diff --git a/IF_TTT_FAILURE_MODE_ANALYSIS_v1.md b/IF_TTT_FAILURE_MODE_ANALYSIS_v1.md
new file mode 100644
index 0000000..9e3b09a
--- /dev/null
+++ b/IF_TTT_FAILURE_MODE_ANALYSIS_v1.md
@@ -0,0 +1,156 @@
+# IF.TTT Failure Mode Analysis v1 (Why traceability is not optional)
+
+**Author:** Danny Stocker (`ds@infrafabric.io`)  
+**Canonical (static mirror):** `https://infrafabric.io/static/hosted/IF_TTT_FAILURE_MODE_ANALYSIS_v1.md`  
+**Repo source:** `https://git.infrafabric.io/danny/hosted/src/branch/main/IF_TTT_FAILURE_MODE_ANALYSIS_v1.md`  
+**SHA256 (sidecar):**
+- `https://infrafabric.io/static/hosted/IF_TTT_FAILURE_MODE_ANALYSIS_v1.md.sha256`
+- `https://git.infrafabric.io/danny/hosted/raw/branch/main/IF_TTT_FAILURE_MODE_ANALYSIS_v1.md.sha256`
+
+**Subject:** Failure mode analysis of removing cryptographic chain-of-custody from an AI governance stack  
+**Protocol:** IF.SEC.threat.model.v1  
+**Status:** FORENSIC ASSESSMENT (not evidence; not certification)  
+**Date (UTC):** 2025-12-21  
+
+---
+
+This document answers one question:
+
+What actually breaks when you replace IF.TTT (Traceable/Transparent/Trustworthy) with “standard logs”?
+
+This is not hypothetical theatre. This failure pattern has already harmed people at scale.
+
+---
+
+## 1) The compromise (how it really happens)
+
+An organization removes **cryptographic signatures + immutable audit chaining** and replaces it with mutable JSON/text logging (Splunk/Datadog/etc).
+
+The rationales are not “2ms latency”. The rationales are organizational:
+
+- “Compliance says monthly audit exports are sufficient.”
+- “We don’t want key management in the hot path.”
+- “We can’t afford per-request signing costs at current volume.”
+- “The new CTO comes from a startup background; they optimize for speed, not forensics.”
+
+The shift is simple:
+
+- From **proven identity** (key-based, non-repudiable)
+- To **asserted identity** (string-based, deniable)
+
+---
+
+## 2) Immediate effects (technical)
+
+Removing cryptographic binding causes immediate epistemic collapse:
+
+1) **Identity spoofing**  
+Any actor can write `{"actor":"admin"}` into a log stream. Without signature verification, “admin” is a string, not a proof.
+
+2) **Repudiation becomes cheap**  
+An operator can deny actions. The system has no mathematical contradiction to present.
+
+3) **Post-hoc editing becomes indistinguishable from “cleanup”**  
+Even honest operators edit logs during incident response (redaction, dedupe, “fix formatting”). Without a hash chain, edits are invisible.
+
+4) **Time becomes an argument, not evidence**  
+Wall-clock timestamps are not cryptographic time. A system clock can drift or be altered. If the log is mutable, time is now a narrative.
+
+---
+
+## 3) Cascade failures (systemic)
+
+IF.TTT is the load-bearing wall. When it fails, upstream components lose their ability to prove they acted correctly.
+
+- **IF.GUARD failure**  
+An actor bypasses governance and injects a fake “Approved” record later. Without non-repudiable signatures and an immutable chain, the council cannot prove they *did not* approve it.
+
+- **IF.emotion corruption**  
+Retrieval context can be poisoned (bad document inserted) and the retrieval log can be edited to look “clean”. The system produces harmful output; the investigation cannot recover the root cause.
+
+- **Swarm communications collapse**  
+Inter-agent trust degrades: a message “from Swarm B” is no longer provably from Swarm B. Man-in-the-middle and replay confusion become operationally plausible.
+
+---
+
+## 4) Real incident scenario (how you lose in court)
+
+**The “hot patch” liability event**
+
+1) An LLM system is integrated into a support workflow (medical, legal, crisis, finance).  
