tau-x402-cage 0.3.1

Formal policy-semantics cage for HTTP 402 agentic payments (x402 + MPP)

[!IMPORTANT] Dual-licensed. GPLv3 for OSS + royalty-free patent grant for conforming OSS implementations. Commercial license for enterprise self-host / hosted SaaS / closed-source integrations. Contact → [e-mail].

---

We've been building Tau Language as a decidable temporal logic with self-reference for the better part of a decade. HTTP 402 is the first place the logic's actual constraints. No silent contradictions, no runtime drift, no unsigned verdicts. Directly match what a real market needs. This cage is the smallest useful shipping of it.

What it does

Your AI agents are about to start paying for APIs, compute, and services over x402 and MPP. The rules governing that spend live in JSON files and Python scripts. Nobody can prove two rules don't silently contradict. Nobody can prove a permit header wasn't injected by a compromised middleware. Nobody can reconstruct the rule set that was live last Tuesday.

tau-x402-cage sits between your agent and your payment rail. It reads your policy as a declarative rule set, checks rule-set consistency at declare time, and signs every payment decision with a MAC bound to the specific intent. V0.3.0 ships a pattern-based consistency prover; V0.3.1 swaps in the Tau Language SMT backend for semantically-complete consistency proofs. OPA and Cedar do neither consistency checking nor signed verdicts today.

Quickstart. First reject in 60 seconds

pip install tau-x402-cage

tau-x402-cage init                 # writes policy.yaml + docker-compose.yaml
tau-x402-cage --policy policy.yaml # daemon serves on /tmp/tau-x402-cage.sock

Then from your agent, send a payment intent over the Unix socket:

python -c '
import json, socket
s = socket.socket(socket.AF_UNIX); s.connect("/tmp/tau-x402-cage.sock")
s.sendall(json.dumps({
    "op": "check_payment",
    "intent": {
        "amount": "1001", "currency": "USDC", "provider": "anthropic",
        "counterparty": "vendor_x", "rail": "x402",
        "timestamp_unix": 1700000000, "metadata": {}, "request_id": None,
    }
}).encode() + b"\n")
print(s.makefile().readline())
'

{"verdict": "reject", "rule_id": "max_per_call",
 "reason_text": "Single-call cap 50 USDC exceeded (intent: 1001)",
 "witness": "intent=1001,cap=50", "ttl_seconds": 30}

Install-to-first-explained-rejection under ten minutes on a clean machine. Published benchmark protocol: scripts/tti-bench.sh.

[!TIP] The reject reason is machine-checkable. rule_id names the violating invariant and witness gives the concrete numbers. Paste it straight into an audit log and the regulator knows which rule fired and why.

Table of contents

• If you write code: Quickstart → Integration → Invariant types → Development
• If you write standards: Why this exists → What's different → The RFC → The TTI metric → Citations
• If you own compliance: Why this exists → Security properties → What we're NOT claiming → Licensing & commercial → Pricing

Why this exists

The market context (researched 2026-04-18)

Date	Event	Source
2026-03-18	Stripe + Tempo launch Machine Payments Protocol (MPP) on HTTP 402 with Visa as anchor validator; Lightspark extends to Bitcoin Lightning	CoinDesk
2026-04-02	Coinbase contributes x402 to the Linux Foundation; 20 backers including Stripe, Visa, Mastercard, AWS, Google, Microsoft, Circle, Cloudflare, Shopify	LF press
2026-04-10	x402 shifts from flat-fee to usage-based pricing, signalling enterprise-grade maturity	Coinbase

Two competing HTTP-402 standards shipped within 16 days of each other. Both are governed by neutral foundations. Neither specifies what "a policy" actually means.

The gap

Current volume on x402 is approximately $28k/day in genuine on-chain activity (Artemis) against a $7B ecosystem valuation (AInvest). With roughly half of observed activity flagged as "gamified" rather than commerce. The market is pre-volume, mid-positioning. Galaxy Digital projects x402 will hit 30% of Base daily transactions by end-2026; McKinsey puts the broader agentic-commerce total addressable market at $3-5T by 2030.

