Declared scan: crypto-universality sweep — displacement + streak on SOL/BNB/XRP

Date declared: 2026-07-25, before any computation. Split: Research only (SOLUSDT 2020-08→2022-12 · BNBUSDT 2017-11→2022-12 · XRPUSDT 2018-05→2022-12, all fetched/validated 2026-07-25). Layer 3, Edge Lab, one ledger row per instrument. Purpose: map the two established crypto effects across the asset class — turning "two instruments" into an honest universality picture for both the displacement event class (research object) and the streak candidate (live, look-2 pending). Signal mapping only: none of these venues has a sourced cost model, so nothing here can license an execution test — harvest viability is a separate, later question per Findings A1/E5.

Conditions (all constants frozen, identical to the prior declarations)

Family 1 — displacement continuation (frozen displacement-repro definitions: baseline ≥ 7bp, body ≥ 60%, compression ≤ baseline, range ≥ {4,5}× baseline; signed by bar direction): disp4, disp5 × horizons 15/30/60 min = 6 hypotheses/instrument.

Family 2 — streak persistence (frozen rhythm-scan definitions: freshness == threshold, disjoint sets): up_ge5(+1), up_ge8(+1), down_ge5(−1), down_ge8(−1) × horizons 15/30/60/240 min = 16 hypotheses/instrument.

22 hypotheses × 3 instruments = 66 (cumulative 137 → 203 of 2,000). Universe, four-null battery, engine defaults exactly as prior scans (shuffle degenerate for constant-direction variants, on record).

Pre-declared per-instrument pattern tests (mechanical)

• Displacement REPLICATES on an instrument iff disp5 signed > 0 at ≥2 of 3 horizons with block-bootstrap CI > 0, AND monotone disp4→disp5 at those horizons.
• Streak REPLICATES iff the ETH-declaration rule holds: up_ge8 positive with CI > 0 at 1h AND 4h, monotone vs up_ge5, ≥4/6 year-blocks (SOL: ≥2/3 of its blocks — see power note), down-side shows no comparable clean effect.
• Power note (declared): SOL has 2.4y in-window → its year-block and CI power is materially below the others; a SOL miss is "under-powered, not refuted" unless the point estimate is negative; a SOL pass counts as supporting, never decisive. BNB/XRP carry full weight.

Pre-declared reading

Prior: displacement replicates broadly (it did on ETH near-identically, and the mechanism — post-compression order-flow continuation — is not BTC-specific); streak replication is genuinely uncertain (thinner effect, asymmetric, possibly tied to BTC/ETH-style leverage microstructure; XRP's retail-heavy flow is the interesting test). Score-keeping declared up front: for each family, report the cross-instrument tally (N of 5 replicate, with BTC/ETH on record) — the deliverable is the map, not any single cell. A patchy streak map (≤2 of 5) materially weakens the look-2 case and will be reported as such in the decision memo before Richard answers.

Results (filled 2026-07-25, three ledger-logged runs — sol, bnb, xrp order; 66 hypotheses,

cumulative 133→155→177→199/2000 — see Ledger-count ambiguity below)

Configs: engine/scans/configs/universality-{solusdt,bnbusdt,xrpusdt}.json (both frozen families reused verbatim — displacement expressions byte-identical to displacement-repro.json, streak expressions byte-identical to rhythm.json, engine default nulls 1000/1000/1000, min_events 200). Runner: scripts/run_universality_scan.py (one shared scan.run_scan() call for disp4/disp5, one per-variant call per streak condition — same architecture as run_session_structure_scan.py + the streak scripts' (expr)*0 + sign pattern). Full tables in engine/scans/results/universality-{solusdt,bnbusdt,xrpusdt}.json; 22/22 dry-run hypothesis counts confirmed on all three before the logged run. Loaded spans (verified against the feature-store frames, not assumed): SOLUSDT 2020-08-11→2022-12-31 (~2.4y, 3 calendar-year blocks: 2020/21/22); BNBUSDT 2017-11-06→2022-12-31 (~6.1y, 6 blocks: 2017(partial)/18/19/20/21/22); XRPUSDT 2018-05-04→2022-12-31 (~4.7y, 5 blocks: 2018(partial)/19/20/21/22). 196 tests green throughout (no source changes, config+script additions only).

