- License 许可
- AGPL-3.0
- 最近更新
- 2026-11-14
- Stars
- ★ 25
- Forks
- 9
01 前言 Front Matter
- name
- evidence-ledger
- owner
- pm
- scope
- common
- status
- active
- version
- 1.1.0
- metadata
- { "type": "process", "triggers": [ "evidence ledger", "citation ledger", "claim verification", "source verification", "evidence tracking" ] }
- description
- Fixed-column evidence ledger that traces every claim a decision depends on to a source, a reference, and a verification state. Use when: building or updating an evidence or citation ledger, verifying claims ahead of a gate ruling, or linking evidence into a decision record (ADR-0061).
- last_reviewed
- 1787616000000
03 SKILL.md 正文
Purpose
Defines the common base format for evidence ledgers — the Evidence link of the Agent → Skill → Knowledge → Evidence → Rule → Decision chain (ADR-0061). A ledger row is the smallest unit of proof: one claim, where it came from, how it was checked, and whether it stands. Gate rulings and decision records cite ledger rows instead of re-deriving facts from prose.
Ledger Format
Every ledger uses the same five fixed columns:
| claim | source | url/ref | verification | status |
| Column | Content |
|---|---|
claim |
The single assertion being evidenced — one claim per row, never a bundled paragraph |
source |
Who or what asserts it (publication, filing, dataset, interviewee) |
url/ref |
The retrievable reference — URL, receipt number, docket ID, or file path |
verification |
How the claim was checked, by whom, and when |
status |
Current standing — e.g. VERIFIED, UNVERIFIED, CONTESTED, SUPERSEDED |
Rows are never rewritten to change history — see Append vs Supersede below.
Evidence Location Convention
docs/evidence/ledger.md— the ledger itself (one table, or one section per topic)docs/evidence/findings/— per-topic finding files holding detail too long for a row (methodology notes, excerpts, calculations); the ledger'surl/refcolumn links into them
Variant Overlays
The five columns are the common base. Variants MAY add requirements on top — stricter sourcing, extra columns, jurisdiction-specific citation rules — in the variant's own assets, never by editing this skill.
Canonical overlay example: co-news requires 2+ independent sources per material claim (NEWS-R1, registered inline in its context file) and instantiates the base format with claim / source 1 / source 2 / receipt number / status columns in its source-verification-ledger skill. The two source columns and the receipt-number column are the overlay; the row-per-claim, fixed-column, status-tracked discipline is the base.
Document-form overlay (co-consult, T-20261007-014): the industry-research-pack skill satisfies the same 2+ independent sources requirement in document shape — one document per source with a five-section per-source skeleton and figure-by-figure cross-validation against a named second source; a figure that lacks its second source carries the ⚠ SINGLE-SOURCE warning mark instead of a ledger row. The pack is also a citable substrate: consumer deliverables cite its per-source documents per section rather than restating claims (the co-deck PR #135 contract — 114 slides citing per-slide source docs). Variants reaching the same discipline independently should teach one of these overlay forms rather than inventing a third convention.
Registry-backed overlays (formal-var form)
Some registry-backed variants carry their evidence plane as FORMAL VARIABLES instead of a prose ledger — machine-checked rows in a schema-governed registry (e.g. an evidence-variable registry with run-scoped snapshots), with calibration corpora or finding files attached as provenance. That is a legitimate overlay form of this skill: the row-per-claim discipline becomes row-per-VARIABLE, verification becomes the registry's verification metadata, and supersession happens through the registry's own reversal/amendment protocol. Whichever form a variant uses, decision records evidence_refs[] must resolve inside it — a registry key name for formal-var overlays, a ledger row ID for prose ledgers.
Append vs Supersede
- Append when adding new evidence or a new verification pass — new rows, new finding files.
- Supersede when a claim's standing changes (a source retracts, a verification fails, a stronger source arrives): keep the original row, set its
statustoSUPERSEDED, and add a new row whoseverificationnotes what replaced it. - Never delete a row or overwrite a
verificationentry — the ledger must show what was believed, when, and why it changed.
Linking Evidence to Decision Records
Decision records (see the decision-record skill) cite evidence through their evidence_refs[] frontmatter. Every entry must resolve to a ledger row (quote the claim text or a stable line ID) or a finding file under docs/evidence/findings/. A ruling that cites no evidence is incomplete; evidence that no ruling cites is unexamined background.
04 同仓库更多 Skills
i18n-layout
Text layout and encoding guidance: character encoding (UTF-8, legacy Korean code pages, BOM hazards), line endings, RTL/bidi handling, and script-specific font selection (Hangul focus). Use when: encoding is corrupted or mojibake appears, an RTL locale must render correctly, fonts must be chosen for a script, or CRLF/BOM issues surface.
i18n-layout
Text layout and encoding guidance: character encoding (UTF-8, legacy Korean code pages, BOM hazards), line endings, RTL/bidi handling, and script-specific font selection (Hangul focus). Use when: encoding is corrupted or mojibake appears, an RTL locale must render correctly, fonts must be chosen for a script, or CRLF/BOM issues surface.
i18n-layout
Text layout and encoding guidance: character encoding (UTF-8, legacy Korean code pages, BOM hazards), line endings, RTL/bidi handling, and script-specific font selection (Hangul focus). Use when: encoding is corrupted or mojibake appears, an RTL locale must render correctly, fonts must be chosen for a script, or CRLF/BOM issues surface.
i18n-layout
Text layout and encoding guidance: character encoding (UTF-8, legacy Korean code pages, BOM hazards), line endings, RTL/bidi handling, and script-specific font selection (Hangul focus). Use when: encoding is corrupted or mojibake appears, an RTL locale must render correctly, fonts must be chosen for a script, or CRLF/BOM issues surface.
i18n-layout
Text layout and encoding guidance: character encoding (UTF-8, legacy Korean code pages, BOM hazards), line endings, RTL/bidi handling, and script-specific font selection (Hangul focus). Use when: encoding is corrupted or mojibake appears, an RTL locale must render correctly, fonts must be chosen for a script, or CRLF/BOM issues surface.
i18n-layout
Text layout and encoding guidance: character encoding (UTF-8, legacy Korean code pages, BOM hazards), line endings, RTL/bidi handling, and script-specific font selection (Hangul focus). Use when: encoding is corrupted or mojibake appears, an RTL locale must render correctly, fonts must be chosen for a script, or CRLF/BOM issues surface.