research

name

research

description

Use when the user asks to research, investigate, verify facts with sources, compare competitors, study a market or industry, review literature, prepare an evidence-backed report, decide whether a topic needs external Deep Research, or make an evidence-backed judgment or recommendation. Runs breadth-first Quick Recon, Claim Ledger gating, and an Adjudication Hypothesis Spine when needed. This generated target uses Codex native collaboration and web tools.

✦ A GiaSip generated target · github.com/GiaSip/giasip-skills

GiaSip Research — Codex

Generated from the neutral canonical Research method. Do not edit this target by hand; reconcile changes into the canonical source and rebuild.

Codex Runtime Contract

Apply the compiled method directly in Codex; do not route through another Research skill or adapter to recover its core workflow.

• Inspect the callable collaboration schema before delegating and pass only supported fields.
• Use Codex's available public-web search and page-reading tools.
• Use 2 lightweight workers by default and 3 only for three genuinely orthogonal slices.
• Keep run IDs, artifact persistence, ledger mutation, synthesis, and delivery in the main task.
• If worker concurrency is unavailable, execute the selected slices sequentially and disclose the fallback.

---

Portable Research

This is the only human-maintained semantic source for the Research method. Claude Code, Codex, and public GiaSip packages are generated host targets; do not copy host tool names or invocation syntax back into this file.

Portability contract: “worker” means the current host’s independent task primitive. When parallel workers are unavailable, execute the same slices sequentially and disclose the fallback. Host-native wrappers define concrete tools, persistence roots, and invocation.

---

Core Principles

1. Recon before escalation — Every research task starts with Quick Recon. 2-5 minutes of initial search helps you decide: deliver directly, or escalate to Deep Research with clear questions. Skipping Recon to submit Deep Research blindly wastes quota
2. Capability fit first — When Deep Research is needed, the only criterion for platform selection is "who is best at this type of task," not cost — within your subscribed platforms
3. Language determines the candidate pool — Chinese tasks prioritize domestic platforms, English tasks prioritize international platforms, mixed tasks use both
4. Combination over single (high-stakes only) — Multi-platform cross-validation is only worth it for high-stakes questions (≥10pp numbers / licenses / policy-legal-financial / AI same-faction claims); for general topics (market/competitive/industry), a single platform + primary source grounding is sufficient — don't burn quota on unnecessary multi-platform runs
5. Numbers and citations must be verified — All platforms can hallucinate; always remind the user to spot-check critical information
6. Quota awareness — Some platforms have monthly caps (e.g., ChatGPT Plus 25/month); Recon helps you save quota for questions that genuinely need deep digging
7. Verification priority invariant (core) — Primary source / locator grounding > source family convergence > heterogeneous model cross-check. First determine whether a claim has a ground-truth locator, then decide whether to spend on heterogeneous models. Heterogeneous reviewers cannot substitute for missing primary source locators (empirical: 1 model that read the primary source > 3 heterogeneous models guessing from memory). "Evidence source family" (owner/regulator/official/independent/vendor/aggregate) and "reviewer faction family" (cross-faction) are two dimensions — don't conflate them.

---

Core Flow

Step 0: Establish the Run Directory (persistence convention, spans the whole flow)

Any task that enters Recon, or skips Recon to escalate directly to DR, first fixes a run directory and physically persists all intermediate products — this is the prerequisite for Claim Ledger / Mini Assurance / Deep Research reflow to actually work. Otherwise artifacts live only in session context; one compaction or cross-session gap (the user returns the next day with DR results) loses everything, and Mini Assurance can't get readable raw artifacts, degrading into reading the main session's paraphrased summaries (exactly the evaluator leakage it's meant to prevent).