+2) Under pressure, an engineer hot-patches a prompt/config (“make it more helpful”).  
+3) The system produces harmful guidance.  
+4) A user captures the harmful output (screenshot, screen recording, saved HTML, exported conversation).  
+5) During investigation, the engineer “cleans up” the deployment log line (panic, shame, or misguided privacy redaction).  
+
+Outcome:
+
+- The organization produces logs claiming the system used safe configuration.  
+- The plaintiff produces the harmful output.  
+- The discrepancy becomes **indefensible** because the chain-of-custody is missing.  
+
+In high-liability domains, “we have logs” is not a defense. It is a liability amplifier.
+
+---
+
+## 5) Historical precedents (pattern, not analogy)
+
+These are different domains. The shared mechanism is always the same: **mutable records + privileged access + institutional denial**.
+
+- **UK Post Office Horizon scandal** — forensic dispute about ledger truth and privileged system access.  
+  - Background: `https://en.wikipedia.org/wiki/British_Post_Office_scandal`
+
+- **Boeing 737 MAX / MCAS** — catastrophic outcomes coupled to opaque system behavior and disputed control/alerting assumptions.  
+  - Background: `https://en.wikipedia.org/wiki/Boeing_737_MAX_groundings`
+
+- **Wells Fargo fake accounts scandal** — incentive pressure plus record-keeping failures and downstream denial dynamics.  
+  - Background: `https://en.wikipedia.org/wiki/Wells_Fargo_account_fraud_scandal`
+
+- **Volkswagen emissions scandal** — engineered “compliance theatre” in software, with auditors unable to trust vendor assertions.  
+  - Background: `https://en.wikipedia.org/wiki/Volkswagen_emissions_scandal`
+
+IF.TTT is designed to prevent the *mechanism*, not to claim these incidents had identical root causes.
+
+---
+
+## 6) Defense validation (what IF.TTT prevents)
+
+IF.TTT prevents the “mutable record” class of failure by enforcing:
+
+1) **Non-repudiation**  
+Actions are signed by keys. “Who did it” becomes a proof, not a debate.
+
+2) **Tamper-evident history**  
+Events are hash-chained; deletions/edits break verification.
+
+3) **Bounded completeness (within boundary)**  
+Requests that cross the witness boundary are witnessed in REQ_SEEN with inclusion proofs.
+
+4) **Portable evidence**  
+Incidents can be exported as bundles with manifests + verifier commands, enabling third-party review without trusting the operator.
+
+This turns disputes into verification.
+
+---
+
+## 7) Residual risks (what IF.TTT does not magically solve)
+
+Even with IF.TTT deployed, real risks remain:
+
+1) **Key compromise**  
+If a private key is stolen, an attacker can sign bad actions that verify as “valid”. Mitigation requires hardware-backed keys, rotation, and incident response.
+
+2) **Rubber-stamp governance**  
+The audit trail can be perfect while the decision is wrong (fatigue, negligence, incentives). Mitigation requires IF.GUARD diversity + escalation rules + post-incident review.
+
+3) **Boundary gaps**  
+Events dropped before the witness boundary (edge, network) may be outside scope unless edge witnessing is enforced.
+
+---
+
+## Conclusion
+
+Removing IF.TTT turns the system from **high-assurance intelligence** into a **black-box liability generator**.
+
+The cost of “not having receipts” is not measured in milliseconds. It is measured in ruined investigations, regulatory exposure, and human harm.
+
diff --git a/IF_TTT_FAILURE_MODE_ANALYSIS_v1.md.sha256 b/IF_TTT_FAILURE_MODE_ANALYSIS_v1.md.sha256
new file mode 100644
index 0000000..991cfb1
--- /dev/null
+++ b/IF_TTT_FAILURE_MODE_ANALYSIS_v1.md.sha256
@@ -0,0 +1 @@
+9b78ba5394e2cd85cde2fa0bfd8941bafc593d58593f942d5d2f11d857058c19  IF_TTT_FAILURE_MODE_ANALYSIS_v1.md