search_hybrid vs proteus-align — pipeline comparison & alignment

[Sulstice]

search_hybrid vs proteus-align

Follow a neural term and a schema pair through each system — step by step.

---

search_hybrid — "What does this term officially mean?"

One term goes in. The best matching official concept comes out.

INPUT
  "spike_time_ms"
        │
        ▼
STEP 1 · Keyword Search (BM25)
  Splits into: "spike"  "time"  "ms"
  Keyword hits in ontology:
    → spike timing            142 docs
    → millisecond timestamp    89 docs
    → action potential         58 docs
        │
        ▼
STEP 2 · Meaning Search (Vector Embedding)
  Converts phrase to a number array: [0.82, -0.14, 0.56 ...]
  Finds closest meanings in database:
    → action potential onset    0.94
    → neural firing timestamp   0.91
    → spike interval marker     0.87
        │
        ▼
STEP 3 · AI Re-Ranks Both Lists Together
  Merges all candidates, scores them:
    #1  Neural Action Potential Onset   0.97  ← winner
    #2  Spike Timing Event              0.82
    #3  Temporal Neural Marker          0.71
        │
        ▼
OUTPUT
  ✓ Neural Action Potential Onset
    ID: NLX:0001234 · NIFSTD Ontology

---

proteus-align — "Do these two lab formats describe the same things?"

Two schemas go in. A field-by-field mapping comes out.

INPUT A                          INPUT B
Lab A · MGH                      Lab B · UCSF
──────────────────               ──────────────────────
spike_time_ms                    neural_event_onset_ms
channel_id                       electrode_index
signal_amplitude_uV              volt_amplitude_micro
recording_session_id             session_uuid
        │                                │
        └──────────┬─────────────────────┘
                   ▼
STEP 1 · Parse Each Field Into Concept Tokens
  spike_time_ms         → [spike, time, ms]
  neural_event_onset_ms → [neural, event, onset, ms]

  channel_id            → [channel, id]
  electrode_index       → [electrode, index, id]

  Shared tokens = possible match
                   │
                   ▼
STEP 2 · Score Every Field Pair
  spike_time_ms  ↔  neural_event_onset_ms    0.94  ████████████
  channel_id     ↔  electrode_index          0.89  ███████████░
  signal_amp_uV  ↔  volt_amplitude_micro     0.96  ████████████
  session_id     ↔  session_uuid             0.91  ████████████
                   │
                   ▼
OUTPUT · output.sssom.tsv
  ✓ spike_time_ms        → neural_event_onset_ms
  ✓ channel_id           → electrode_index
  ✓ signal_amplitude_uV  → volt_amplitude_micro
  ✓ recording_session_id → session_uuid

---

At a Glance

	search_hybrid	proteus-align
Input	one term or phrase	two full schemas
Output	official ontology concept	field-to-field mapping file
Asks	what does this word mean?	do these formats overlap?
Scale	up to 4,000 terms / batch	whole schema vs whole schema
Used by	researchers, annotators	data engineers, registry teams

---

Simple rule:

• Use search_hybrid when you have a word and need to know what it officially means.
• Use proteus-align when you have two datasets and need to know if they're talking about the same things.