Welcome to CIRE, the Continuous Improvement Research Engine.

Creating your first project:
1) Click + New Project in the top bar (or + Create First Project if you have none yet).
2) Fill in the project fields. See Research Cycles for what each field means.
3) Select the project from the sidebar dropdown to make it active.

Orientation: The sidebar groups are Research Climate (synthesized findings), Discovery (active investigation), Memory (disproven paths), Governance (self-check and review), and Control (sources, cost, transactions, and project management).

Your account balance appears at the top of the sidebar. New accounts start at $0.

Every clickable control has a ? widget for quick contextual explanation.

A research cycle is one round of investigation on a single hypothesis: CIRE searches real evidence, reasons over it, and updates belief state.

Project setup fields:
Name, Domain, End state, Why, Constraints, Excluded, and Budget.

Running cycles: Click Run Queue (or use auto-cycles when enabled). Each cycle formulates a query, searches real academic sources, reasons over retrieved evidence, then updates confidence and records findings.

Daily pacing and cost safety: Auto-cycles are capped per project per day by the auto-cycle setting (default 5/day), and all runs still respect daily query and dollar limits. This is intentional cost/pacing control. If your queue appears to "run out" for the day, that usually means the daily cap was reached, not that anything is broken.

Domain determines source mix automatically (for example arXiv, Semantic Scholar, OpenAlex, PubMed Central, CORE, Unpaywall, bioRxiv, medRxiv).

Grounding labels:
Grounded = evidence-backed, Partial = incomplete support, Ungrounded = no supporting evidence found.

Confidence reflects current support strength. Fitness reflects longitudinal performance used for prioritization. Both can move up or down as new evidence arrives.

Use Add Source or Import Content to include lawful user-provided material when needed.

Active: Runs normally and can execute cycles (balance permitting).
Closed (Frozen): Stops new cycles and cost accrual while preserving all state; reversible via Reopen.
Soft Deleted: Hidden from main list and moved to Recently Deleted; recoverable via Restore.
Pending Hard Delete: Starts a 15-day countdown. Recoverable via Cancel Deletion during countdown. After expiration, deletion is permanent and irreversible.

You can export a project as raw JSON or PDF report regardless of lifecycle state.

CIRE uses real compute resources. Funding is a prepaid balance model through cost-recovery donations to EM Foundation (nonprofit), not subscription purchase semantics.

New accounts start at $0. Browsing is free; running new cycles requires balance.
Click + Add Funds to donate via Stripe and credit balance immediately.

Completed cycles deduct actual observed cost. GLM verification and quantum execution are itemized separately in transaction history.

If balance runs low, projects pause automatically, you receive email notification, and execution resumes after top-up.

GLM verification is an optional second independent AI pass at project level.

When enabled, GLM performs its own retrieval and reasoning for the same hypothesis, producing an independent assessment rather than a rubber-stamp review.

Use it for higher-stakes or contested topics where independent agreement matters. It is off by default because it generally increases cycle cost by running a second model path.

In Verification Network and Adversarial Agent outputs:
Agree = aligned conclusions, Partial = partly aligned, Disagree = meaningful split with rationale and evidence differences surfaced.

Quantum is used only for genuinely quantum-suited questions and never triggers automatically without explicit user approval.

Flow:
1) Conservative recognition during cycle reasoning.
2) Structured approval request: what was found, what was hypothesized, why classical methods are insufficient, and what quantum would rule in/out.
3) User decision to approve or decline (decline has no penalty).
4) If approved and funded, submission to IBM Quantum or AWS Braket.
5) Real result retrieval and clear quantum-informed labeling in findings.

Current execution is labeled clearly where applicable (for example Amazon SV1 simulator), with transparent itemized settlement in transactions.

Governance features keep research auditable and self-correcting:

Review Queue / Human Review Board: escalated items requiring direct user judgment.
Adversarial Agent: actively challenges findings to surface weaknesses and disagreement.
Wrong Goal Detector: flags drift from declared End State.
Research Constitution: integrity rules for evidence grounding and honest labeling.
Continuity Receipts: preserved reasoning trail so resumes do not lose context or repeat dead ends.

This page shows both the real retrieval sources CIRE is configured to use and the live evidence counts currently stored in production. Search and filter the integrated sources below, then compare that reference against actual historical usage.

Live evidence inventory is pulled from the real production sources table. Zero means the provider is configured but has not yet produced any stored retrieval rows.

Checklist status: 1) Per-row table actions (hypothesis/source/transaction): Deferred as N/A where rows are read-only and contain no action buttons in current UI build; automatic button binding will cover any row actions if introduced. 2) Modal form field-level widgets: Covered for modal field groups (.fg). 3) Review-board decision buttons: Covered (modal decision controls bind automatically). 4) Long-tail section action controls: Covered via route + mutation rebind.

Pending rollout: none in mechanism scope. Remaining work is guide content authoring and editorial tuning.