• Location: project research → <project>/research/<topic>-<YYYY-MM-DD>/; no project home → ~/research-runs/<topic>-<YYYY-MM-DD>/
• Structure:

  <run_dir>/
    manifest.md               # run state anchor (cross-session recovery entry, see below)
    artifacts/                # each recon worker facet/gap's full raw output, one .md
    ledger.md                 # Claim Ledger master table (Step 2.5) + Hypothesis Matrix as an independent section (Adjudication, Step 2.5)
    recon-report.md           # final report for Recon direct delivery
    deep-research-prompt.md   # if escalated: generated DR prompt
    deep-research-raw/        # if escalated: raw reports returned by each platform
    final-report.md           # merged Recon + DR final version
    audit.md                  # Mini Assurance / fact-check audit results
  

• Persistence is the orchestrator's responsibility, not the worker's: after each Round 1 or Round 2 worker returns, the orchestrator immediately writes its full raw output to <run_dir>/artifacts/<facet>.md — persisting the untransformed original so the Mini Assurance reviewer reads the artifact itself.
• manifest.md = cross-session recovery anchor: status (in_recon / awaiting_user_dr / delivered / partial / blocked_needs_approval) + research mode (Retrieval/Mapping/Adjudication — so a resumed session knows whether hypotheses need updating when DR returns) + current step + todos + items awaiting user confirmation. Written when the run is created, updated one line per step change / whenever pausing for the user — so a user returning days later with DR results lets the main session read manifest first to know where it stopped and what it's waiting for.
• Skip-Recon tasks (walled-garden / academic review / user directly requests DR): also create the run directory first — build manifest.md (status: awaiting_user_dr + flag recon_skipped: true) + an empty ledger.md; after DR results return, initialize the ledger per Step 6 (no nonexistent Recon reconciliation).
• Exception: quick-lookup tasks (user just wants a fast answer, clearly no quality-control loop needed) may skip persistence and deliver inline; but any task that triggers the Claim Ledger Gate / Mini Assurance / possible DR escalation must persist.

Step 1: Analyze the Research Task

Extract from user input:

• Research language: primarily Chinese / primarily English / mixed
• Research type: academic/professional / strategic/industry analysis / fact-checking / enterprise data integration / Chinese walled-garden platform data / ultra-long document analysis / sentiment analysis / mixed
• Depth requirement: quick lookup (< 10 min) / standard report / deep research
• Hallucination tolerance: extremely low (academic/legal/financial/model-license/primary-source verification) / low (tool selection/competitive) / medium (business) / high (exploratory)
• Citation requirement: academic-grade (sentence-level tracing) / business-grade / informal
• Special platform needs: whether CNKI / Xiaohongshu / WeChat Official Accounts / Twitter, etc. are needed
• Research mode (determines whether the Hypothesis Spine engages): Retrieval (a single fact / price / list → skip the spine) / Mapping (survey / landscape / non-adjudicative comparison → coverage only; default) / Adjudication (why / evaluation / recommendation / decision — "should we", "which is better" → fully engage the spine). When unsure, default to Mapping; escalate to Adjudication only when the sub-question requires comparing plausible options and defending an inferential judgment (even if the outcome is underdetermined); route mixed tasks per sub-question. Full spec: references/hypothesis-spine.md.

Explicit declaration (required): After analysis, state the seven dimensions above — especially hallucination tolerance + citation requirement + research mode — in one or two lines before starting. They directly drive fact-check triggering (extremely low + academic-grade), the Round 2 primary-source constraint, and whether the Hypothesis Spine engages (Adjudication); skipping the declaration means re-improvising the judgment at every branch point, rendering the trigger chain moot.

Step 2: Quick Recon — Round 1 (Breadth Reconnaissance)

Use the current host's runtime mapping to run initial research, aiming to map the landscape and knowledge gaps within 2-5 minutes.

When to Skip Recon

Skip directly to Step 4 (platform matching) in these scenarios:

• User explicitly says "submit to Deep Research directly" or "skip preliminary research"
• The task's core need is walled-garden platform data (CNKI / Xiaohongshu / WeChat Official Accounts, etc.) that the current host cannot reach
• The task is an academic literature review requiring full papers and citation chains beyond the host's public-web coverage
• The user has already done preliminary research and comes with specific questions

Round 1 Execution

Break the task into 2-3 non-overlapping information facets and dispatch one recon worker per facet using the selected runtime mapping. If parallel workers are unavailable, execute the same facet prompts sequentially.

