⚖️ Verdict and Decision Table

The campaign’s answer: what earns edge in the 27-signal catalog, what does not, and the full weight table. Dated 2026-09-27 · HEAD 3108547 · segment 4 closed.

Verdict

• What earns edge: episodic on-chain flow lead-ins — whale_tx_spike, whale_flow_composite, whale_pump_frontrun, shock_continuation ride whale-accumulation ramps that precede strikes. Blowoff is handled by a funding-overheat mute (the composite vetoes longs once EMA18 funding exceeds 0.0006 ≈ 6 bps), not by climax fades; capitulation pays only with inflow confirmation. Segment-2 rewrites add false-lead-in resistance (family 2 emits ~30 % false leads with identical pre-strike flow columns): partial pre-strike size with strike confirmation, momentum-confirmed sizing, bounded episodes, IV/isolated-tail gating.
• What does not: shorting the drains — trend_down/distribution (constant-short distribution Sharpe −1.90 IN; bear_regime_short degrades IN −0.44…−1.92 at every weight), contrarian funding (every funding_momentum book failed held-out/TEST, −0.41…−0.51), and shock fades (every fade design negative across weights 1–5 on IN/HO).
• Largest structural win: inverse-volatility sizing of the combined position (24 h window, 0.02 hourly target, 5× cap) — the biggest ledger step, IN median 3.131 → 4.213 (+1.08) at run #9.
• Generalisation status (segments 2–3, two families): pooled in-sample 3.31 — family 1 4.59 → 4.82, family 2 1.32 → 2.09 after the whale/shock rewrites; selection HO1 3.22 / HO2 3.44; TEST 4.23 is diagnostic only (non-blind). Segment 4 added the third generator family (jump diffusion, latent beta, separable false lead-ins) and closed with the run #16 book restored — the run #17 descent was flagged and reverted.
• The honest out-of-family number: single-touch fresh seeds (2026-09-27, all now burned) give family medians f1 3.39/3.79, f2 1.79/0.08 (worst single path −1.21), f3 4.84/4.90 — the honest flat edge is ~1.8–4.8 Sharpe in-family (realistic costs 2.5 / 0.5 / 2.6), not the 4.8 in-sample family-1 figure taken at face value.
• Standing caveat: synthetic generator families, real feeds untested; the family-2 funding-clamp fixture flaw was fixed in segment 3, yet funding logic still adds nothing on the repaired fixture; flat stablecoin supply leaves stablecoin_supply_growth vacuous on family 2; synthetic dev_commits performance does not validate the real feed (Santiment flagged GitHub dev metrics as unreliable, Aug 2026). A measurement harness and playbook, not a profitability claim.

Decision table — all 27 signals

W = current WEIGHTS (funded ⇒ > 0). Feeds are the providers cited by each spec, not live exchange wiring — no spec wires an exchange API. shines/fails are catalog claims, not measured rankings; measured per-regime numbers live in the regime map.

Signal	TF	Horizon	Shines in	Fails in	W · role
ema_cross_fast	1h	swing	trend_up, trend_down, vol_shock	range, whale_dump, whale_pump	0 · diagnostic
ema_cross_slow	4h	trend	blowoff_top, whale_dump	range, whale_pump, capitulation, accumulation	0 · diagnostic
macd_hist	4h	swing	blowoff_top, capitulation, whale_dump, vol_shock	range, trend_down, distribution	0.5 · funded
rsi_mean_reversion	1h	reactive	range, trend_down, vol_shock	whale_pump, blowoff_top, whale_dump, trend_up	0 · diagnostic
bollinger_fade	4h	swing	range, trend_down, vol_shock	whale_dump, whale_pump, blowoff_top, trend_up	0.25 · funded
donchian_breakout	1d	trend	trend_up, trend_down, vol_shock, blowoff_top	range, whale_dump, whale_pump, capitulation	0 · diagnostic
vol_breakout	1h	reactive	vol_shock, whale_dump, whale_pump, capitulation, blowoff_top	range, accumulation, distribution	0 · diagnostic
adx_trend_strength	4h	swing	trend_up, trend_down, vol_shock	range, accumulation, distribution, whale_dump, whale_pump	0.25 · funded
volume_price_confirm	1h	reactive	trend_up, trend_down, vol_shock, blowoff_top	range, accumulation, distribution	0 · diagnostic
roc_momentum	1d	trend	trend_up, trend_down, whale_pump, blowoff_top	range, capitulation	0 · diagnostic
zscore_reversion	4h	swing	range, distribution, blowoff_top	whale_pump, whale_dump, vol_shock, trend_down	0 · diagnostic
keltner_squeeze_break	4h	swing	range, accumulation, vol_shock, whale_dump, whale_pump	trend_up, trend_down	0 · diagnostic (was 0.25 funded, zeroed run #14)
exchange_netflow_pressure	1h	reactive	whale_dump, whale_pump, capitulation	blowoff_top, trend_up, trend_down	0 · diagnostic
whale_tx_spike	1h	reactive	whale_dump, whale_pump, capitulation	blowoff_top, accumulation, trend_up	5.0 · funded
whale_net_accum_trend	1d	trend	accumulation, trend_up, distribution	capitulation, vol_shock	0 · diagnostic
active_addresses_momentum	1d	swing	trend_up, accumulation, vol_shock	range, distribution	0 · diagnostic
sopr_regime	1d	swing	trend_up, blowoff_top, whale_pump	capitulation, range, distribution	0 · diagnostic
mvrv_zscore	1d	swing	capitulation, blowoff_top, distribution, accumulation	trend_up, trend_down	0 · diagnostic
stablecoin_supply_growth	1d	trend	trend_up, accumulation, blowoff_top	whale_dump, trend_down, capitulation	2.0 · funded
funding_momentum	4h	swing	blowoff_top, whale_dump	range, trend_down, distribution	0 · diagnostic
oi_price_divergence	4h	swing	vol_shock, whale_dump, capitulation, blowoff_top	trend_up, trend_down	0 · diagnostic
oi_funding_squeeze	4h	swing	trend_down, whale_dump	range, accumulation, whale_pump, vol_shock	0 · diagnostic (funded-swap candidate rejected)
dev_commit_trend	1d	trend	accumulation, whale_pump, capitulation	whale_dump, trend_down, distribution	5.0 · funded
shock_reversal	1h	reactive	capitulation, whale_dump, whale_pump, vol_shock, blowoff_top	trend_up, trend_down	0 · diagnostic
shock_continuation	1h	reactive	vol_shock, blowoff_top, capitulation, distribution	whale_dump	3.0 · funded
whale_flow_composite	4h	swing	whale_pump, whale_dump	vol_shock, range, trend_down	3.0 · funded
whale_pump_frontrun	1h	reactive	whale_pump	accumulation, trend_down, whale_dump	3.0 · funded

Funded weight sum: 22.0 across 9 of 27 signals.