forked from chiki2bum2/SniperGold_ML
34 lines
No EOL
1.5 KiB
Markdown
34 lines
No EOL
1.5 KiB
Markdown
# P3-S25.1 — ORACLE COMPLETION (status)
|
|
|
|
```text
|
|
2026-08-26 | Recovery #3 | Status: ORACLE IMPLEMENTED + TESTED ON FIXTURES ONLY;
|
|
full-source oracle equality NOT yet run (requires a completed ingestion)
|
|
```
|
|
|
|
## Oracle status
|
|
|
|
- Independent vectorised oracle (`s251_oracle.py`): implemented — `oracle_m15`
|
|
and `oracle_m30` are conceptually different from the streaming state machine
|
|
(stable sort + bucket-key change detection + segment reduction).
|
|
- Fixture-level agreement with the streaming path: **verified in test suite
|
|
T05/T07** (oracle equality on `mt100` fixture) on fixtures only.
|
|
- Full-source M15/M30 oracle output: **PENDING** — the production ingestion
|
|
into M15/M30 must complete first; then oracle arrays can be computed and
|
|
compared (stream vs oracle) over the real output.
|
|
|
|
## What must happen for "oracle completion"
|
|
|
|
1. Complete a bounded, checkpointed ingestion run over the real source.
|
|
2. Compute stream M15/M30 outputs (files) + independent oracle on the same
|
|
source windows (batch).
|
|
3. Compare stream vs oracle across the whole covered period — agreement must
|
|
be exact (timestamp/open/high/low/close/ticks/n_m15).
|
|
4. Record the comparison result in this document + machine-readable evidence.
|
|
|
|
## Current state
|
|
|
|
Nothing numeric about the real source is reported as oracle-verified yet;
|
|
nothing was guessed or fabricated. Oracle equality is validated on fixtures
|
|
only (tests S25.1-T05 / S25.1-T07).
|
|
|
|
*End of P3-S25.1 oracle completion status.* |