Cross-instrument map (all 5 instruments; BTC/ETH pulled from

engine/scans/results/displacement-repro-btcusd.json, reports/scan-displacement-events/results-ethusdt.json, engine/scans/results/rhythm-btcusd.json, engine/scans/results/upstreak-eth.json — verified against those files, not transcribed from prior status prose)

n columns are (disp5 events / up_ge8 events).

Displacement tally: 4/5 REPLICATE (BTC, ETH, BNB, XRP), 1/5 UNDER-POWERED-not-refuted (SOL). Streak tally: 2/5 decisively REPLICATE (BTC, ETH) + 1/5 REPLICATES-but-only- supporting (SOL, power-limited) + 2/5 DO NOT REPLICATE (BNB, XRP) — a patchy map exactly matching the pre-declared honest prior ("streak replication is genuinely uncertain... a patchy streak map (≤2 of 5) materially weakens the look-2 case").

Family 1 — displacement: per-instrument mechanical rule check

("REPLICATES on an instrument iff disp5 signed > 0 at ≥2 of 3 horizons with block-bootstrap CI > 0, AND monotone disp4→disp5 at those horizons")

• SOL — UNDER-POWERED, not refuted. disp5 n=65 (well under min_events=200; disp4 n=173 also flagged underpowered). Zero of 3 horizons clear CI>0 (15m [−0.429,0.638], 30m [−0.453,0.765], 1h [−0.743,0.761] all straddle zero). Per the declared power note ("a SOL miss is under-powered, not refuted unless the point estimate is negative"): 2 of 3 point estimates are positive (15m +0.065, 30m +0.145) and the third is negligibly negative (1h −0.007, effectively zero) — not the clean-negative case the note reserves for an actual refutation, so this reads as under-powered rather than a miss.
• BNB — REPLICATES. 2 of 3 horizons clear CI>0 with the same sign: 30m signed +0.656 CI [0.146, 1.184]; 1h signed +0.820 CI [0.185, 1.472] (15m +0.263 CI [−0.128, 0.675] crosses zero, supportive but not counted). Monotone disp4→disp5 at both qualifying horizons: 30m disp4 +0.378→disp5 +0.656; 1h disp4 +0.352→disp5 +0.820 (also monotone at 15m: +0.099→+0.263). Rule met at 2/3 as required.
• XRP — REPLICATES, and the cleanest of all five. All 3 of 3 horizons clear CI>0 (15m +0.553 CI [0.143, 1.028]; 30m +0.829 CI [0.330, 1.382]; 1h +0.890 CI [0.290, 1.532]) with monotone disp4→disp5 at every horizon (15m +0.364→+0.553; 30m +0.600→+0.829; 1h +0.612→+0.890). No CI here is as tight-crossing as BTC's/ETH's 15m cell — this is the strongest single-instrument displacement reading in the project to date.

Displacement verdict: broad crypto-class replication holds (4/5, all instruments with adequate power), consistent with the prior BTC→ETH replication and the pre-declared reading ("the mechanism — post-compression order-flow continuation — is not BTC-specific"). SOL's miss is a power artifact of its short listing history, not evidence against the effect; BNB's own disp5 count (176) is itself sub-threshold but its CI still clears at 2 of 3 horizons, so it counts as a pass under the declared rule as written (the rule conditions on CI-clearing, not on the min_events power flag).

