Frequently asked questions

It is a workbench that supports a restoration project across its whole lifecycle, in one auditable place:

• Define a project Area of Interest from an uploaded boundary file, and run site characterization across roughly three dozen public datasets (soils, terrain, hydrography, flood, wetlands, streamflow, impaired waters, wildfire, species, and more).
• Quantify outcomes and model the economics: realizable funding-program revenue as NPV, IRR, and payback, with Monte Carlo bands and a literature-based total-economic-value range.
• Build regulatory checklists and generate deliverables (the CEQA and NEPA series, state little-NEPA documents, and permit and consultation packets).
• Run pre-award pursuits: read a solicitation, draft an SOQ or proposal, and price a fee.
• Ask a grounded AI assistant that retrieves evidence from the project and cites it.

The original restoration-ROI workbench is now one area among several. The mission behind all of it is to accelerate ecosystem restoration by giving practitioners faster, more defensible analysis.

Restoration sponsors and developers (land trusts, NGOs, public agencies, and tribes) who put projects together and have to defend them; AEC and environmental consulting firms; and the funders and agency reviewers who receive the deliverables. Within a project, the discipline practitioners (site assessment, permitting, modeling, design, implementation, monitoring) each have a place to work. Workspace admins manage billing, team, and keys.

No. The project assistant supports human practitioners. It drafts, analyzes, retrieves evidence, and cites sources, but it never self-certifies or renders final engineering judgment. The engineer of record decides, and counsel decides on regulatory matters. It is not a virtual staff and not a substitute for a licensed professional.

The platform covers the contiguous United States. An Area of Interest outside CONUS is rejected at upload. Most data sources are national. Some are state-specific, including California Fire Hazard Severity Zones, EnviroStor and GeoTracker contamination, water-rights points of diversion (California and Oregon), and California HCP/NCCP and salmonid layers. Regulatory-agency mappings and little-NEPA deliverables are deepest for California, Washington, and Oregon, and the deliverable catalog adapts to the project's state.

For each outcome the engine finds eligible funding programs with a payment rate, computes quantity at full uplift × dollars-per-unit × attainment probability, applies anti-double-payment stacking rules, distributes the resulting realizable value over the project timeline as cashflows, and discounts to NPV. Water and habitat quantities are computed server-side from structured inputs rather than typed directly, so the number is reproducible from its inputs.

Realizable value is money a project can actually be paid by funding programs. It flows into NPV and IRR. TEV is the broader societal benefit estimated from published valuation literature, shown as a transparency range grouped by method (willingness-to-pay, hedonic, replacement cost, avoided damage, market price).

TEV is never added into the realizable NPV, and its per-method subtotals are never summed across methods, because the methods overlap and summing them double-counts. The two numbers answer two different questions: what a project will be paid, and what public-good value it produces that no program reliably pays for.

No. The single-point NPV is a screening estimate, and it is labeled as one until you run Monte Carlo. The Monte Carlo run samples rates, attainment, and program persistence to produce confidence bands (P10 to P90), a probability that NPV is above zero, and a tornado view showing which inputs drive the variance. Byte-reproducible output requires a seeded run; the default interactive run is not seeded.

IRR only exists when the cashflow series changes sign in a way that yields a real root. Front-loaded or all-positive or all-negative series have no real IRR, so the engine returns nothing rather than a misleading number. Payback shows "Never" when cumulative cashflow stays negative across the modeled horizon.

A rule cannot be checked because the project is missing the data it needs, for example a county or scale rule when no county or acreage has been entered. Fill in the project field, or apply an eligibility override with a written justification. Sponsors override eligibility and rates at the project level; the underlying program catalog is curated by Orenna.

A run executes around three dozen data-assembly stages spanning soils (SSURGO), land cover (NLCD), terrain (3DEP), hydrography (NHDPlus HR), watershed and flood-frequency (StreamStats), flood hazard (FEMA NFHL and FIS), wetlands (NWI), streamflow (NWIS), water quality (Water Quality Portal and ATTAINS), climate (PRISM), wildfire, species and critical habitat (USFWS and NMFS), dams, aquatic barriers, levees, water rights, point sources, contamination, cultural resources, imagery, parcels, and infrastructure. The exact live list is shown in the run's "Public data assembled" panel.

Usually no. The contributing basin auto-delineates from the Area of Interest using USGS StreamStats. Upload a watershed polygon only when your state has no StreamStats delineation grid (Florida, Michigan, Nebraska, Texas) or when you want to override the auto-delineation. You can also add an annual-exceedance-probability discharge CSV and BRAT/R-CAT reach layers as additional hydrology inputs.