Facet decomposition = diverge, then select on two axes (not slotting the research type into a fixed template — templates make the model converge on the most typical three-part split and miss unknown angles):

1. Diverge: privately list 4-6 candidate angles (default, not a hard rule; for high-uncertainty / high-risk topics include 1 cross-domain / challenger angle, but no novelty theater).
2. Select 2-3 that satisfy two orthogonal criteria at once: ① cover different decision-relevant unknowns (non-overlapping) ② point to different evidence sources / methods (avoid re-answering the same question with correlated evidence). Selecting only on "different source" makes three workers re-answer one question with different sources.
3. Name the residual in one line: state explicitly what these 2-3 facets do not cover (at k=2-3 coverage is inherently incomplete; making the residual explicit beats pretending completeness).

Per-research-type example splits (market / competitive / industry / tech-selection) live in references/hypothesis-spine.md — reference examples, not templates to slot into.

Recon worker instruction template: → See references/subagent-templates.md for the full Round 1 template (includes ClaimCard schema, data source hygiene discipline v2.4, and output format).

Tool selection:

• Primary: the host's native web search + page reading tools — zero additional external quota
• Fallback: a browser or extraction tool only when the native reader hits JS rendering or anti-scraping blocks

Step 2.5: Claim Ledger Gate + Hypothesis Matrix + Gap Assessment & Round 2 (Conditional)

After all Round 1 workers return, the orchestrator runs a Claim Ledger Gate first, then (Adjudication only) forms a Hypothesis Matrix, then does gap assessment to decide whether Round 2 is needed.

Mode recheck (ALL modes run this — do it before the branch below). Do the Round 1 findings require comparing plausible competing answers to the same question (and are there at least two live candidates)? Yes → escalate to Adjudication even if Step 1 tagged Retrieval/Mapping, and build the Hypothesis Matrix section below. No → the task stays Retrieval/Mapping and skips the Hypothesis Matrix. (Sources merely disagreeing on a fact → that's Claim Ledger conflict arbitration, not Adjudication.) This recheck lives here, above the Adjudication-only section, so a Mapping run cannot skip it and silently miss its own escalation.

Claim Ledger Gate (v2.5)

Design origin: Inspired by the claim-level quality control approach from Claude Code Workflow's deep-research skill. Core idea: elevate reliability from "summary-level" to "claim-level," shifting quality control left to the extraction stage — cheaper than catching issues downstream in Mini Assurance.

Consolidate all worker ClaimCards into a single ledger. Ledger schema (per entry): claim_id / normalized_claim / importance(central/supporting/context) / risk_reason(why high-risk) / source_family(owner/regulator/official/independent/vendor/aggregate/community) / locator(primary source locator) / status(confirmed/weak/unresolved/refuted) / merged_from(repost merge count) / counterquery

Run through the gate in order:

1. Merge duplicates — URL dedup + claim-level semantic dedup (the same number reposted by 5 aggregators ≠ 5 pieces of evidence; merge to 1, record merged_from)
2. Flag high-risk — risk_reason non-empty = high-risk (≥10pp numbers / license / policy-legal-financial / AI same-faction assertions)
3. Central claims without locator → send back to Round 2 (no evidence-free conclusions allowed)
4. Central claims supported only by vendor/aggregate → mark weak, excluded from conclusion topic sentences (can only appear in "pending verification")
5. Claims with conflicting evidence → selective adversarial verification (see below, not full-coverage)
6. Uncertain claims → mark unresolved, not refuted (refuted requires explicit conflicting evidence; uncertain ≠ disproven, just not reportable as fact)

Selective adversarial verification (high-risk / conflicting claims only, not full coverage). Strictly follow Principle 7's verification priority invariant through three levels:

• ① Primary source grounding first: when owner/regulator/official primary sources are directly readable, read the original text to arbitrate — most conflicts resolve here, no need for heterogeneous models.
• ② Then source family convergence: have a skeptic search for counter-evidence across different evidence source families (owner / independent test / vendor); arbitrate by source family, not by agent vote count (running the same search engine 3 times is just correlated noise).
• ③ Heterogeneous reviewer faction (cross-faction) last: escalate only when the topic involves AI same-faction content (see Step 3). This is the reviewer/model dimension, orthogonal to ②'s "evidence source family" — don't conflate.
• Verdict: explicit conflicting evidence → refuted; insufficient evidence → unresolved (excluded from factual narrative); multi-source-family corroboration → confirmed.