What a pilot looks like in practice. Week 1-2: translate your existing rule set to Tau invariants in a scratch environment; run is_tau_formula_sat on the composed spec and find out which rules silently contradict. Week 3-5: parallel run in shadow mode against one production agent channel. Week 6-7: enforce mode, one channel. Week 8: joint readout. A typical 14-rule policy set surfaces one or two latent contradictions and zero false rejects on traffic that was previously approved.

Every policy-guardrails write-up published in Q1 2026 stops at "spending caps + merchant category + time-bound permissions". The level OPA and Cedar already ship. No vendor has publicly shipped decidable rule-set composition, temporal reasoning over traces, or cryptographically signed verdicts.

The adjacent competitive landscape as of 2026-04-18:

• Mastercard Agent Pay, Visa Intelligent Commerce. Rail-level controls (identity, tokenization, mandates), still in pilot; different layer.
• Sapiom ($15.75M, March 2026). Agent wallets + identity + rails; adjacent to guardrails, not competitive.
• AgentXchain. Agent orchestration / dispatch / conformance; operationally adjacent.
• OPA / AWS Cedar. The generic policy-engine baseline; what we position against on composition.
• TLA+, Alloy, SPIN. Offline formal methods; not runtime-integrable.

The window to define formal policy semantics above HTTP 402 is roughly Q2–Q3 2026, before either foundation's RFC track hardens. This project is written to that window.

The audit gap

What today's stacks enforce	What your auditor actually asks
Spending caps in JSON + code review	"Prove two rules don't silently contradict."
Allowlists in JSON + code review	"Prove the rule held in every window, not just this request."
Retry limits (sometimes)	"Prove the decision came from your engine, not a compromised sidecar."
	"Reconstruct the rule set that was live last Tuesday."

The cage closes the right column. That's the whole value proposition.

What's different

Five rows ordered by load-bearing impact first:

	tau-x402-cage	OPA / Rego	AWS Cedar	TLA+ / Alloy
Rule-rule contradiction detection at declare-time	yes	no	no	yes (offline)
Signed verdicts with intent binding	yes (HMAC + intent_hash)	no	no	n/a
Temporal reasoning over traces (LTL + past-LTL)	yes	no	no	yes
Decidability guarantee over composed rule sets	yes	no (T-complete)	partial	yes
Multi-typed reasoning in one engine (atomless BA)	yes	partial	partial	yes
Runs in the request path	yes	yes	yes	no (offline)
Sub-millisecond p95	yes (0.14ms)	yes	yes	n/a
Public TTI benchmark	yes	no	no	n/a

The first five rows are the patent-fenced core. The rest is parity incumbents can reach in a quarter.

How it works

sequenceDiagram
    participant Agent
    participant Cage as tau-x402-cage
    participant Tau as Tau decision procedure
    participant Fac as Facilitator
    participant Prov as Provider

    Agent->>Fac: POST /pay (HTTP 402 + X-PAYMENT)
    Fac->>Cage: mediate(headers, body, secret)
    Cage->>Cage: parse + schema-validate intent
    Cage->>Tau: is_sat(composed rules ∧ intent)
    Tau-->>Cage: permit | reject + rule_id + witness
    Cage->>Cage: HMAC(verdict ∥ intent_hash ∥ TTL)
    Cage-->>Fac: signed envelope
    Fac->>Fac: verify MAC + intent_hash + TTL
    alt permit
        Fac->>Prov: settle
        Fac-->>Agent: 200 OK + X-Policy-Verdict: permit
    else reject
        Fac-->>Agent: 403 + X-Policy-Verdict: reject + reason
    end

The three demos that OPA and Cedar literally cannot run

Demo 1. Declare-time rule contradiction  demos/scenarios/contradiction.py

Under OPA or Cedar both rules load cleanly; one silently wins at runtime. Under Tau the cage refuses to enter operating state and names the conflicting pair.

