# Synthetic Sentinel research examples

Every property value and point observation in these files is a **synthetic test
fixture**. These are not real assessor, FEMA or customer records, and the sample
address/parcel has not been verified to exist. Official URLs demonstrate the
reference shape; they do not substantiate the fictional values. Replace the
entire locator before making an authorized live request.

| File | Purpose |
| --- | --- |
| `sentinel-research-request.json` | Property + independent point + two caller scenarios |
| `sentinel-research-compact.json` | Complete claim evidence with no repeated source-detail block |
| `sentinel-research-full.json` | Same claims/evidence plus bounded source details |
| `sentinel-research-offer-compare-request.json` | Two prices and concessions, both with unknown mortgage payoff |
| `sentinel-research-offer-compare-response.json` | Actual deterministic service output over synthetic inputs |

The response files were generated with `build_research_packet` and validated
against `ResearchPacketResponse`. Optional source functions returned typed
synthetic fixtures; no public source or production account was queried. They
target the API 1.5.0 research contract. Check deployed OpenAPI before making a call.

For the compact/full pair, the same values yield 10 reported facts, two
calculations and 20 references. Facts, calculations, identity, references,
unknowns, questions, conflicts, section states, attributions and limitations
are equal across projections. Only projection metadata, generation time and
additional source details differ. Normal serialized response sizes are 18,805
and 29,410 UTF-8 bytes respectively; these indented downloads are larger. This
36.06% reduction is one controlled fixture result, not a model-token or cost claim.

In the two-unknown-payoff example, both `net_proceeds` values remain null. The
intermediate amounts are explicitly estimated proceeds **before mortgage
payoff**, using the listed caller assumptions. Prices are not valuations, and
this arithmetic does not compare contingencies or other contract terms.

Keep reference IDs with every factual assertion and reproduce required notices.
An official-source observation is not independent verification of current
conditions. The sample coordinates remain caller-supplied and unverified against
the property. Missing source data never proves an absence or a safe condition.

The n8n and server-side Python starters are in `docs/integrations` in the source
repository. They contain no API key. The existing integration pilot and customer
AI usage are separate purchases; these examples do not announce a new plan.