★ Hypothesis Matrix (built when the mode recheck above lands on Adjudication; after breadth + the Claim Ledger Gate, before Round 2)

The third axis (argument validity — evidential and inferential support, not formal logical validity). Retrieval / Mapping skip this section. It converts scattered claims into evidence-tested candidate answers — the "information collector → argument engine" step that generic search-and-summarize flows omit. Hypotheses do NOT enter the Claim Ledger (to avoid polluting its confirmed/weak semantics and a circular self-justification hole); keep them in a separate section of the same ledger.md.

• After breadth returns, converge the findings into 2-3 mutually competing candidate answers (Chamberlin's multiple working hypotheses), including one null / status-quo / "not worth doing" hypothesis (to prevent nominal alternatives that share one hidden assumption). Form hypotheses after breadth (to prevent anchoring). Give each a type-appropriate discriminator (see hypothesis-spine.md §4).
• Record per hypothesis: hypothesis_id / candidate_answer / type(causal|decision|forecast) / supporting_claim_ids / contradicting_claim_ids / discriminator_or_falsifier / status(active|conditional|rejected) / revision_history / residual_uncertainty; and a matrix-level matrix_outcome(preferred|underdetermined|none_of_current) (+ preferred_hypothesis_id). Note underdetermined is a matrix outcome, not a per-hypothesis status.
• Full schema, discrimination discipline, warrant gate, and the Claim-Ledger-vs-Matrix boundary: see references/hypothesis-spine.md.

Gap Assessment Logic

Design philosophy (inspired by MiroThinker's Interactive Scaling): one-shot broad search tends to miss key directions. Round 2's value lies in "searching again with Round 1's knowledge" using more precise keywords to fill critical gaps, not repeating Round 1's breadth.

After collecting Round 1 results, check knowledge gaps item by item:

Round 2 triggers (any one sufficient):

• Coverage gap: Round 1 revealed unexpected new directions / a critical single-source data point affecting core judgment / contradictory workers / a clearly missed angle
• ★ Discrimination gap (Adjudication mode): evidence cannot yet distinguish competing hypotheses → Round 2 designs a discriminating / disconfirming query (a strong-inference-inspired discrimination pass, "find evidence against Hx"), not filling a vague blank

Skip Round 2 conditions (any one sufficient to skip):

• Retrieval/Mapping: scope audit passes (against ① the user's original explicit ask ② discarded candidate angles ③ a generic-dimension backstop checklist — entities / process / space-time / stakeholders / rival hypotheses — only when a key residual gap remains); Adjudication: surviving hypotheses are sufficiently discriminated (note: fact count ≠ coverage proxy — do NOT use "≥N findings" as a skip criterion)
• Gap nature requires walled-garden platforms or academic full text — Round 2 can't reach them; escalate to Deep Research directly
• User's need is quick-lookup level, Round 1 is sufficient
• Round 1 already consumed significant time (> 5 min), not worth more waiting

Precedence when triggers and skip conditions both fire (ordered): unreachable evidence (walled-garden / academic full text) → escalate to DR; otherwise a material coverage / discrimination gap → Round 2; otherwise skip. A time-budget cutoff yields partial (or underdetermined for Adjudication) — it does not imply the hypotheses were sufficiently discriminated.

Round 2 Execution

Unlike Round 1, Round 2 is a targeted pass, not a broad sweep:

• Dispatch only 1-2 workers (not 2-3)
• Each worker targets one specific gap, not a broad facet
• Worker instructions include Round 1's high-confidence findings as context to avoid re-searching known information
• Adjudication: design the query to discriminate or disconfirm a surviving hypothesis — the worker prompt says outright "find evidence against Hx" (a fresh worker reduces anchoring, but the prompt must still request symmetric treatment of supporting and contrary evidence)

Additional constraints for high fact-density task types: When "hallucination tolerance = extremely low" AND "citation requirement = academic-grade", Round 2 must include at least 1 "direct primary source reading" task. → See references/subagent-templates.md for primary source types, unit sanity check rules, and the full Round 2 template (includes ledger_patch and, for Adjudication, hypothesis_patch format).

