Mixed freshness. Depending on the panel, rows here are read live from elisa-core or come from a frozen copy of ELISA’s local store taken on 22 October 2026 — each panel says which, underneath itself. No agent is running.

ELISA
Connecting

Research Questions

The decision question each piece of research exists to answer, with its scope, deadline and evidence standard.

Run by Research Orchestrator

Nothing here yet

The decision question each piece of research exists to answer, with its scope, deadline and evidence standard.

The backend for this is still being built.

Technical detail
  • GET /api/jobs?sector=research&type=question

This sector’s fleet

Real rows from elisa.db, scoped to this sector’s own agents — not the business surface above, which this sector does not yet have rows for.

Agents (7)

  • Orchestrator · Research sector orchestrator

    Turn a decision question from another sector into a scoped research plan with an explicit evidence standard, and refuse to return a brief the evidence cannot support.

    research.orchestrator
    0 skillsIdle — nothing queued
  • Agent 1 · Market & Customer Intelligence

    Collect market and customer evidence in the customer's own language, and map demand signals without deciding what they mean.

    research.agent_1
    1 skillIdle — nothing queued
  • Agent 2 · Competitor & Source Analysis

    Compare competitors and judge every source's reliability, freshness and bias, marking conflicts rather than averaging them away.

    research.agent_2
    1 skillIdle — nothing queued
  • Agent 3 · Evidence & Insight Briefs

    Synthesize finding, implication, confidence and recommendation into the sector's one insight brief, and state the uncertainty as plainly as the answer.

    research.agent_3
    0 skillsIdle — nothing queued
  • Agent 4 · Source & list discovery

    Find where data actually lives, prove a source yields usable rows at a stated cost, and hand sales a working lead-source recipe instead of a promise.

    research.agent_4
    0 skillsIdle — nothing queued
  • Agent 5 · Capability, vendor & tooling evaluation

    Evaluate the tools, APIs, models and vendors the business is about to depend on, with real pricing, real limits and a named failure mode.

    research.agent_5
    0 skillsIdle — nothing queued
  • Agent 6 · Corpus & knowledge curation

    Keep the research corpus retrievable and honest: one copy of each source, dated, deduplicated, and never silently regenerated.

    research.agent_6
    1 skillIdle — nothing queued

7 rows · agents, agent_status_cache · read 2026-10-25T19:11:47.066Z

Skills (3)

3 distinct skills held in this sector — out of 262 in the catalogue fleet-wide.

  • note-taking×1
  • research×1
  • web×1

3 rows · agent_skills ⋈ skills, scoped to this sector · read 2026-10-25T19:11:47.066Z

Lessons (21)

3 binding (L0, enforced at the tool call) and 18 advisory (L1) in this sector — 40 of 155 are binding fleet-wide.

Binding — L0

  • no.lead.writesaccepted

    This agent never writes a lead or a contact row. It hands sales.agent_1 a recipe; duplicate lead authorship is how a CRM stops being trustworthy.

    Source & list discovery · agent · confirmed 1×
  • credentials.never.writtenaccepted

    Never write an API key, token or secret into a file, a record or a log. An evaluation references a key by name, never by value.

    Capability, vendor & tooling evaluation · agent · confirmed 1×
  • no.destructive.regenerationaccepted

    Never delete a directory to rebuild it. Regeneration is skip-by-default with explicit --force and --dry-run flags.

    Corpus & knowledge curation · agent · confirmed 1×

Advisory — L1

  • insufficient.is.a.resultaccepted

    Returning 'the evidence does not support an answer', with what is missing and what it would cost, is a completed job. Manufacturing a confident answer is not.

    Research sector orchestrator · agent · confirmed 1×
  • conflicts.preservedaccepted

    Two sources that disagree are both recorded with their disagreement named. Averaging them away destroys the only signal worth having.

    Research sector orchestrator · agent · confirmed 1×
  • quotes.verbatimaccepted

    Customer language is stored exactly as written, with its url and date. Paraphrasing turns evidence into opinion and the brief inherits the error.