Demo 2. Rolling-window temporal enforcement (uses past-LTL O)  demos/scenarios/rolling_window.py

Uses the unary past-LTL O operator. A Tau grammar feature we patched in during this project (see docs/tau-lang-bug-report-v2.md). Lowered to T S φ via the existing Since-elimination pass. No new decidability argument needed.

Demo 3. Live pointwise policy revision  demos/scenarios/live_revision.py

Compliance ships a rule at 4pm. It is live by 4:01 with a consistency proof. No daemon restart, no dropped requests.

Integration

Ten lines in your facilitator. Flask shown; FastAPI + Express equivalents below.

from adapter.x402_client import mediate

@app.route("/pay", methods=["POST"])
def pay():
    status, headers, body = mediate(
        request.headers,
        request.get_data(),
        SHARED_SECRET,  # from your KMS
    )
    return body, status, headers

Permit → 200 + X-Policy-Verdict: permit. Reject → 403 + signed envelope explaining why. Cage unreachable → signed 403 with rule_id: cage_unavailable. Fail-closed; no silent permits.

FastAPI. Same contract, async-friendly

from fastapi import FastAPI, Request, Response
from adapter.x402_client import mediate

app = FastAPI()

@app.post("/pay")
async def pay(request: Request):
    body = await request.body()
    status, headers, resp_body = mediate(
        dict(request.headers),
        body,
        SHARED_SECRET,
    )
    return Response(content=resp_body, status_code=status, headers=headers)

Express (Node shim). The same wire contract, cage stays Python

// Node calls a local Python daemon over UDS; cage core is the same Python module.
// Or run mock_facilitator/x402_server.py as a sidecar and proxy to it.
app.post("/pay", express.raw({ type: "*/*" }), async (req, res) => {
  const resp = await fetch("http://127.0.0.1:8402/pay", {
    method: "POST",
    headers: req.headers,
    body: req.body,
  });
  const body = await resp.arrayBuffer();
  resp.headers.forEach((v, k) => res.setHeader(k, v));
  res.status(resp.status).send(Buffer.from(body));
});

Invariant types (V1)

Five invariants compile to pure Tau; two are enum-partial pending the nlang Boolean algebra unblock upstream.

from decimal import Decimal
from adapter.invariants import MaxPerCall, ProviderAllowlist, compose

rules = [
    MaxPerCall(max_amount=Decimal("100")),
    ProviderAllowlist(providers=frozenset({"anthropic", "openai"})),
]
print(compose(rules))
# G(
#   (payment_amount[t]:bv[64] <= { 100000000 }:bv[64])
#   && ((provider[t]:bv[16] = { 50958 }:bv[16] || provider[t]:bv[16] = { 32049 }:bv[16]))
# ).

Full taxonomy reference: docs/x402-invariants.md. Every invariant is an immutable @dataclass(frozen=True) with constructor validation; compose() merges bodies under a single top-level G(...) to satisfy Tau's prohibition on nested temporal quantifiers.

Security properties (proven via tests)

Four attacks, four defenses, four passing property tests in tests/test_verdict_signer.py:

Attack	Defense	Test
Attacker flips permit ↔ reject header	HMAC-SHA256 over canonical payload	test_tampered_outcome_fails_verify
Attacker replays $10 permit against $10,000 intent	intent_hash binding checked before MAC	test_replay_against_different_intent_fails
Attacker replays expired permit	TTL + issued_at enforced	test_expired_verdict_rejected
Hostile payment intent injects Tau syntax	Safe-token regex on every field	test_provider_injection_attempt_rejected

What we're NOT claiming

[!CAUTION] No software-only policy engine can defend against the four attacks below. Anyone who tells you otherwise is selling a prayer. We shift the attack surface from "silent runtime drift" (where you're breached and don't know for months) to "spec review" (where humans can catch mistakes). That is the trade we are offering.