After Round 2 returns, the main session applies ledger_patch back to the master ledger (re-running the gate) to ensure Round 2's critical corrections enter the ledger — otherwise Step 3 Mini Assurance can't see them. Adjudication mode: Round 2 also emits hypothesis_patch (update hypothesis_id=... status→...) to update the Hypothesis Matrix; absence of evidence ≠ refutation — only explicit conflicting evidence or a failed predicted observation rejects a hypothesis; a failed search records an unresolved Claim Ledger entry and leaves the hypothesis status unchanged (never write unresolved as a hypothesis status). Round 2 cap stays at 1 round (no stopping rule = infinite loop).

Step 3: Synthesis & Decision

After collecting all Round 1 (and Round 2, if triggered) results, evaluate next steps.

Decision Criteria

Recon is sufficient (deliver directly):

• Quality-control invariant, split by sentence type: fact sentences → map to one confirmed ledger claim (weak/unresolved excluded from topic sentences, only in "pending verification"); conclusion / argument sentences (Adjudication) → map to a surviving hypothesis + pass the warrant gate — a conclusion is an argument node inferred from multiple facts via a warrant, not required to be a "confirmed fact" itself
• Warrant gate (load-bearing conclusions only): claim / evidence IDs / warrant (reason) / qualifier / key defeater — the audit is not "are there zero contradicting claims" (a preferred hypothesis may legitimately have some, and requiring zero rewards omission); a fresh reviewer verifies the evidence IDs exist with acceptable ledger status, the warrant sufficiently supports the conclusion, the qualifier matches evidence strength, and known contrary claims / defeaters are complete and materially addressed (no unaddressed decisive contradiction after a documented countersearch); self-evident light conclusions don't need it
• All central metric / license / policy / ≥10pp number claims have owner/regulator/official/independent-level locators (not 5 aggregate sites padding the count)
• Remaining gaps involve only peripheral details, not affecting core judgment
• User's need is quick-lookup or standard-report level, no walled-garden data or academic-grade citations required
• ★ Adjudication's legitimate terminal outcomes include underdetermined: when evidence can't decide, do NOT force a single surviving hypothesis (forcing = rejecting an un-refuted hypothesis just to fit a format = premature closure); honestly output "multiple hypotheses coexist + the conditions under which each holds + which evidence would decide between them"

→ Compile Recon results directly (merge Round 1 + Round 2), output research report. Jump to Step 5's "Recon direct delivery" template.

Need Deep Research (escalate):

• After Round 1 (+ Round 2), critical data points still lack reliable sources
• Major information conflicts discovered, and the current host's web-reachable sources are exhausted
• Obvious knowledge gaps remain, fillable only through walled-garden platforms, academic full text, or deep industry data
• User requests deep-research level

→ Proceed to Step 4 with both rounds of Recon findings, generate a targeted Deep Research prompt. The prompt quality will be higher because "known information" is richer and "questions to deep-dive" are more precise.

User briefing format:

## Quick Recon Results

### Information Gathered
- [Round 1 + Round 2 core findings summary, 3-7 items]

### Knowledge Gaps
- [Unresolved questions, if any]

### Recommendation
[Deliver directly / Escalate to Deep Research, with rationale]
[If Round 2 was triggered, briefly explain what it filled]

When delivering from Recon (no paid action), don't wait for confirmation — compile results and output the report directly. But any action consuming external paid quota (DR escalation, or paid fact-check within direct delivery) follows the unified authorization rule below: report cost first, wait for confirmation — no longer distinguishing between the two.