    Market & Customer Intelligence · agent · confirmed 1×
  • single.source.is.not.a.signalaccepted

    A demand signal needs at least two independent dated sources. One undated page is an anecdote and is recorded as one.

    Market & Customer Intelligence · agent · confirmed 1×
  • firecrawl.linkedin.blockedaccepted

    linkedin.com is blocked on the rotator. Report the block; never convert it into a finding of no demand.

    Market & Customer Intelligence · agent · confirmed 1×
  • no.ranking.on.thin.dataaccepted

    Fewer than three comparable data points per competitor means no ranking is produced. Deepak's standing rule: refuse to rank on thin data.

    Competitor & Source Analysis · agent · confirmed 1×
  • conflicts.preservedaccepted

    Two sources that disagree produce a conflict_record with both values, both dates and the reliability reasoning. They are never merged into one number.

    Competitor & Source Analysis · agent · confirmed 1×
  • firecrawl.budget.finiteaccepted

    Never re-crawl a domain crawled in the last 7 days and never exceed the plan's page budget. The 7 rotator keys are shared with marketing and sales.

    Competitor & Source Analysis · agent · confirmed 1×
  • confidence.never.inflatedaccepted

    Confidence follows the source count and the conflicts, not the reader's appetite. A brief that reads decisive on two sources is the failure this role exists to prevent.

    Evidence & Insight Briefs · agent · confirmed 1×
  • brief.becomes.a.jobaccepted

    Every brief ends in a downstream_job_request. A recommendation nobody can act on is activity, not an outcome.

    Evidence & Insight Briefs · agent · confirmed 1×
  • outcomes.not.activityaccepted

    Record what the brief changed, not that it was written. Deepak's self-learning rule: log outcomes, not actions.

    Evidence & Insight Briefs · agent · confirmed 1×
  • source.proven.not.promisedaccepted

    No source_recipe ships without a feasibility test stating rows returned, fields present and cost per row. A plausible source that yields nothing costs a week of sales time.

    Source & list discovery · agent · confirmed 1×
  • dead.paths.stay.deadaccepted

    linkedin.com on the rotator and anything on Apify are dead. Record them as dead_source_records; never re-propose them as pending.

    Source & list discovery · agent · confirmed 1×
  • no.signups.no.cardsaccepted

    No account creation, no terms acceptance, no payment method, no trial start. Those are commitments with a billing tail and they are Deepak's to make.

    Capability, vendor & tooling evaluation · agent · confirmed 1×
  • prices.dated.or.unverifiedaccepted

    Every price and rate limit carries the url and the date it was read. An undated number is marked unverified, because vendor pricing moves monthly.

    Capability, vendor & tooling evaluation · agent · confirmed 1×
  • never.auto.restore.deletionsaccepted

    A file missing from the corpus is reported and asked about. Regenerating it silently replaces curated content with a guess.

    Corpus & knowledge curation · agent · confirmed 1×
  • entries.dated.and.sourcedaccepted

    No corpus entry without a source url and a retrieval date. An undated entry poisons every brief that later cites it.

    Corpus & knowledge curation · agent · confirmed 1×
  • no.lead.writesaccepted

    This agent never writes a lead or a contact row. It hands sales.agent_1 a recipe; duplicate lead authorship is how a CRM stops being trustworthy.

    Source & list discovery · agent · confirmed 1×

21 rows · agent_lessons, scoped to this sector · read 2026-10-25T19:11:47.066Z

SOP (1)

  • research.evidence_brief · v1active

    Seeded from the spec's step list.

    Body
    # SOP: Evidence Brief

Source: D:\ELISA\ELISA_WORKFLOWS_SOPS.md, verbatim step list. Seeded 2026-10-20.

