Trust posture

How we report what we know.

Water Hawk surfaces four classes of content: live data records (bids, utility scoring, candidates and clusters), capability offerings (the Salian Defense stack and the bespoke engagements), stat claims that summarize the market, and cohort-aggregated outputs (lane benchmarks, supplier-share concentration). Each class carries a specific discipline. The discipline is documented below so any reader can calibrate the strength of any claim they encounter on the site.

1. Live data records: verified or in review

Every live record we surface (a bid, a candidate, a cluster) carries one of two quality badges. Verified means the record has been reconciled against the actual source document. In review means the buyer or site is matched and the lane is matched, but the record has not yet reconciled fully against the source document (typical gaps are exact dates, line items, or document references).

A separate quality dimension tracks the source itself: Direct Municipal, State Portal, Verified Aggregator, or Unverified (which is filtered out at the consumer surface). Subscribers see both dimensions on every record; the public preview surfaces the verification badge with a tooltip carrying the source quality.

A defensive filter at the data-access boundary drops records that are logically impossible (posted dates in the future, closed status with overdue dates, plausible records with no source quality). Anything filtered is also flagged upstream for fix. See the bid-corpus ingest feedback in our v1.0 engineering ask: V1_BID_INGEST_FEEDBACK.

2. Capability claims: engagement-scoped, not product-shipping

The Salian Defense backbone behind Water Hawk is delivered through scoped engagements, not off-the-shelf product subscriptions. Each capability layer (hardware OT encryption, synthetic-content detection, continuous regulatory posture, managed orchestration, post-quantum cryptography, threat-intelligence fusion) is described in plain language on /defense and specifically marked “scoped per engagement.” The implementation, named components, and integration detail are confirmed in writing during scoping before any deployment begins.

The engagement tiers (Signal subscription, Assess fixed-price posture review, Protect managed defense) are real work, not service-catalog entries. Tier I (Water Hawk subscription) ships through the platform; Tier II is bounded and fixed-price; Tier III is custom-scoped per environment with no published list price. Beyond Defense, the broader Salian relationship extends into custom AI integration, custom software, and security engineering, each custom-scoped per engagement.

3. Stat claims: published sources where applicable

Where a stat originates from a public federal source, the source is named inline (e.g., “29,766 U.S. community water systems, EPA SDWIS”; “$49B SRF pipeline, EPA Needs Survey”). Verbs are calibrated to what we actually do: “indexed” for records we have ingested, “scored” for utilities the methodology has run against, “tracked” for ongoing monitoring. We avoid “analyzed” or “monitored” unless the downstream analysis or monitoring is actually wired and continuous.

Operator-network observations (e.g., indicative P3 delivery acceleration, indicative rate-base avoidance) are reported as ranges with a clear “Indicative” prefix and a methodology footnote that names the underwriting basis. Specific per-engagement values are scoped in the bespoke deliverable, not in the public report.

4. Cohort-aggregated outputs: antitrust safeguards

Chemical-lane benchmarks, supplier-share concentration, and lane price history are cohort-aggregated outputs. Three safeguards govern the aggregation, enforced at the query layer:

• k=5 minimum cohort. No cohort-derived value is published unless at least five distinct contributors sit inside the aggregation.
• 25% concentration cap. No single contributor accounts for more than 25% of the aggregated value; values that would exceed the cap are clipped and the clipping is flagged.
• 90-day forward delay. Forward-looking cohort outputs respect a 90-day delay so the platform never functions as a real-time price-signaling channel.

Public-facing lane spend figures are additionally banded to the nearest $50M so magnitudes read without exposing operator-curated illustrative numbers verbatim. The chemical-lane public index at /topics/lanes documents this; the long-run replacement of illustrative lane data with real contributor-aggregated benchmarks is on the v1.0 engineering roadmap (see Part E of the system audit handoff).

5. When the discipline fails

We get things wrong. If you encounter a claim that does not match a source you can verify, a record badged Verified that fails reconciliation against the portal, or a stat that is overstrong relative to what we can actually ship, write to [email protected] with the URL, the field, and what you found. We will investigate, correct where reasonable, and credit the finding in the changelog where the disclosure is structural. Subject-matter-expert spot-checks are how we have caught the largest issues to date.