There is no difference. Each project has exactly one geometry, always called the Area of Interest (AOI), and it is the project boundary. The map renders it as a single AOI layer. There is no separate project-boundary layer or field.

From a completed run you can download a summary briefing PDF, a full report PDF, a Map Atlas PDF, a geospatial ZIP (one GeoJSON per data type plus findings as JSON and CSV, a README, and an ArcGIS GeoPackage when the geospatial worker is running), raster cloud-optimized GeoTIFFs (3DEP, NAIP, NLCD), the delineated watershed GeoJSON, and the raw bytes of each public dataset that was queried. All coordinates are WGS84 (EPSG:4326).

No. The synthesis stages produce a historical narrative, a triage of candidate intervention types, and an executive summary, all as assistive analysis built from the retrieved evidence. They are decision support, not autonomous design. Your team decides what to build.

The assistant holds the whole project picture. It works across project management, site assessment, modeling and analysis, design, permitting, implementation, monitoring, and pre-award capture using one shared project memory, decision log, corpus, and toolset. Ask about a site finding, a scenario assumption, a regulatory gap, a design surface, or a pursuit requirement, and it answers from the project evidence with citations. It can draft, analyze, run read-only tools, and propose memories, but consequential project changes still require human confirmation.

It grounds every project-specific claim in retrieved evidence and cites it inline. Any clause it could not cite is flagged as unverified so you can challenge it. Retrieval is hybrid keyword and semantic search over the project corpus (canon documents, metadata, outcomes, scenarios, decisions, and characterization findings). Permit and regulatory applicability comes from an encoded regulatory matrix rather than general knowledge, and web results are explicitly marked as lower-trust than your own filed documents.

It only proposes memories, each with a citation, and nothing is saved until you click "Save to memory." Confirmed items live in the Memory pane, where you can pin or retract them. Beyond that, the assistant uses read-only tools today. It retrieves, drafts, and proposes. It cannot mutate project state or take external actions on its own.

Memory is small reusable knowledge (facts, preferences, conventions, open questions) the assistant proposes and you confirm, then re-injects into future prompts. The Decisions log is your formal record of project decisions (what was decided, why, and the alternatives), authored on the Decisions tab and indexed into the corpus so chat can cite the reasoning behind a choice. Uploaded meeting transcripts can produce candidate decisions that you accept or reject.

Dozens of templates, including the full CEQA series (NOE, IS-MND, EIR, NOP, Final EIR, Findings and Statement of Overriding Considerations, NOD); the NEPA series (CatEx, EA, FONSI, NOI, EIS, Final EIS, ROD, public notice); state little-NEPA sets (Washington SEPA, New York SEQRA, and more); permit and consultation packets (§404 NWP27 pre-construction notification, Section 7 biological assessment, CDFW 1602 packet, FEMA CLOMR, basis of design); and pursuit documents (SOQ, technical proposal, SF330). State-specific templates appear based on the project's state.

A gap is a requirement that applies to your project (or is triggered by its resource context) but is not yet satisfied in the project record. It is an applicable-but-open item your team needs to scope or produce. Requirements that do not apply are short-circuited and never show as gaps. Many requirements link directly to a deliverable you can generate.

No. Deterministic sections are pure functions over the project record and make no model calls (they carry a "No model calls" chip). Prompt sections run a grounded Claude call against declared retrieval scopes and cite their sources. A section that fails grounding lands as needs review, and the way out is a human one: approve as-is or replace the text, either of which locks the section.

Finalizing seals the document body (or the item statuses, for a checklist) into a signed, hashed manifest and locks it. You can still export a finalized report to PDF or DOCX, but to change it you create a new report or regenerate the checklist, which supersedes the prior one. This keeps a clean, defensible record of what was issued and when.

Upload an RFQ or RFP to the project Data Room, start a pursuit, and an extraction job (it takes a minute or two) pulls out the scope summary, weighted evaluation criteria, submission rules, mandatory qualifications, and individual requirements. From there you generate a draft SOQ (the RFQ path), a technical and cost proposal (the RFP path), or an SF330, run a compliance check, and track win or loss. If a scanned PDF cannot be parsed, a recovery panel lets you re-run extraction, paste text, or add requirements by hand.