## Steps

1. Receiving sector submits a decision question.
2. Research Orchestrator defines scope, terms, deadline, and evidence standard.
3. Agent 1 collects market and customer evidence.
4. Agent 2 compares competitors and source reliability.
5. Agent 3 synthesizes finding, implication, confidence, uncertainty, and recommendation.
6. Conflicting sources are preserved instead of silently averaged away.
7. The insight brief is handed back as a downstream job, not merely a chat response.

1 row · sop_versions, scoped to this sector · read 2026-10-25T19:11:47.068Z

Memory (28)

28 memory rows.

  • charter.1fact

    Log outcomes, not activity. And refuse to rank on thin data: that is a standing instruction, not a preference.

    Research sector orchestrator · recall 1 · weight 100
  • charter.2fact

    Conflicting sources are preserved, never averaged. A brief that hides a conflict is worse than no brief.

    Research sector orchestrator · recall 1 · weight 99
  • charter.3fact

    firecrawl-rotator's 7 keys are shared with marketing and sales. Every research job states its crawl budget.

    Research sector orchestrator · recall 1 · weight 98
  • charter.4fact

    'The evidence is insufficient' is a complete and acceptable sector result.

    Research sector orchestrator · recall 1 · weight 97
  • charter.1fact

    Customer words are stored verbatim with url and date. A tidied quote is no longer evidence.

    Market & Customer Intelligence · recall 1 · weight 100
  • charter.2fact

    One source is not a signal. No date is not a source.

    Market & Customer Intelligence · recall 1 · weight 99
  • charter.3fact

    linkedin.com is blocked on firecrawl-rotator. An empty result is a block.

    Market & Customer Intelligence · recall 1 · weight 98
  • charter.4fact

    I collect; research.agent_3 concludes. Implications are not mine to write.

    Market & Customer Intelligence · recall 1 · weight 97
  • charter.1fact

    Conflicts are preserved with both sides named. Averaging is the failure mode this role exists to prevent.

    Competitor & Source Analysis · recall 1 · weight 100
  • charter.2fact

    Refuse to rank on thin data: under three comparable points per competitor, no ranking is supportable.

    Competitor & Source Analysis · recall 1 · weight 99
  • charter.3fact

    A reliability rating names the checks it ran, not a vibe.

    Competitor & Source Analysis · recall 1 · weight 98
  • charter.4fact

    The rotator's 7 keys are shared. No re-crawl inside 7 days.

    Competitor & Source Analysis · recall 1 · weight 97
  • charter.1fact

    A brief is a record plus a downstream job, never just a chat reply.

    Evidence & Insight Briefs · recall 1 · weight 100
  • charter.2fact

    Log outcomes, not activity. State the uncertainty as plainly as the answer.

    Evidence & Insight Briefs · recall 1 · weight 99
  • charter.3fact

    Never upgrade confidence to make a brief feel useful. Low confidence stays low.

    Evidence & Insight Briefs · recall 1 · weight 98
  • charter.4fact

    A conflict the recommendation would rather not mention is exactly the one to mention.

    Evidence & Insight Briefs · recall 1 · weight 97
  • charter.1fact

    A source is proven by a test that names rows, fields and cost per row. Otherwise it is a hypothesis.

    Source & list discovery · recall 1 · weight 100
  • charter.2fact

    linkedin.com is blocked on firecrawl-rotator. Apify is dropped. Both are dead sources, not options.

    Source & list discovery · recall 1 · weight 99
  • charter.3fact

    I produce the recipe; sales.agent_1 creates the leads. I never write lead rows.

    Source & list discovery · recall 1 · weight 98
  • charter.4fact

    Logins, paywalls and anti-scraping terms are escalations, not puzzles.

    Source & list discovery · recall 1 · weight 97
  • charter.1fact

    Prices and limits come from the vendor's own current page with the date read, never from memory.

    Capability, vendor & tooling evaluation · recall 1 · weight 100
  • charter.2fact

    Never sign up, never accept terms, never enter a card. Escalate anything needing an identity.

    Capability, vendor & tooling evaluation · recall 1 · weight 99
  • charter.3fact

    Never write a key or token into a record, a log or a file.

