Touchstone — Build Plan & Roadmap

Status note (2026-08-21): the Phase 0/Phase 1 calendar below is a historical planning record, not current deployment or operations state. Statements such as “mainnet is unscheduled,” hand-started-only publication, and the scripted two-act demo describe the plan when written and have been superseded. Current verified state is recorded in site2/_data/facts.json, site2/data/stats.json, docs/OPERATIONS.md, and docs/SUBMISSION-DRAFT.md.

Name chosen by the owner on 2026-08-13: Touchstone — the stone assayers rubbed gold against to verify what it really was before anyone accepted it. That claim was corrected on 2026-08-14: existing crypto projects do use the name — see docs/BRAND-CLEARANCE.md, which records a Morpho vault curator branded Touchstone for gold-RWA lending (announced 2026-07-15, contracts live on Ethereum) and a PyPI SDK touchstone-verify implementing signed, hash-chained disclosure verification (2026-07-12). Known adjacent mark: Touchstone Investments (Cincinnati mutual-fund/ETF firm, Western & Southern) — different business (asset management vs. verification software), but a trademark opinion from counsel is required before commercial launch. DNS quick-check: touchstone.finance, touchstonelabs.xyz, touchstone-rwa.com show no DNS records (registrar confirmation pending); touchstone.xyz and usetouchstone.com resolve and are likely taken. Domain purchase, handles, and any announcement remain owner actions. (Prior internal codename ARGUS retired — collision with Cobo Argus.)

Mission. Touchstone is the policy and control plane for tokenized assets. It turns issuer, regulator, custodian, oracle and onchain evidence into explicit policy for the contracts, wallets and AI agents that act on it. The current USTB vertical begins with issuer-published evidence: AI proposes cited, machine-checkable controls; deterministic surveillance evaluates them; accountable humans approve policy; signed results are published to X Layer for enforcement.

Grant wedge. Touchstone makes RWA liquidity conditional on verifiable evidence.

Touchstone does not issue assets, custody funds, recommend investments, assign credit ratings, or claim facts beyond the evidence class it has actually verified.

The sentence the project is built around:

"I accepted this RWA because its required evidence was current, attributable, and independently verifiable through Touchstone." — a future X Layer application

Product principles

1. Evidence classes are never inflated. An issuer webpage is issuer disclosure — not independent attestation. An API observation is not proof of reserves. Retrieval failure is not issuer failure.
2. AI proposes meaning; deterministic systems control state. AI compiles evidence into typed controls. Schema validation, approved policies, and deterministic evaluation decide registry transitions.
3. Every accepted claim is traceable to an exact evidence span, source URL, artifact hash, retrieval time, compiler version, and control version.
4. Abstention is a valid result. UNVERIFIABLE is more valuable than a confident hallucination.
5. Onchain state must have a consumer. The registry is incomplete without a contract that admits, pauses, or rejects based on its state.
6. Corrections are append-only. Reports may be superseded, never silently rewritten.
7. Public claims remain conservative. Touchstone reports what it observed and verified — not whether an asset is safe, compliant, solvent, or suitable.
8. Every deployment, post, grant application, partnership message, and submission requires owner approval.

System architecture

Official issuer sources
        │
        ▼
Allowlisted isolated fetchers
        │
        ▼
Content-addressed evidence store
        │   raw bytes + source metadata · exact hashes + retrieval time
        │   immutable version history
        ▼
AI control compiler
        │   typed candidate control · exact supporting span
        │   confidence + abstention · compiler/prompt/model version
        ▼
Deterministic validator + approval gate
        ▼
Versioned approved control set
        ▼
Source-specific surveillance adapters
        ▼
Deterministic evaluator
        ▼
Signed observation report
        │   Ed25519 signature · hash-chained transparency log
        │   offline-verifiable bundle
        ▼
X Layer registry
        │   control-set root · evidence root · status + freshness
        │   immutable transition history
        ├───────────────┐
        ▼               ▼
   AssetGate      Living dossier

Category and product boundary

Touchstone is not an oracle, credit rating or legal-compliance oracle. It determines whether a specific, predeclared evidence policy is currently supported and gives a consumer an enforceable answer. Price and NAV providers answer what value was reported; custodians answer how an asset is held; permissioned-token standards answer whether an investor or transfer is eligible. Touchstone answers whether the complete evidence policy required for this action currently holds.