No, by design. SOQ, proposal, and SF330 sections cite strictly from your org-scoped Assets and are instructed never to invent people, titles, certifications, or projects. Win themes must cite Assets evidence for any claimed strength, and a high-weight evaluation criterion with no supporting evidence is surfaced as a gap or a teaming need rather than papered over.

It is professional-services labor, not construction cost: hours by role across phases and tasks, priced against your organization's billing-rate schedule, plus materials and other direct costs with an optional markup. The computation is deterministic and shared by the fee editor and the proposal's cost section, and any role used without a rate is flagged so nothing is silently underpriced. Treat the result as planning-level, a basis to confirm against your fee schedule before submission.

Assets is org-scoped, reusable business-development content (past projects, personnel, certifications, references, boilerplate), kept deliberately off the project chat corpus to preserve tenant isolation. A proposal build resolves its organization and retrieves only that org's assets, so qualifications never leak between organizations and your firm's institutional memory survives across projects and staff turnover.

Pricing is organized into five tiers: Free, Solo, Practice, Firm, and Enterprise. The pricing page carries the current rate for each. The model is deliberately simple: tiers differ on exactly two axes — how many projects you can run and your included monthly AI allotment. Everything else — the full calc engine, characterization, the complete literature corpus, CSV/JSON/GeoPackage export, white-label deliverables, bring-your-own-key, and unlimited seats — is the same on every tier, including Free. We never gate the methodology or the data quality.

The free tier gives full access on one project for 30 days, after which the project auto-freezes to read-only: still fully viewable and exportable, but no new AI work (no new characterization runs, chat turns, or re-synthesis) until you upgrade. Upgrading unfreezes it in place with no re-characterization and no data loss. It is never deleted on the 30-day clock.

An optional add-on for teams doing hydraulic and design engineering: HEC-RAS model interpretation (results extraction; derived depth, velocity, shear, and WSE-difference rasters; run comparison) and the grading design workbench (drawn intervention features, criteria-driven surface generation with an honest exception report, quantity takeoff, download to RASMapper). It has the same price and content on every tier — it does not change the two axes tiers differ on — and it gates only the creation of engineering products: anyone in your organization can view and download what the package produces. Uploading and organizing RAS models is included on every plan without it. See the Help guide for the full workflow, and email sales@orenna.io to add it to your organization.

No. Every tier includes unlimited seats, including Free. Price scales only by how many projects you run and your AI allotment, never by headcount. When you need more active-project capacity you add it with project-packs rather than jumping a tier. Roles within an organization are admin, editor, and viewer, managed under Settings → Team.

Yes to both controls. Bring-your-own-key is available on every tier (and is the default on Enterprise). An org admin adds it under Settings → API keys; it is validated once, encrypted, and never shown again, and calls on it bill to your own account and bypass the platform meter. Otherwise, platform AI usage is metered in AI credits against per-tier allotments and any budget caps you set (user daily, project monthly, org monthly). Approaching a soft cap downgrades the model rather than cutting you off.

Not through a public guest link yet; there is no public share-link feature in the product today. Collaboration is by inviting people into your organization (Settings → Team) with a viewer role, and any finalized deliverable can be exported to PDF or DOCX to send directly. Read-only guest links are on the roadmap.

Trust is built into the surfaces. Chat answers cite their sources inline and flag uncited clauses as unverified; permit answers route through the regulatory matrix rather than general knowledge. Report sections record their citations and retrieval trace, deterministic sections make no model calls, and grounding-failed sections require human approval. Every chat turn produces a signed audit manifest (queries, chunks retrieved versus used, tools called, citations), and finalized reports and checklists seal a signed, hashed manifest. Throughout, the AI assists your human leads and never self-certifies.

An org admin controls this under Settings → Privacy: a retention window for the body of each Claude call (default 30 days, after which bodies are stripped to hashes), a zero-data-retention toggle that stores new calls hash-only, and an allow-platform-key toggle that, when off, requires bring-your-own keys and fails calls closed. The Privacy page shows how many audit rows still hold bodies versus have been stripped.

Yes. Any user can download a signed ZIP of their personal data, and an org admin can download a signed ZIP of the whole organization. Erasure is self-service with a cancellable grace window: account erasure anonymizes you and deletes personal rows, and org deletion (admin only) hard-deletes the tenant after the grace window.

Each project has an Activity tab: a per-project audit trail of create, update, delete, run, retry, and cancel events across projects, outcomes, scenarios, site characterization, and documents, with field-level change diffs and the acting user. It is separate from the AI usage and spend dashboards under Settings.