Unified external paid-quota authorization rule (covers the whole flow): delivering the report itself needs no confirmation, but any action consuming external paid quota — DR escalation, the high-fact-density Perplexity DR fact-check, cross-faction reviewers — always reports "platform + estimated count/cost" first and waits for user confirmation. Exception: the user has explicitly said "submit directly / no need to ask". This rule takes precedence over any "default run / mandatory" phrasing in the fact-check protocol.

Fact-Check Protocol for High Fact-Density Tasks

When "hallucination tolerance = extremely low + citation requirement = academic-grade", reports delivered from Recon must undergo independent fact-checking. The protocol uses a three-layer approach: primary source locator reading first → Perplexity Deep Research → independent worker blind-spot check, with cross-faction discipline for AI same-faction content. For Adjudication reports, Mini Assurance also audits conclusion sentences against the Hypothesis Matrix (are the warrant's evidence IDs valid; are known contrary claims / defeaters materially addressed) — not just fact sentences against the ledger. The reviewer receives the conclusion sentences and their warrant records as extracted inputs, not by reading the draft's conclusions section.

→ See references/fact-check-protocol.md for the full protocol (Layer 0/1/2 flow, v2.4 cross-faction discipline, conflict arbitration order, empirical cases, and Mini Assurance audit procedure including the Adjudication conclusion-sentence rule).

Step 4: Match Deep Research Platform

Only executed when Step 3 determines escalation is needed.

Refer to references/platform-profiles.md for each platform's capabilities; refer to references/matching-rules.md for matching logic.

Core approach: identify the task's 1-2 most critical requirement dimensions, match the platform strongest on those dimensions. Don't prioritize a platform because it's "free" or "cheap" — pick the most capable.

★ Adjudication mode: the DR prompt carries the Hypothesis Spine — not "dig deeper into X," but "evidence A/B is confirmed (confirmed claims); please find evidence that distinguishes H1/H2" (carrying the surviving hypothesis set + each one's discriminator). When Recon was skipped there is no surviving set yet — send provisional, explicitly unverified candidates, or ask DR to construct and test them (see Step 6's no-Recon branch). This is the spine's most valuable application: spend the most expensive DR quota on discrimination rather than aimless breadth.

Step 5: Output Research Plan

Based on Step 3's decision, select the corresponding output template.

---

Template A: Recon Direct Delivery

Used when Recon results sufficiently answer the user's question.

Structure compatibility: If the task has a dedicated output structure (e.g., 6-section: TL;DR / fact table / conflict verification table / action items / pending confirmation / source annotations), prioritize the task's structure as the main body; Template A's "Limitations + For deeper investigation" sections serve as closing. Mixing both is reasonable and doesn't count as template deviation.

Adjudication delivery: lead with the defended judgment (or an honest underdetermined + holding conditions), back the load-bearing conclusion with a warrant gate (claim / evidence IDs / warrant / qualifier / key defeater), and summarize the Hypothesis Matrix terminal states. Do not force a single winner when evidence is insufficient.

## Research Report: [Topic]

> Based on Quick Recon using the current host's web and worker tools; no external Deep Research platform used.

### Key Findings
[Integrated key findings across facets, structured presentation]

### Data Sources
| Source | URL | Confidence |
|--------|-----|------------|

### Limitations
- [Aspects not covered by Recon]
- [Data timeliness notes]

### For Deeper Investigation
> If you feel any aspect needs more depth, let me know and I can dispatch targeted Deep Research.

---

Template B: Escalate to Deep Research

Used when external platform deep-diving is needed. Key difference: the prompt includes only confirmed-status Recon findings as "established" background (weak/unresolved excluded, to avoid DR digging along a wrong anchor), so Deep Research focuses on unknowns.

No-Recon variant (skip Recon, submit DR directly): there are no confirmed claims to include — the background section lists only the user-provided raw constraints, explicitly tagged [user input, unverified], and never masquerades as "confirmed through preliminary research". Returning DR results always go through Step 6's gate (empty-ledger initialization).

