# FLOW Data moves through the Fabric only when actualized by a specific process; until then, paths are latent — possible but not real. --- ## Inference Vivification Flow (actualized) Raw inference text enters; a scored, categorized semantic unit exits — the Fabric grows one node. - Raw text arrives via STDIN or argument → `vivify.py` - Left-semantic pass: Claude API extracts 8-12 keyword clumps capturing felt meaning, not surface words - **Tension node — structural**: the inference is placed in `inferences/unclustered/` — the Fabric has not yet decided where it belongs - `keyword_graph.py` accumulates co-occurrence edges across all inferences — graph grows with corpus - `categorize.py` reads graph seeds (left keywords only) → builds emergent 2-3 layer category tree → assigns `category_paths` - **Tension node — conflict**: `right_pass.py` runs a digital/structural pass; keywords describe the same inference analytically, not semantically - `tension_score.py` scores three numbers — `predicted` from operator coordinates, `confirmed` from `right_pass` claimed-vs-actual discrepancies, `calibration_delta` between them - **Tension node — resolution**: `beneficial_signals()` surfaces inferences where divergence exceeds threshold — these are intervention candidates - Phase 5 `prediction_output()` shapes the result as a navigational signal, not a conclusion --- ## Secret Storage Flow (actualized) A secret enters encrypted by the client; the server strips routing layers and stores a human-readable JSON file. - Secret created on client → `vivify.py` + `secrecy.py` (PBKDF2/Fernet, client-side) - `freeze.py` serializes payload to base64 JSON for IP transit - SSH tunnel carries payload to phone (star topology hub) - `server.py` (Flask, Android/Termux) receives → strips name + app-name layers → recreates directory path - JSON stored at `db/{username}/{app_name}/secret.json` — filesystem is the database - Blobs separated; raw sensitive material never stored unencrypted --- ## Tension Nodes (cross-flow) Tension is structural pressure pointing toward the next evolution — four types, each visible in the flow. - **Structural** — the unclustered drop point in vivify: inference exists but Fabric hasn't autovivified to place it; graph must accumulate before categorization is possible - **Conflict** — un-truth measured by `tension_score.py`: `predicted` is the operator's judgment, `confirmed` is ground truth from the record, and a negative `calibration_delta` is the operator under-smelling what the record confirms - **Functional** — rules in `pillars/CLAUDE.md` and `pillars/doc_standard_v1.json`: constraints that resist certain flows (no external taxonomies, no schema-first design, no concluding statements in exploratory sections) - **Resolution** — `beneficial_signals()` and Phase 5 output: the pull toward beneficial outcome; what the system is oriented toward, not what it has achieved --- ## What Isn't (unactualized) These paths exist as structural pressure — latent in the Fabric, not yet actualized. - **Typed tension JSON**: `tension_score.py` outputs three numbers rather than the original single float; the typed shape `{type, magnitude, between, resolution_vector}` is specified but still not in the code - **Normalization standard**: conflict inputs enter raw — no standard yet for stripping proper nouns, collapsing entities to typed roles, offsetting dates; data contaminates before it vivifies - **Cloudflare Workers runtime**: the Fabric is described in terms of Worker isolates and Durable Objects but runs locally in Python; the translation is unactualized - **Animal Crossing instantiation**: same underlying model, different population (elders, children) — the instantiation doesn't exist; only the shared model does - **Reify feedback loop**: `reify.py` holds two inferences in synthesis simultaneously but doesn't write back to the graph or update `category_paths`; synthesis is terminal, not recursive - **Cross-inference autovivification**: the moment a Flow decides the Fabric needs a new branch is not yet explicit — it happens by corpus accumulation, not by a triggered decision - **`right_pass.py --all` duplication**: inferences appear across multiple category paths when run in bulk — bug, not feature - **Secret Server ↔ vivify-inferences connection**: the two flows are independent; no path yet from a vivified inference into the secret storage layer ---