The long-term product spans the full policy lifecycle, while liquidity admission remains the first commercially legible application:

Asset onboarding
      ↓
Evidence collection
      ↓
Policy compilation and accountable approval
      ↓
Market, collateral or agent admission
      ↓
Continuous surveillance and automatic suspension
      ↓
Recovery, reactivation, audit and governance

The north-star platform has eight connected layers. They are a sequence, not a claim that an enterprise suite exists today:

No layer graduates because its UI exists. It graduates when an independently operated consumer uses it, its failure mode is measured, and its authority boundary is auditable.

Evidence status entering Phase 0 (as of 2026-08-13)

Preliminary probes are encouraging but do not replace the formal audit:

• PAXG (Paxos): transparency page fetched directly — monthly attestation reports, KPMG LLP since Feb 2025. Direct repeatable PDF retrieval from the deployment environment unverified.
• OUSG / USDY (Ondo): secondary sources report end-of-business-day NAV updates, daily third-party reporting, daily Ankura Trust attestations for USDY. Exact endpoints and historical accessibility unverified.
• USTB (Superstate): documented public NAV/holdings API. Whether every required endpoint is unauthenticated and stable from the VPS unverified.
• BENJI/FOBXX (Franklin Templeton): official page exposes prospectus material and daily yield fields. Stable machine retrieval unverified.
• RWA.xyz may corroborate but is never the authoritative source for controls.

Phase 0 — Evidence and brand kill gates (Aug 13–14)

Purpose: prove Touchstone can monitor real evidence before any product code expands.

Source audit of PAXG, OUSG, USDY, USTB, BENJI + one spare. Per source, record: asset identity and canonical contract addresses; publishing legal entity; first-party URL; evidence class; MIME type and true file signature; login/cookie/JS/anti-bot requirements; retrieval from the production VPS (repeatedly — browser-only access is insufficient); redirect behavior and URL stability; publication cadence, timezone, holidays, grace period; version semantics; historical-version availability; terms-of-use constraints; observable structured fields; proposed controls with exact evidentiary limits; failure behavior.

Asset acceptance gate. An asset enters the build only with: an attributable official source; repeatable no-login retrieval; ≥2 honest machine-observable controls; explicit update semantics; a defensible asset-identity mapping; hashable, retainable evidence; an adapter feasible within the sprint; no manual-download dependence.

Portfolio selection: one hero asset (strongest machine-readable daily evidence — provisional order: USTB, then USDY, then OUSG/BENJI), one second daily asset (cross-issuer repeatability), one contrast asset (likely PAXG — monthly cadence and staleness semantics). PAXG is never the hero: monthly evidence may not change in the judging window.

Deliverables: SOURCE_AUDIT.md; accepted source manifests; rejected candidates with reasons; raw sample artifacts + hashes; proposed controls; cadence and failure rules; evidence-class glossary; initial threat model; brand-collision report. No issuer-specific negative findings published at this stage.

Abort conditions. Reject any asset requiring fragile interactive sessions, aggregator-only data, unknowable cadence, speculative legal interpretation, unretainable evidence, uncertain contract mapping, or controls claiming more than the source proves. Abort Touchstone entirely if fewer than two candidates pass. Do not preserve the concept by lowering evidence standards.

Brand clearance for "Touchstone" (deadline Aug 14, before the project X account opens): completed 2026-08-14 and recorded in docs/BRAND-CLEARANCE.md. The earlier 2026-08-13 finding that no existing crypto/web3 project uses the name was wrong — two live projects use it, one of them in this project's own verification lane; Touchstone Investments (US mutual-fund/ETF firm) is the known adjacent financial mark, different service class but requiring a counsel trademark opinion before commercial launch. Remaining checks: USPTO, WIPO, EUIPO/TMview, Nigerian and operating-market registries, GitHub, npm/PyPI, CoinGecko/CoinMarketCap/DefiLlama, explorers, X/Telegram/Discord/Farcaster/LinkedIn handles, app stores, ENS; phonetic and visual similarity. DNS quick-check found touchstone.finance, touchstonelabs.xyz, and touchstone-rwa.com unresolving (registrar confirmation pending); touchstone.xyz and usetouchstone.com likely taken. Domain purchase and handle registration are owner actions.

