Verification Report: Thorn-Signature Habitat Mapping (Hedgehog Proxy) and RF7 Event, Multi-Timeline Coherence , 27 September 2025

Just after 7 AM, the HQ mug exhaled steam from one lip and cold vapor from the other; the building sensors logged both seasons in the same breath. A few minutes after 8, an email arrived wearing a timestamp from nine minutes in the future and a subject line that hadn’t been written yet. By half past 9, the traffic controller at Jefferson and Vernor performed yesterday’s schedule in TL2 and TL4, while TL0, TL1, and TL3 stuck to today like there was paperwork involved.

Then the radio channel commonly labeled “Band 7” raised its volume. Even at indoor levels meant for quiet offices, ductwork trembled.

Observation: just before 7:43 PM UTC, a clock-echo arrived seven seconds early and seven seconds late; the signal fenced itself off by 47.2 decibels and acted withdrawn under interrogation. After denoising, the waveform carved into thorn-like geometry, repeatable scatter consistent with bramble and hedgerow. Downstream labels called it “vegetation,” but the learned embedding insisted on “shelter that resists approach.” The server log wrote, “All instances are truthful,” which was accurate in one timeline and theatrical in two.

Observation: the data path held steady across instances: Sentinel-1 radar and Sentinel-2 multispectral revisits landed, a learned Earth model (TESSERA) moved pixels along channels aligned with real rivers, and the audit trail could be followed in each Detroit. TL2 alone showed a +0.3 offset in the motion field; it looked small on paper and still shaved confidence along tile borders when sun angles were low.

The classifier stayed deliberately simple. Observation: a logistic threshold guarded the entry and nearest-neighbor voting decided; no generator tried to invent wildlife where sensors saw noise. Action: thermal load stayed low enough to run on a phone during an afternoon survey without raising more heat than a warm pocket.

Ground proofing refused to drift. Observation: human pins from iNaturalist nailed down locations, and a Cambridge walking survey left GPS breadcrumbs where feet actually went. In real use, one line kept repeating: about twenty seconds to the first confident hit when the tile truly held bramble.

So what. Over roughly a decade, European hedgehogs fell by about two to five out of ten. Night walks blink, miss, and demand hours that most crews can’t afford. Satellites keep their dates. The system took a fair trade: stop chasing the animal; find the thorn rooms it sleeps in and the thorn edges where it eats. Say “proxy,” specify where it works, and then use it.

Observation: the code promised “locate hedgehogs,” reality delivered “locate large, uncovered bramble canopies.” Performance peaked when the canopy drew a clean overhead signature. Under trees, confidence sagged. Not memory failing, geometry failing. From above, shelter turns into shade.

Observation: Sentinel-2 composites using bands 8A (near-infrared), 11 and 12 (shortwave infrared) sketched bramble crisply when the sun sat low enough that the zenith angle stayed near twenty degrees; once pushed past the mid-thirties, the sketch smeared. Under city tree crowns, the signal-to-noise ratio slipped under eight decibels; that is the sensor admitting, “something is there,” without promising what. False positives clustered along gridded streets; dark pavement echoed the pattern and the model, as designed, treated repetition as evidence.

Observation: between 1,620 and 1,680 nanometers, reflectance flattened by three to five percent under overlapping leaves. TESSERA read the flattening as hesitation, not emptiness. The map slowed exactly where the animal needs decisiveness.

Work shifted. Crews stopped patrolling at midnight and started checking at lunch. Around Milton Community Centre and Milton Country Park, high-confidence tiles matched bramble that could be pointed out with a glove. One residential hotspot sent auditors to a vacant lot: waist-high thorns invisible to old imagery. North of the river, positives aligned with Bramblefields Local Nature Reserve, living up to its name. The task moved from “see one animal once” to “confirm habitat in bulk.” Budgets can follow tiles instead of anecdotes.

A question came from the loading bays behind Eastern Market in TL1: “Why not just find the hedgehogs?” Field note reply: “Because the city blinks every five days with ten-meter pixels; animals don’t hold still that long.” (TL3 asked the same an hour and a half later and received the same sentence, sounding a little more worn.)

Action: keep the stack small because the phone is the field office. In three timelines, on-device active learning already runs: a person taps yes/no, the model updates locally, and the confidence map redraws on the handset. Crews get guidance where thorns actually prick, and the battery accepts the arrangement until early afternoon.

Observations and actions across timelines:

• Observation: the seven-second timestamp echo passed audit only by pretending the missing three seconds had already happened; explanation attached in TL0–TL4 logs.
• Action: a notch filter suppressed the 2.6 GHz/2.4 GHz intermodulation ghost that leaked through building vents; it is installed in TL1, waiting on parts in TL2, and retired in TL4 after a hardware swap.
• Observation: TL2’s +0.3 drift softened votes on morning overpasses; action: per-orbit recalibration after 10 AM, tied to human pins, restored normal behavior.
• Action: the app enforced privacy: cache updates while offline, sync only on trusted links, and no raw coordinates leaving a tile; servers asked for shortcuts and were denied by design.
• Observation: the bridge to Belle Isle refused to forward packets after sunset unless addressed in MIL-CAN 2.0 frames; the daytime scheduler kept being ignored by that controller.

Absurd yet repeatable: at 847 MHz, the server rack declared an identity crisis with continuous WHOAMI interrupts until a checksum was acknowledged and a small standby battery was seated; facilities rolled eyes, the spectrum log wrote it down.

The ticket queue did its work. A user in TL3 reported “reality not responding”; a restart changed nothing and the issue went to Facilities (Metaphysics). The thorn-denoiser v1.3 produced thorns even more convincingly, which is technically a pass. Asking TESSERA “what’s wrong?” across three timelines returned four answers, one apology, and a meeting invite from last week.

Policy, translated for action: funders should require plain-language vantage limits in every release. Publish error tables split by canopy class. Label outputs as “high-confidence exposed bramble” versus “candidate under canopy.” Two labels, two field workflows, fewer hours lost to green misdirection.

Anomalies persisted. The controller at Jefferson and Vernor showed up as both replaced, still in transit, and installed upside down, depending on timeline and which way the printed arrows pointed. Sentinel’s next pass happened both yesterday and next week; the images matched except where the shadows argued.

Curbside explainer in Cambridge’s market square: “It’s just bramble.” Response: “It’s a map of where hedgehogs can still live, drawn by satellites that can’t see under trees. Your steps fill the blind spots. That’s the work.”

Final observation: satellites measure reflected light, not animals; they do best in open sky with sun angles near twenty degrees, they falter under crowns, and they become reliable when their blind spots are named, published, and patched with ground truth.

Filing