Dashboard β€Ί Three-Axis Probes

substrate Β· findings/evolution/shared_data/

πŸ—„ Crypto Shared Data Pool

The crypto-only central data layer every probe reads from β€” locked scope, read-only ClickHouse, and the SSoT registries that make the probes deterministic. Non-audit: it holds the substrate, not the verdicts. Every item in the folder is carded below; each finding opens its own plain-English page.

The locked substrate

dimensionvalue
Access contractREAD-ONLY ClickHouse (SELECT only) Β· opendeviationbar_cache.open_deviation_bars Β· runs on bigblack
SymbolsBTCUSDT Β· ETHUSDT Β· BNBUSDT Β· LTCUSDT Β· ADAUSDT Β· XRPUSDT
Symbol Γ— threshold21 (symbol, threshold) combos β€” thresholds 100/250/500/750 dbps (750 = spearman-only tier)
Time slices21 two-month windows, 2021-01 β†’ 2026-02 (3 Γ— 2020 windows dropped: no BTC/ETH bars)
Regime epochs7 epochs: bull_2021 Β· bear_2022 Β· stress_2022 Β· recovery_2023 Β· bull_2024 Β· range_2025 Β· current_2026
Convergence floorsLOOKBACK_COUNT=200 (cell skip) Β· FLOOR_1000 (h_norm/ΞΎ/CODEC reliability)
Open scope question: Terry's orthogonality substrate names BTC/ETH/EURUSD; the pool is crypto-only since 2026-05-30 and the narrowing (msg id=209) is unverified against a Terry source β€” confirm before treating crypto-only as his ruling.

Everything in the pool

One card per item in findings/evolution/shared_data/ (29 items). Each finding opens its own plain-English explainer page; probe/gate/guard cards link to their living pages where one exists.

πŸ†• Newest engine β€” auto-rotation coverage

rotating_slice_orthogonality_probe.py PROBE NEW Β· AUTO-ROTATION
A coverage-ledger rotation engine that always tests the least-covered market condition next β€” asset Γ— symbol Γ— threshold Γ— volatility Γ— session Γ— trend β€” until every feasible condition is covered 20Γ—, fusing Spearman |ρ| + Chatterjee ΞΎ on real bootstrapped bars. Promotes a feature only if it survives the worst condition and replicates across crypto + forex. Cross-repo engine (currently on the feat/crypto-rotation-mechanism branch).
Read in plain English β†’

πŸ“„ Findings β€” each has its own plain-English page (13)

purpose.md FINDING
Why the pool exists β€” one locked grid, read once, reused by every audit so results stay comparable and cheap.
Read in plain English β†’
README.md FINDING
Overview / hub for the locked, read-only shared data layer (441-cell grid, queried live from ClickHouse).
Read in plain English β†’
pre-spec.md FINDING
The locked "pre-registration" of exactly which crypto data the probes may read (v2, crypto-only, 2021+).
Read in plain English β†’
slicing-methodology.md FINDING
Why the data is cut into a 441-cell grid, and how each slice is sized so cells hold enough bars to trust.
Read in plain English β†’
slicing-robustness-findings.md FINDING
The 441-cell robustness map: which cells hold enough bars, and how the thin/empty ones are quarantined.
Read in plain English β†’
REGIME-SLICING-ROBUSTNESS-FIX.md FINDING
Swapping calendar windows for 10 real market regimes (back to 2018) so features must survive crashes and blow-offs.
Read in plain English β†’
data-quality-notes.md FINDING
Per-cell data-quality findings (e.g. LTC@100 broken) + the rules that stop missing data faking a real zero.
Read in plain English β†’
candidate-intake-standard.md FINDING
One ledger row + one preflight per candidate, so no idea is evaluated twice and every verdict stays traceable.
Read in plain English β†’
orthogonality_probe.md FINDING
Plain-English guide to the Axis-1 orthogonality probe β€” is a feature genuinely novel or just redundant?
Read in plain English β†’
scope-expansion-options.md FINDING
Pre-computed bigger versions of the pool (more coins / thresholds / forex) so expanding later is a constant edit.
Read in plain English β†’
S1_GCM_OINFO_PIN_SPEC.md FINDING
The S1 "tamper test" spec for the GCM + O-information cascade β€” break the code, the tests must catch it.
Read in plain English β†’
forex-availability-probe.md FINDING
The archived probe that confirmed EURUSD forex data exists (5.5M bars) and locked the 42-column fetch list.
Read in plain English β†’
manifest.md FINDING
Schema & usage for manifest.json β€” the (legacy) "table of contents" for the data slices.
Read in plain English β†’

πŸ”¬ Probes, gates & guards (13)

orthogonality_probe.py PROBE
Axis-1 engine: per-cell Spearman |ρ| + h_norm + regime-stratified cross-slice aggregation.
Open the Orthogonal probe page β†’
parameterless_probe.py PROBE
Axis-2 engine: AST gate (job 1) + shared-knob promotion sweep (job 2).
Open the Parameterless probe page β†’
agnostic_probe.py PROBE
Axis-3 engine: no-dispatch scan + two-implementation convergence oracle + doc/code check.
Open the Agnostic probe page β†’
three_axis_gate.py GATE
THE gate: composes the three probes, cost-ladder short-circuit, one promote/skip verdict.
Open The Gate page β†’
codec_foci_eval.py PROBE
ΞΎ β†’ CODEC β†’ O-information cascade + leakage-free MOWFO walk-forward.
Open the CODEC/FOCI runs β†’
selection_gates.py PROBE
Selection gates S2-ΞΎ (analytic null) / S3 (walk-forward agg) / S5 (jackknife CIs).
Open the CODEC/FOCI runs β†’
selection_finalize.py GATE
Selection verdict assembler with a BLOCKING leakage preflight.
Open the CODEC/FOCI runs β†’
campaign_summary_adapter.py GATE
Campaign records β†’ gate dialect: panel + SIBLING/Finish stages, fail-closed worst-of, coverage floor, auto panel from the ledger.
test_leakage.py GUARD
Structural anti-leakage gate: future-perturbation invariant + noise null (proven teeth).
mutation_oracle.py GUARD
S1 mutation-testing oracle for the cascade primitives β€” deliberately breaks the code to prove the tests bite.
Read the S1 spec β†’
cascade_pins.py GUARD
Locked "pin" constants for the ξ→CODEC→O-information cascade (S1 pin-invariance).
Read the S1 spec β†’
candidate_ledger.py LEDGER
Ledger CLI: stages, dedup --check, --preflight (the pre-loop alignment as one command).
Read the intake standard β†’
regen_grounded_page.py TOOL
Regenerates the grounded-candidates dashboard page from the ledger.

πŸ—ƒ Registries β€” SSoT data (3)

constants_registry.toml SSOT
SSoT of every constant touching a candidate: derived / project-wide / feature-tunable.
How it's used (Parameterless) β†’
candidate_knobs.jsonl SSOT
Append-only knob inventory feeding the promotion sweep (one line per tested candidate).
How it's used (Parameterless) β†’
candidate_ledger.jsonl SSOT
Candidate lifecycle SSoT: every candidate's stage + append-only history.
Read the intake standard β†’

How a new candidate flows through the pool

1. add knobs       β†’ candidate_knobs.jsonl          (one line, append-only)
2. static gates    β†’ parameterless_probe + agnostic_probe   (laptop, free)
3. empirical sweep β†’ orthogonality_probe on bigblack        (read-only CH)
4. one verdict     β†’ three_axis_gate                        (promote / pending / blocked)
5. periodic        β†’ parameterless_probe --sweep            (constants to promote + rescue list)
← Forwardβ€”