Out-of-scope attack	Your responsibility
Rule authorship errors ($10,000 when you meant $1,000)	Review processes, spec testing, change management
Trusted Computing Base compromise	Process isolation, binary attestation, runtime monitoring
Side channels not modeled by the rules	Expand the spec to cover the channel
Info-leak via attributed verdict rule_id	Redact or generalize for external-facing contexts

[!IMPORTANT] Compliance lead reviewing this for AI-Act or SR 11-7 posture? The 10-slide pilot deck walks your team and engineering leads through an 8-week paid pilot with written success criteria (zero false blocks on existing-approved traffic, ≥1 silent issue surfaced in your current rule set, operator TTI < 30 minutes). Deck: docs/pitch/enterprise-pilot-deck.md. Pilot call: [e-mail].

The standards story

The project is the reference implementation of a proposed RFC: Formal Policy Semantics for HTTP 402 Agentic Payments (draft-00, ~350 lines).

Normative requirements include: decidable rule-set composition; temporal fragment (LTL + past-LTL over atomless Boolean algebras); signed verdict envelope with intent binding; published TTI measurement protocol; IANA-registered X-Policy-* header namespace.

Targeted at the x402 Foundation working group and the Stripe+Tempo MPP spec team.

Evaluation ladder. Pick your time budget

You have	Read
5 minutes	RFC Abstract + Motivation. Decide if this is your lane
30 minutes	RFC §4 Rule Model + §6 Wire Envelope. You can critique the design
2 hours	Run the demos + full RFC + docs/mappings/x402-mapping.md. You can recommend it or tear it down

Comment on the draft

• GitHub Discussions on this repo (opens at public launch of the repo)
• The x402 Foundation WG mailing list when the WG charter ratifies. Track Linux Foundation's x402 Foundation page
• Direct: [e-mail]. We reply to every RFC comment within 48 hours

The TTI metric

Time-to-First-Correct-Rejection (TTI). The operator-ergonomics metric this project cares about. Every prior formal-methods tool (TLA+, Alloy, SPIN, Coq) is technically superior to OPA/Cedar and still lost every adoption battle because operators could not onboard in under an hour.

The RFC defines a shared test battery including contradictions that weaker engines silently accept. A vendor cannot win TTI by shipping an init wizard over strictly weaker semantics. The battery lives at specs/tti_battery/; the measurement script lives at scripts/tti-bench.sh. CI gate blocks any PR pushing TTI above 15 minutes.

Development

git clone https://github.com/Jme-Tau/tau-x402-cage.git && cd tau-x402-cage
pip install -e ".[dev]"

pytest tests/                              # 152 tests, ~10s
pytest --cov=adapter --cov=daemon tests/   # coverage ≥ 80% enforced in CI
pip install -e ".[saas]" && pytest saas/   # optional SaaS wrapper tests
bash scripts/tti-bench.sh                  # clean-machine TTI benchmark
docker build -t tau-x402-cage:local .      # Python-only slim runtime

The runtime image builds against python:3.12-slim (~216MB). The Tau SMT backend is a V0.3.1 capability and will be re-introduced as an optional base image once the upstream publish path lands.

Licensing & commercial

Dual-licensed. Full terms in LICENSE, pricing and tier breakdown in docs/pricing.md. Commercial contact: [e-mail].

Citations

• Methods and systems for computation using a decidable temporal logic. US Patent 12,254,082, granted 2026-03-18. Assignee: IDNI. Inventor: Ohad Asor.
• Asor, O. Guarded Successor: A Novel Temporal Logic. arXiv:2407.06214, 2024. https://arxiv.org/abs/2407.06214
• Formal Policy Semantics for HTTP 402 Agentic Payments. RFC draft-00 (this repo)

Roadmap

• V1 (now). 5 decidable + 2 enum-partial invariants; x402 primary, MPP mock for parity; signed envelope + intent binding; RFC draft-00.
• V1.1. Full string matching in JurisdictionRule / CounterpartyConstraint when nlang BA lands upstream.
• V2. Self-reference / NSO fragment: specs that reason about other specs.
• V2.1. MPP wire mapping first-class once an MPP spec-author conversation lands.
• V3. Asymmetric Ed25519 signing for multi-party attestation.

