hyperresearch-9-evidence-digest
GitHubHyperresearch V8流水线第9步，用于整合高置信度主张和原始引语，生成证据摘要文件。作为草稿生成的主要证据源，提供比抓取摘要更高保真度的信息索引。
Trigger Scenarios
Install
npx skills add NeverSight/learn-skills.dev --skill hyperresearch-9-evidence-digest -g -y
SKILL.md
Frontmatter
{
"name": "hyperresearch-9-evidence-digest",
"description": "Step 9 of the hyperresearch V8 pipeline. Assembles the top load-bearing claims and verbatim quotes from the claims JSONs into $TEMP_DIR\/evidence-digest.md — a single high-fidelity evidence index the draft sub-orchestrators read as primary evidence (higher fidelity than fetcher summaries). Invoked via Skill tool from the entry skill (full tier)."
}
Step 9 — Evidence digest
Run-scoped artifact contract
Use the active RUN_MANIFEST created by the entry skill. Substitute $ARTIFACT_DIR, $TEMP_DIR, $CLAIMS_DIR, $QUERY_FILE_PATH, and $FINAL_REPORT_PATH from that manifest before reading, writing, spawning subagents, or running Bash. Do not read or write legacy top-level run artifacts such as research/scaffold.md, research/prompt-decomposition.json, or research/temp/*.
Tier gate: Run for full. Skip for light.
Goal: assemble the top load-bearing claims and verbatim quotes from the claims JSONs into a single digest file the drafter reads as primary evidence — higher-fidelity than fetcher summaries.
Recover state
Read these inputs:
$ARTIFACT_DIR/scaffold.md— vault_tag$ARTIFACT_DIR/prompt-decomposition.json— atomic items, response_format, pipeline_tier- All
$CLAIMS_DIR/*.jsonfiles $TEMP_DIR/consensus-claims.json(if step 3 ran)$TEMP_DIR/contradiction-graph.json(if step 3 ran)
Procedure
-
Read all claims files from
$CLAIMS_DIR/*.jsonfor every non-deprecated note tagged with the vault tag. If no claim files exist (e.g., fetchers didn't produce them), skip this step. -
Filter and rank. Keep claims where
confidenceis"high"ORevidence_typeis"empirical"or"statistical". From the remainder, prefer claims with non-emptynumbersarrays and non-emptyquoted_support. Cap at 80–120 claims total forfulltier. -
Group by atomic item. Match each surviving claim to the atomic item it is most relevant to based on topic overlap — do not rely on exact field matching. A claim about "United Health Group regulatory exposure" serves the atomic item "UNH risk factors" even though no field matches exactly. Use the claim's
entities,stance_target,scope_conditions, andclaimtext holistically to judge relevance. When uncertain, include the claim under the most relevant item rather than dropping it to Ungrouped. Claims that genuinely don't map to any atomic item go into an "Ungrouped" section at the end. -
Include consensus and contested claims.
- If
$TEMP_DIR/consensus-claims.jsonexists, include its claims marked as[consensus]. - If
$TEMP_DIR/contradiction-graph.jsonexists, include the top 3–5 contested claim pairs with both sides'quoted_supportpassages.
- If
-
Write
$TEMP_DIR/evidence-digest.md. Format: one H3 per atomic item, bullet list of claims. Each bullet includes:- The
claimtext - The
quoted_supportverbatim passage (block-quoted) - The
source_note_id
Keep it scannable — this is an evidence index, not a narrative.
Example:
### Atomic item: Market growth in Southeast Asia - Annual growth rate of 12.4% in 2024 (empirical) > "Southeast Asian e-commerce GMV grew from $89B to $100B between 2023 and 2024, a 12.4% YoY expansion." [source-note-12] - Vietnam led by penetration rate (statistical) > "Vietnam reached 64% e-commerce penetration in 2024, the highest in SEA, surpassing Singapore (61%)." [source-note-19] - The
Exit criterion
$TEMP_DIR/evidence-digest.mdexists- Contains at least 30 claims for
fulltier - Grouped by atomic item with verbatim quoted_support and source_note_id
If fewer claims exist in total, include all of them.
Next step
Return to the entry skill (hyperresearch). Invoke step 10:
Skill(skill: "hyperresearch-10-triple-draft")
Step 10 is the most important step in the pipeline. Re-read the entry skill before invoking if needed — the triple-draft ensemble must spawn 3 draft-orchestrators for full tier.
Version History
- e0220ca Current 2026-07-05 23:31