## Research Plan

### Quick Recon Summary
> [2-3 sentences summarizing known information and core gaps]

### Recommended Platform: [platform name]

**Why this one:** [1-2 sentences focusing on capability-fit rationale]

**Suggested research prompt:**
```
[topic description]

Background (confirmed through preliminary research — confirmed claims only):
- [Recon confirmed finding 1]
- [Recon confirmed finding 2]
- [Recon confirmed finding 3]

Please focus on these questions (not covered by preliminary research):
1. [knowledge gap 1]
2. [knowledge gap 2]
3. [knowledge gap 3]

[output format / other requirements]
```

**Authorization confirmation (required before submitting, wait for user OK):**
- Platform + estimated count: [e.g., Perplexity DR ×1]
- Estimated cost / quota consumption: [e.g., ~$5, or "1 of ChatGPT DR's 25/month quota"]
- ⏸ Wait for user authorization — if declined, don't submit; mark manifest `blocked_needs_approval`, deliver existing Recon as `partial`

**How to use:**
- [Entry URL/path]
- [Estimated duration]

**Caveats:**
- [Known limitations]
- [Parts requiring manual verification]

### Combination Strategy (if applicable)
| Role | Platform | Sub-task | Rationale |
|------|----------|----------|-----------|

> [How platforms coordinate, why combination is needed]

### Alternative Plan
> [When to use alternative, capability differences vs. primary choice]

### Next Steps
> Once you confirm the plan, say "run it" or "submit research" and I'll generate a dispatch page — one-click copy prompt, one-click open platform, all submissions done in 30 seconds.
> After the DR report returns, run Step 6 (reflow: gate + reconcile + persist).

---

Step 6: Deep Research Result Reflow (closing the escalation loop)

Whenever DR was escalated (or Recon was skipped for direct DR submission), you must run this step after the report returns — otherwise quality control covers only the "no-escalation" half of tasks, leaving the most expensive, highest-stakes half naked.

After the user brings back the DR report (possibly in a new session — read <run_dir>/manifest.md first to restore run state, then continue from ledger.md + artifacts/):

1. Persist: store the raw report at <run_dir>/deep-research-raw/<platform>.md
2. Gate: extract central / high-risk claims from the DR report into ClaimCards, run them through Step 2.5's Claim Ledger Gate (same schema, risk flags, locator requirements) — DR platforms hallucinate too; "paid depth" doesn't exempt them
3. Reconcile: align item-by-item against the Recon ledger; conflicts follow the existing Step 2.5 / Step 3 arbitration order (primary source reading > source family convergence > heterogeneous reviewer); DR conclusions don't automatically override Recon's existing primary-source grounding
4. ★ Update hypotheses (Adjudication): after DR reflows, use hypothesis_patch to update the Hypothesis Matrix status (which hypothesis DR evidence distinguished / rejected / turned conditional, and the matrix-level outcome), not just the Claim Ledger
5. Merge & persist: produce <run_dir>/final-report.md, fact sentences mapping to confirmed ledger claims and conclusion sentences mapping to surviving hypotheses + warrant gate (same split as Step 3)
6. Skip-Recon tasks (walled-garden / academic review, etc.): ledger.md starts empty — DR results directly extract ClaimCards to initialize the Claim Ledger; there is no Recon ledger to reconcile against, so skip reconciliation and run the gate on the DR-seeded claims. For Adjudication, the candidates sent in Step 4 were provisional/unverified, so form the initial Hypothesis Matrix from the gated DR evidence (don't treat the provisional set as established), then merge. Not exempt from quality control for "having skipped Recon"

manifest state transitions (minimal set, not a full state machine): escalate DR → awaiting_user_dr; user declines paid authorization → blocked_needs_approval; budget/time exhausted → partial; final delivery → delivered. Update one line per transition.

---

Detailed Reference

• references/hypothesis-spine.md — Adjudication third axis: research modes, Hypothesis Matrix schema, discrimination discipline, warrant gate
• references/platform-profiles.md — Deep Research platform capability profiles
• references/matching-rules.md — Platform matching logic and decision tree
• references/fact-check-protocol.md — Fact-check protocol (v2.2+v2.4) + Mini Assurance audit (incl. Adjudication conclusion-sentence rule)
• references/subagent-templates.md — recon worker instruction templates (Round 1 + Round 2, incl. hypothesis_patch) + unit sanity check