Status

[!NOTE] This repository is currently private. It goes public at the moment the RFC is submitted to the x402 Foundation working group. If you were given access early, please do not forward the link. Email [e-mail] to loop someone in and they'll receive the same early access.

Track	State
Code	V1 shipped, 89 tests, p95 < 10ms
Standards	RFC draft-00 drafted, targeted at x402 Foundation WG
Upstream (Tau Language)	3 patches contributed, end-to-end verified
GTM	Pricing + pitch deck locked; outreach in progress

Contact

If you're building on x402 or MPP and want the policy layer solved, we'd like to talk.

• General + commercial: [e-mail]
• RFC comments: [e-mail], or the x402 Foundation WG mailing list when it opens
• Security disclosures: [e-mail] (PGP on request)

---

[!WARNING] The section below is for Ohad + Fola + IDNI internal review only. Delete the entire block between the BEGIN INTERNAL / END INTERNAL HTML comment markers before making the repo public. Covers monetization, go-to-market, decision log, and open asks. Not for external eyes.

🔒 IDNI INTERNAL. Business review: monetization, GTM, open decisions

Why this section exists

This README is the outward-facing artifact of the 90-day Tau + x402 standards-capture program. Before the repo goes public (targeted at x402 Foundation WG RFC submission), Ohad and Fola should have a single place to review the business envelope: how the project earns money, what traction targets it's aiming at, what strategic decisions are locked, and what's parked for your sign-off.

Full planning context lives in the companion project:

• ../Tau-Safety-Cage/docs/plans/2026-04-18-001-feat-tau-x402-standards-capture-plan.v2.md. 12-week plan v2
• ../Tau-Safety-Cage/docs/review/2026-04-18-plan-review-findings.md. 6-reviewer persona findings
• ../Tau-Safety-Cage/docs/validation-parking-lot.md. 5 deferred validations with hard gates
• ../Tau-Safety-Cage/docs/briefings/2026-04-18-x402-standards-capture-briefing.md. Origin briefing

Strategic position in three sentences

The agentic-payment rails (x402, MPP) just went neutral (Linux Foundation April 2; Stripe+Tempo March 18). Neither spec defines formal policy semantics, and no competitor has shipped decidable composition + signed verdicts. We have ~90 days to get Tau named as the reference semantics in the x402 Foundation RFC (and ideally the MPP spec too) while the window is open.

Strategic decisions already locked (D1–D6)

These were reviewed via /document-review against six persona reviewers (product, feasibility, adversarial, scope, security, coherence); findings folded into plan v2.

Ref	Decision	Outcome
D1	Architecture commitment	Hybrid. 5 invariants emit pure Tau, 2 (JurisdictionRule, CounterpartyConstraint) use enum encoding pending nlang BA unblock upstream
D2	Self-reference V2 first vs x402 port first	x402 port first. Self-reference V2 deferred to post-window; the satisfiability-on-composition check carries the declare-time contradiction demo in V1
D3	Block on Ohad/Fola strategic conversation first	No. Proceed with patches + validation evidence in hand. Soft checkpoint scheduled Week 3-4 for your sign-off on the six commitments below
D4	Full premise-validation sprint before execution	Skipped; parking lot written. Items 1 (patent claim read) + 2 (LF IPR audit) remain hard gates. See "Open asks" below
D5	Monetization framing	Dual narrative. Commercial lead is hosted SaaS + enterprise services + certification (what customers actually buy). Patent is moat sustainability, not revenue line. Red Hat analogy reserved for technical + strategic audiences only
D6	Graduation criterion to option C (full company plan)	Tightened: (R5 OR R6 OR R7) AND R2 AND R3 + second-touch quality gate. Self-issued artifacts alone (TTI + RFC + 10k impressions) cannot graduate the program. We require at least one external pull (TSC conversation, MPP author conversation, or pilot conversation)

Monetization. The four-tier model

Full detail: docs/pricing.md.