Phase 1 — Hackathon build (Aug 13–21)

Dependency order (corrected): evidence audit → state and control semantics → golden fixtures → contract interface → live source adapter → compiler/evaluator → registry publishing → consumer gate → product UI → operations hardening. Contracts encode settled semantics; they are not designed before the evidence model.

Aug 13 — Evidence kill gate

Audit sources; select hero/second/contrast; capture real artifacts + metadata; define allowed evidence classes; start brand clearance; minimum repo structure for audit artifacts. Exit gate: hero source retrieves repeatedly from the intended server; ≥2 honest controls identified; one historical observation exists for comparison; no unresolved ambiguity about what the source proves.

Aug 14 — Semantics, fixtures, threat model

Freeze control-language v0 before contracts. Control fields: asset_key, control_id, control_version, predicate_type, subject, source_id, source_authority_class, evidence_span, cadence, grace_period, observation_adapter, comparison_operator, expected_value, effective_from, effective_until, compiler_confidence, approval_state. Asset states: CONFIRMED / STALE / INCONSISTENT / UNVERIFIABLE. Operational events kept separate: RECONFIRMED / EVIDENCE_CHANGED / SOURCE_ERROR / CORRECTION_PUBLISHED. A transient outage is SOURCE_ERROR, never instant inconsistency; only an expired evidence deadline moves status toward STALE. Also: golden evidence fixtures, expected compiler outputs and abstentions, state-transition test matrix, adversarial-document tests, contract storage model, UI wireframe, submission-copy skeleton.

Aug 15 — Contracts and testnet foundation

TouchstoneRegistry: canonical asset key, control-set root, evidence root, status, observedAt, validUntil, publisher, monotonic sequence, report URI. Protections: publisher authorization, replay protection, monotonic enforcement, chain/domain separation, correction events, key rotation, no custody or value-transfer logic. AssetGate: allowed statuses, maximum observation age, required publisher, optional control-set version. Tests cover every transition, stale boundary, unauthorized publisher, replay, correction, rotation. Deploy fixture-driven system to X Layer testnet only after tests pass.

Aug 16 — Hero compiler and source adapter

Full hero ingestion path: allowlist → raw retrieval → content hash → normalization → exact evidence-span extraction → AI candidate control → schema validation → confidence gate → explicit abstention → deterministic observation evaluation. Security: all documents are adversarial input; the model gets no shell, network, wallet, or contract tools; instructions embedded in evidence cannot self-approve a control or move it into evaluation — narrowed 2026-08-15: steering is a stated residual, because a well-formed injected candidate is accepted as a proposal and only the approval gate stops it (see docs/THREAT-MODEL.md T9 and R-9); allowlisted URLs, blocked redirects; MIME/magic-byte/size/decompression/page/time limits; isolated parsing worker; model ID, prompt hash, compiler version, input hash, raw output all recorded. Exit gate: one real source compiles into an exact cited control; ambiguous examples abstain; golden cases reproduce.

Aug 17 — Signed surveillance and onchain publication

Scheduled epochs; Ed25519-signed reports; hash-chained transparency log; offline verification bundle; idempotent submission; receipt tracking; restart reconciliation; failure recovery without duplicate publication; hero report on testnet; working AssetGate consumption. By end of day the narrow vertical works end to end — if not, stop adding assets.

Aug 18 — Product experience

Hero living dossier: state + freshness, accepted controls, evidence excerpts + hashes, "what Touchstone verified" vs "what it did not verify", transition timeline, explorer links, offline-verifier download, evidence-grounded "why this state?" interaction, developer integration page. No wallet required to browse. Second source only after the hero flow is polished.

Aug 19 — Coverage and operations

(Amended 2026-08-16: the second daily asset and the contrast asset are cut. See the completion-metrics table — the two-adapter target is recorded as missed, not retargeted.) Public heartbeat; source health; last successful epoch; next scheduled epoch; gas runway; incident history; watchdog; alerting; backup/restore; production-like deployment rehearsal. Contract interface frozen at end of day.