Family 2 — streak: per-instrument mechanical rule check

("REPLICATES iff up_ge8 positive with CI>0 at 1h AND 4h, monotone vs up_ge5, ≥4/6 year-blocks (SOL: ≥2/3) sign-stable, down-side shows no comparable clean effect")

• SOL — REPLICATES (supporting only, per the declared power note; SOL never counts as decisive). up_ge8 1h +0.343 CI [0.102, 0.602] and 4h +0.665 CI [0.033, 1.321] — both exclude zero (4h narrowly: lower bound +0.033). Monotone vs up_ge5: 1h +0.106→+0.343; 4h +0.115→+0.665. Year-blocks (3 total: 2020/21/22) — 1h: 2020 −0.353, 2021 +0.613, 2022 +0.439 → 2/3 positive, clears the SOL-adjusted ≥2/3 bar; 4h: 2020 −0.605, 2021 +1.509, 2022 +0.108 → 2/3 positive, clears. Down-side: streak_down_ge8 1h +0.198 CI [−0.074, 0.493] and 4h +0.348 CI [−0.230, 0.966] — both cross zero, no clean down-side effect. All four legs of the rule clear under SOL's own bar.
• BNB — DOES NOT REPLICATE (narrow miss). up_ge8 1h +0.221 CI [0.055, 0.389] clears, but 4h +0.261 CI [−0.122, 0.661] crosses zero — criterion (a) requires CI>0 at BOTH 1h AND 4h, and 4h fails outright despite the point estimate being positive. Monotone vs up_ge5 holds at both horizons regardless (1h +0.105→+0.221; 4h +0.159→+0.261) and the year-block count would have cleared even the strict ≥4/6 bar (1h 6/6 positive: 2017 +0.015, 2018 +0.183, 2019 +0.320, 2020 +0.109, 2021 +0.188, 2022 +0.484; 4h 4/6 positive: 2017 −0.480, 2018 +0.629, 2019 −0.181, 2020 +0.244, 2021 +0.723, 2022 +0.008) — but the rule is conjunctive and the CI leg fails, so the mechanical verdict is non-replication. Down-side is not clean either (1h +0.030 CI [−0.195, 0.259]; 4h +0.051 CI [−0.413, 0.493] — both cross zero), consistent with the asymmetric pattern but doesn't rescue the miss.
• XRP — DOES NOT REPLICATE, and the sign inverts. up_ge8 1h +0.119 CI [−0.152, 0.377] already crosses zero. 4h −0.406 CI [−0.809, −0.012] — negative and CI-clean, the opposite sign to BTC/ETH/SOL's continuation finding. Monotone-vs-up_ge5 breaks down: 1h up_ge5 −0.054→up_ge8 +0.119 (sign flips, not a same-sign dose response); 4h up_ge5 −0.143→up_ge8 −0.406 (both negative, magnitude increases with streak length — dose response toward reversal, not continuation). Year-blocks confirm the inversion is not a fluke of pooling: 4h signed_by_year 2018 −0.511, 2019 −0.805, 2020 −0.776, 2021 +0.205, 2022 +0.045 → only 2/5 positive, i.e. the negative pooled mean reflects broad-based negativity across years, not one outlier year. Down-side (streak_down_ge8) shows no clean effect either (1h +0.077 CI [−0.161, 0.312]; 4h +0.027 CI [−0.354, 0.442]).

Streak verdict: does not replicate cleanly across the asset class. Two instruments (BTC, ETH) show the clean continuation pattern; SOL supports it but is power-limited by declaration; BNB misses narrowly on one CI leg; XRP inverts outright with its own CI-clean, multi-year negative effect at the same cell (up_ge8 @ 4h) where BTC/ETH/SOL are positive. Per the pre-declared reading, this is exactly the "patchy... ≤2 of 5" outcome flagged in advance as materially weakening the look-2 case for the up-streak-≥8 lead.