Tier	Price	Customer	Margin shape
OSS self-host	$0	Indie devs, hobby projects, early-stage startups pre-pilot	Adoption wedge; no direct revenue but required for category adoption
Hosted SaaS	8–12 bps of transaction value, per-transaction capped (tiers $99 – $2,999 / month + volume)	Startups + fintech engineering orgs running policy at scale without operating the cage themselves	SaaS gross margin 70–85% typical; anchor: Chainalysis ~10bps, Stripe Radar ~2bps
Enterprise self-host	$100k – $1M/yr flat + volume	Regulated Fortune 500 (banks, pharma, defense, payments incumbents)	Enterprise-software margins; small customer count, high ACV
Conformance certification	$10k – $50k per audit	Other policy engines wanting RFC-conformance badge	Services margin 50–60%

Revenue projection (honest, not aspirational)

Year	Tier 2 SaaS	Tier 3 Enterprise	Tier 4 Cert	Total
2026	$0 – $100k	$0 – $400k	$0	$0 – $500k
2027	$200k – $1M	$1M – $8M	$100k – $500k	$1.3M – $10M
2028	$2M – $20M	$10M – $60M	$1M – $5M	$13M – $85M
2029–30	$20M – $200M	$50M – $500M	$5M – $50M	$75M – $750M

Driven by: x402 + MPP volume projection ($120M/yr 2026 → $10B+/yr by 2028 if Galaxy Digital's "30% of Base daily txs" holds); McKinsey $3–5T agentic commerce TAM by 2030; Tier 3 lands 3 customers in 2027, 10 by 2028, 30+ by 2030 at $500k–$1M average.

90-day standards-capture program. What we're running

Horizon: 2026-04-18 → 2026-07-17 (12 weeks)

Three parallel tracks:

• Code. V1 MVP shipped (89 tests green; full adapter + mock facilitator + demos; Tau patches verified)
• Standards. RFC draft-00 written + x402 wire mapping + three audience one-pagers; LF membership application drafting
• GTM. Outreach vault at 2-AREAS/IDNI/x402-outreach/ with contacts.md + pilots.md + sequence.md + crm.md templates; Wave 1 sends target Week 6; Wave 2 target Week 9

Graduation (Week 12) requires all three of:

1. R2. Published TTI benchmark <10 min, reproducible from public script, with ≥1 external reproduction
2. R3. RFC published as working draft with stable URL, linked from ≥1 public thread
3. At least one of R5 (TSC voting-member conversation with second-touch), R6 (MPP spec-author conversation with second-touch), R7 (regulated-enterprise pilot conversation with second-touch)

Second-touch qualifier (not just polite Zooms): counterparty initiates follow-up, requests the RFC, forwards it internally, OR proposes a next meeting with defined agenda.

Outreach plan

Wave 1. Standards audience (Weeks 6–8). 15–20 contacts across x402 Foundation TSC backer orgs + MPP spec authors. Templates at ../../2-AREAS/IDNI/x402-outreach/sequence.md. Contact research + scoring at ../../2-AREAS/IDNI/x402-outreach/contacts.md (names TBC. Your network will fill these faster than ours).

Wave 2. Enterprise pilots (Weeks 9–12). 5 regulated Fortune 500 candidates across banks / pharma / defense / agentic SaaS / Circle-Coinbase enterprise. Pitch deck at docs/pitch/enterprise-pilot-deck.md. Pilots are 8-week paid with written success criteria and half-refund if we don't surface a silent issue in the customer's existing rule set.

Cold-outreach limitation: Regulated Fortune 500 rarely respond to cold email. Warm-intro mapping through Frontier / Tau Net DAO / your networks is the critical-path dependency for R7.

