ForagePal
Forecast transparency

Stage A · model card v1

Fruiting conditions, explained honestly.

ForagePal ranks whether recent and forecast environmental conditions support mushroom fruiting. It is a deterministic, rule-based planning aid — not a probability, an identification, a habitat confirmation or food-safety advice.

Rule-basedNot calibratedCounty-level14 calendar days

01 · Intended use

A conditions outlook, not a find prediction

The service ranks whether environmental conditions are supportive of fruiting across the 32 counties of Ireland and Northern Ireland for fourteen Europe/Dublin calendar days. Its score is a bounded conditions index. It is not the probability that a mushroom is present and it is not a confidence percentage.

The outlook does not identify a specimen, confirm habitat at a location, assert that fruit bodies are present, estimate yield, determine access rights or provide an edibility verdict.

The question it answers

“Are the broad recent and forecast conditions supportive of fruiting for this release-approved species in this county?”

02 · Method

A fixed, versioned path from inputs to snapshot

The server runs the same deterministic Stage-A rule engine as the portable app implementation. Every published manifest pins the content, model, weather and boundary versions that produced it.

  1. 01

    Verify the release

    The publisher checks the content bundle and its approval evidence against their SHA-256 manifests, then confirms that both belong to the same release.

  2. 02

    Assemble county conditions

    Met Éireann forecasts and ancillary observations are combined with versioned normal inputs. Continuous daily series are averaged across each county before any thresholds or rolling features are calculated.

  3. 03

    Apply deterministic rules

    The Stage-A engine combines published phenology, ecological guild, moisture, air and soil temperature, dry-spell, incubation, VPD, heat, cooling and photoperiod modifiers.

  4. 04

    Publish an immutable snapshot

    Complete county, species and spatial files are written first. The manifest is the commit marker, and the current pointer advances last so an incomplete candidate cannot become current.

03 · Scores and states

Every output says what kind of answer it is

A scored result uses a 0–100 conditions index and a 0–4 display band. The labels describe environmental support only; even “Excellent” is not a promise that a species is present.

  • Very Low

    0–19

  • Low

    20–39

  • Moderate

    40–59

  • High

    60–79

  • Excellent

    80–100

scored

A conditions score is available

The response carries a 0–100 index, one of the five bands, and short categorical explanations.

unavailable

A required weather input is missing

The service says that the result is unavailable. It does not substitute zero or present a seasonal value as current weather.

not_scored

The model deliberately withheld a score

For example, a catalog record without release-bound production approval remains visible but is marked content_not_approved.

04 · Inputs and provenance

Versioned sources, explicit approval

Raw weather, numeric factors, intermediate triggers and formulas stay server-side. The client receives categorical explanations and the version identifiers needed to trace a result.

Weather and normal inputs

Met Éireann point forecasts and ancillary observation feeds supply precipitation, air and soil temperature, humidity, radiation and related inputs. Versioned climatology and normal inputs support the lower-information tail of the horizon.

Published content and release evidence

The model reads the atomically published mushroom ecology and phenology bundle only after its bundle and approval-evidence hashes, record set, decisions and warning disposition have been verified. Release approval does not mean expert review; the service claims specialist review only when the evidence actually records it.

County boundaries and sampling

The fixed boundary roster uses 43 canonical sample points across 32 counties. Larger counties use two or three points. Republic of Ireland boundaries derive from Tailte Éireann’s 2019 Generalised 20m county boundaries under CC BY 4.0; Northern Ireland boundaries derive from OSNI/Land & Property Services Open Data County Boundaries under OGL v3.0.

05 · Limitations

What the outlook cannot establish

  • County values are spatial summaries, not a site-level habitat check or a precise foraging location.
  • The outlook covers the fruiting half of Habitat × Fruiting. It does not confirm a compatible host, substrate, woodland, access permission or habitat at a site.
  • The environmental rules have not been calibrated against observed fruiting frequency.
  • Forecast uncertainty grows with lead time and with the availability of weather inputs.
  • Met Éireann supplies roughly the first ten days. Days 11–14 use versioned climatology and normal inputs, so they are lower-information outlooks.
  • Seasonal priors describe typical windows and may not capture an unusual early or late season without supporting environmental conditions.
  • Microclimate, recent picking, disturbance, detectability and the presence of fruit bodies are outside the service’s evidence.
  • Weather warnings and local safety guidance remain authoritative and should be checked independently before travel.

06 · Operations and change control

Fail closed, keep the last known good result

Production publication is scheduled every six hours with bounded retries, a dead-letter queue, two-week logs, error and duration alarms, handled-reader-error monitoring and a stale-publication alarm.

A systemic upstream failure cannot replace a useful snapshot with an entirely unavailable candidate. Partial county failures may publish only when affected cells explicitly sayunavailable.

Any rule change must increment the model version and pass direct regression tests against the previous Stage-A semantics. A future calibrated model requires a new model card and validation evidence; it cannot reuse this page’s calibrated: falsecontract.

Change history

Stage-A production ownership

The existing Stage-A semantics were preserved while weather and versioned normal inputs moved behind authenticated immutable snapshots, with release-evidence gating, atomic publication and explicit unavailable and not-scored states.