Ambiguity resolutions (explicit)

• Ledger-count mismatch (declaration vs actual): the declaration's framing assumed a starting cumulative of 137 (137 → 159 → 181 → 203). The ledger's actual cumulative immediately before this sweep was 133 (last row: upstreak-eth, cumulative 133/2000 — verified via ledger.cumulative_count() and runs/experiments.db directly), a constant 4-hypothesis discrepancy that predates this sweep and is not something this scan can or should correct after the fact (no ledger rows were edited, per the hard rules). Actual observed sequence: 133 → 155 (sol) → 177 (bnb) → 199 (xrp) — each step still +22 as declared; only the starting offset differs. Flagged here rather than silently reconciled.
• XRP's 5 vs BNB/BTC/ETH's 6 year-blocks: the declaration's power note explicitly frees only SOL from the ≥4/6 bar ("BNB/XRP carry full weight"), but XRP's data (listed 2018-05-04) covers only 5 calendar-year blocks (2018 partial + 2019–2022), one fewer than BTC/ETH/BNB's 6 (all starting with a partial 2017). This is a data-availability fact, not a chosen adjustment — XRP's year-block counts above are reported as "x/5", and the ≥4/6 bar is read as an absolute count (≥4 sign-stable blocks) rather than a proportion, which is the stricter of the two readings for a 5-block instrument. It did not end up load- bearing for XRP's verdict either way (XRP fails on the CI leg at 4h, not the year-block leg — indeed its year-block count there, 2/5, itself confirms the inversion rather than contradicting it).
• BNB's disp5 event count (176) is below min_events=200 (flagged underpowered) even though BNB is not the power-adjusted instrument. The declaration's mechanical displacement rule conditions only on CI-clearing and monotonicity, not on the min_events flag, so BNB's displacement pass stands as declared; this is noted for the record since it means the map's "4/5 replicate" count includes one instrument whose event count is itself thin (176 < 200), same caveat class as SOL's but far less severe (176 vs 65, and BNB's CI actually clears where SOL's does not).
• "disp5 15m/30m/1h" vs "1h" label: read horizons: [15, 30, 60] exactly as in displacement-repro.json (60min = the "1h" key in scan.run_scan's output), not a separate 60m-vs-1h distinction — same convention as every prior displacement scan.
• Streak freshness / per-variant direction pattern: identical resolution to run_rhythm_scan.py/run_upstreak_eth_scan.py — each streak counts once, at the bar where it first reaches the threshold (up_streak == 5/== 8, not >=), disjoint event sets per side; (expr)*0 + sign parenthesization used throughout to avoid the documented operator-precedence bug.
• n_rand/n_shuffle/n_boot/min_events at "engine defaults" (1000/1000/1000/200), not displacement-repro.json's own bespoke-reproduction values (200/200/1000) — the declaration says "engine defaults exactly as prior scans," and every prior scan since session-structure/rhythm (which itself reuses the exact disp4 event definition) uses 1000/1000/1000/200; the 200/200 values in displacement-repro.json were specific to matching the retired bespoke script's null draws byte-for-byte and are not part of the "frozen definitions" (the event/universe/direction expressions) being reused here.

Consequence

The cross-instrument map is asymmetric exactly as the declaration predicted it might be: displacement continuation is crypto-class structure (4/5 clean replications, 1 under-powered-not-refuted — no instrument shows a clean negative, unlike gold's outright inversion) while streak persistence is not (2/5 clean, 1 supporting-only, 2 non- replications including one outright sign inversion at the exact cell, up_ge8@4h, where the other three are positive). This narrows the mentor-adjacent up-streak-≥8 lead from "BTC+ETH replication, project's next-highest-priority lead" (per the 2026-07-24 ETH scan) to a BTC/ETH/SOL-only pattern that does not generalize to BNB or XRP — materially weakening the case for a look-2 pursuit of that lead, as pre-declared. Displacement's broader replication strengthens the case that Family 1 is the more universal crypto structure and the better candidate for any future execution-geometry work, subject to the same cost-sourcing caveats already on record (no venue here has a sourced cost model; this remains a signal-mapping exercise, not a license for an execution test).