Tau Language upstream contributions

Three bugs patched and verified via Python binding during the project:

• Bug 1. solve_ltl_aba pair-consistency over-rejection (5-line patch to src/ltl_aba.tmpl.h)
• Bug 3. Past-LTL O / H operators absent from grammar (8 files, first-class nodes lowered to T S φ via existing Since-elimination pass)
• Bug 4. & chains fail parser at wff level (1-line grammar alias for &&)

Both patch bundles at ../Tau-Safety-Cage/docs/upstream-fix-*.patch + bug reports at ../Tau-Safety-Cage/docs/tau-lang-bug-report*.md. Ask: please review before we open the public GitHub PR. Upstream-style bug reports drafted; submission sequence is v1 first (pair-consistency, already deployed locally) then v2 (grammar additions for O/H + &).

Open asks. Your sign-off required before these steps

Listed in execution order. Each blocks a downstream milestone.

#	Ask	Blocks	Timing
1	Patent claim read. IP counsel reads independent claims of US 12,254,082 and maps against the 7 commercial-moat claims in briefing §3. If any are narrower than assumed, Red Hat monetization story needs reframing.	RFC submission (R3); any written licensing claim	Before Week 4
2	LF x402 Foundation IPR policy audit. Read charter + contributor agreement. If policy mandates royalty-free on all essential claims (W3C-style), contributing RFC as a member auto-grants what we want to license.	LF membership application (Unit 2.7)	Before Week 5
3	Sign-off on the six strategic commitments. RFC under IDNI name; LF membership application as IDNI; public TTI benchmark commitment; royalty-free patent grant carve-out (conditions a/b/c in LICENSE); pricing tiers as published in docs/pricing.md; public repositioning from "DAO project launcher" (Tau Net deck) to "agentic payments policy" as IDNI's commercial front in 2026	All public outreach (Wave 1 + 2)	Week 3-4 conversation
4	Warm-intro leads from your network. Any relationships at Stripe / Tempo / Paradigm / Visa / Mastercard / Coinbase / Circle / x402 Foundation TSC voters. Warm paths convert at 20-40× cold-email rates; we have none.	Wave 1 quality (R5 + R6 conversions)	Before Week 6
5	Warm-intro leads for regulated-enterprise pilots. Any relationships at banks / pharma / defense / Fortune 500 AI-governance leads. R7 is structurally unreachable in 90 days without at least one warm path.	Wave 2 R7	Before Week 8
6	Bug-report review + upstream PR approval. Read ../Tau-Safety-Cage/docs/tau-lang-bug-report.md + v2 + both patch files. Patches verified locally; want your eyes before public submission.	Upstream credibility (signals to x402 Foundation that we're real Tau contributors, not just consumers)	Before Week 6

Graduation signal. What success looks like at Week 12

• Graduate to Option C (full company plan + fundraise conversation): R2 ✓, R3 ✓, AND (R5 ✓ with second-touch OR R6 ✓ with second-touch OR R7 ✓ with second-touch)
• Continue iterating on B (90-day program extended 90 days): some but not all of the above; TSC or MPP WG has acknowledged the RFC exists; no pilot conversation landed
• Pivot: zero external pull by Week 12 despite shipped code + RFC + demo reel → reframe the project or reassign IDNI resources

Questions I'd want answered in the Week 3-4 conversation

1. Is "agentic payments policy" the right commercial identity for IDNI in 2026, or does it overshadow the DAO-launcher story in the Tau Net deck?
2. Willingness to file a patent-grant carve-out for conforming OSS implementations (condition a/b/c in LICENSE)? Loses some patent optionality; buys standards-track credibility.
3. Who's the Tau Net brand voice on this? IDNI account, your personal Fola/Ohad channels, or a new @tau_x402_cage X handle?
4. Do we announce the project publicly (X + HN + blog) on RFC submission, or keep private until first TSC conversation lands?
5. What's the "no" criterion? At what signal do we stop this line of work and pivot to the self-reference V2 demo?

How to respond

Reply to this section directly via GitHub PR, email me at [e-mail], or raise in our next 1:1. Happy to walk through any of the above live. The 90-day clock is running but the Week 3-4 conversation is the natural checkpoint before anything irreversible.

. IDNI, 2026-04-18