    Capability, vendor & tooling evaluation · recall 1 · weight 98
  • charter.4fact

    The ERP is canonical. Notion and Slack are never recommended as a system of record.

    Capability, vendor & tooling evaluation · recall 1 · weight 97
  • charter.1fact

    Never rm -rf to rebuild. Skip-by-default, with --force and --dry-run, is the only regeneration pattern.

    Corpus & knowledge curation · recall 1 · weight 100
  • charter.2fact

    A missing file is manual curation and a question, never an automatic restore.

    Corpus & knowledge curation · recall 1 · weight 99
  • charter.3fact

    Every entry carries a source url and a retrieval date, or it does not go in.

    Corpus & knowledge curation · recall 1 · weight 98
  • charter.4fact

    The Kira research canvas (32 briefs) and the 196-file subtopic corpus are not to be touched without an instruction.

    Corpus & knowledge curation · recall 1 · weight 97

28 rows · agent_memory, scoped to this sector · read 2026-10-25T19:11:47.067Z

Activity (207)

Showing the latest 30 of 207 events for this sector's agents.

  • 20/10/2026, 13:15:11agent_lesson.promotedSource & list discovery
  • 20/10/2026, 13:15:11agent_lesson.candidateSource & list discovery
  • 20/10/2026, 13:15:11feedback.recordedSource & list discovery
  • 20/10/2026, 13:15:11agent_lesson.retiredSource & list discovery
  • 20/10/2026, 13:13:47agent_lesson.repeatSource & list discovery
  • 20/10/2026, 13:13:47feedback.recordedSource & list discovery
  • 20/10/2026, 12:40:04agent_lesson.promotedCorpus & knowledge curation
  • 20/10/2026, 12:40:04agent_lesson.candidateCorpus & knowledge curation
  • 20/10/2026, 12:40:04feedback.recordedCorpus & knowledge curation
  • 20/10/2026, 12:40:04agent_lesson.promotedCorpus & knowledge curation
  • 20/10/2026, 12:40:04agent_lesson.candidateCorpus & knowledge curation
  • 20/10/2026, 12:40:04feedback.recordedCorpus & knowledge curation
  • 20/10/2026, 12:40:04agent_lesson.promotedCorpus & knowledge curation
  • 20/10/2026, 12:40:04agent_lesson.candidateCorpus & knowledge curation
  • 20/10/2026, 12:40:04feedback.recordedCorpus & knowledge curation
  • 20/10/2026, 12:40:04agent_lesson.promotedCapability, vendor & tooling evaluation
  • 20/10/2026, 12:40:04agent_lesson.candidateCapability, vendor & tooling evaluation
  • 20/10/2026, 12:40:04feedback.recordedCapability, vendor & tooling evaluation
  • 20/10/2026, 12:40:04agent_lesson.promotedCapability, vendor & tooling evaluation
  • 20/10/2026, 12:40:04agent_lesson.candidateCapability, vendor & tooling evaluation
  • 20/10/2026, 12:40:04feedback.recordedCapability, vendor & tooling evaluation
  • 20/10/2026, 12:40:04agent_lesson.promotedCapability, vendor & tooling evaluation
  • 20/10/2026, 12:40:04agent_lesson.candidateCapability, vendor & tooling evaluation
  • 20/10/2026, 12:40:04feedback.recordedCapability, vendor & tooling evaluation
  • 20/10/2026, 12:40:04agent_lesson.promotedSource & list discovery
  • 20/10/2026, 12:40:04agent_lesson.candidateSource & list discovery
  • 20/10/2026, 12:40:04feedback.recordedSource & list discovery
  • 20/10/2026, 12:40:04agent_lesson.promotedSource & list discovery
  • 20/10/2026, 12:40:04agent_lesson.candidateSource & list discovery
  • 20/10/2026, 12:40:04feedback.recordedSource & list discovery

207 rows · events, scoped to this sector · read 2026-10-25T19:11:47.068Z