Aug 20 — Hardening and release candidate

Full contract suite; pipeline tests; golden compiler evaluations; adversarial-document tests; browser smoke tests; restart/recovery; duplicate-publication tests; source-outage simulations; key-rotation rehearsal; restore rehearsal; full hero-demo rehearsal. Prepare: verified addresses, release commit, deployment manifest, rollback procedure, public limitations, submission form draft, X posts (drafts). (Amended 2026-08-16: the line that stood here staged a mainnet deployment for approval on Aug 20 — before the Aug 21 testnet proof it was supposed to depend on. Mainnet is unscheduled and conditional on a proven testnet loop.) Stage the testnet replacement-registry deployment package and request owner approval.

Aug 21 — Testnet release candidate (internal deadline 18:00 UTC)

Amended 2026-08-16. This date previously scheduled a mainnet deploy and production canary. That was not a credible plan and contradicted this repository's own owner gates, which permit mainnet only "after a proven testnet loop" — and nothing had then published to testnet at all. Scheduling an action the same documents forbid is how an operator ends up taking it.

Aug 21 is now a testnet target: prove the loop end to end on X Layer testnet — deployed registry, one autonomous live epoch, an offline-verifiable bundle, and a consumer contract gating on the published state. After explicit approval only, and no feature work on the day.

Mainnet is unscheduled. It is conditional on the completed testnet loop above being proven, not on a date, and returns as a separate owner decision.

Hackathon completion metrics

Three real assets documented; ≥2 fully autonomous live adapters; ≥6 accepted controls; 100% of accepted claims span-cited and hash-bound; zero unproven production claims; one live consumer contract gating on Touchstone state; every transition independently verifiable; all contract authorization boundaries tested; public interface wallet-free; watchdog + incident history + reconciliation operational; every scheduled epoch through Sept 1 completed or publicly recorded as an incident.

Current completion counts. Counts below render from repository data where available. The targets are not restated downward; what is missed is recorded as missed.

The minimum hero demo (90–120 seconds, two acts)

Act 1 — AI makes disclosure executable. Judge opens the hero dossier (no wallet); sees source, evidence class, limitations; runs "Compile disclosure" on the labeled testnet demo path; Touchstone fetches the official source; AI extracts one bounded commitment into a typed control; the UI highlights the exact supporting span; deterministic validation confirms span-in-artifact, schema, source identity, operator, confidence; the accepted control root publishes to testnet. If the compiler cannot support the control, it abstains live — never hidden.

Act 2 — current evidence changes contract behavior. A demo AssetGate shows REQUIRES_REFRESH (its freshness requirement is deliberately stricter than the last observation age); Touchstone retrieves the current official daily observation; UI shows previous vs current values (an ordinary NAV movement is never called a risk event); report signed; transparency log appends; freshness renews on X Layer; the gate flips to ACCEPTED; judge opens the explorer transaction; judge asks "why did the gate accept this asset?" and the dossier answers solely from accepted graph nodes. The gate reacts to verification freshness, never declares the asset safe.

Pre-staged: deployed contracts, publisher authorization, funded account, verified adapters, previous genuine evidence snapshot, previous signed epoch, approved control set, demo consumer, hash-bound cached fallback artifacts, recorded fallback walkthrough. Never pre-staged: modified issuer evidence, fabricated reports, synthetic NAV changes presented as real, hidden manual state changes, self-funded activity described as user activity. If the live source fails, switch visibly to verified historical replay mode and log the incident.

Aug 21 – Sept 1 operations calendar

Daily: poll each source per its documented window; offchain source-health checks; one signed observation per asset; anchor fresh observations on X Layer; one aggregate daily epoch root; heartbeat update; honest RECONFIRMED/EVIDENCE_CHANGED/failure surfacing; onchain reconciliation; gas and watchdog checks.

Reliability objectives: 100% of scheduled epochs complete or produce a public incident; 99.5% availability in the window; alert ≤5 min on daemon failure; safe auto-restart ≤15 min; no duplicate registry sequence after restart; daily encrypted backup; one tested restore before submission and one during the window; gas runway through Sept 3+; no incident ever removed from public history.

Phase 2 — Product proof (Sept–Oct 2026)

Objective: prove teams other than Touchstone want to consume the primitive.

Deliverables: 8–10 production-grade assets across ≥3 RWA categories (same evidence standard — never a vanity count); verification API v1; Python and TypeScript SDKs; webhooks/event subscriptions; commissioned rescans via x402; integration sandbox; external contract security review; initial legal review (research language, source terms/copyright, financial-promotion boundaries, liability and corrections, issuer naming/trademark); methodology, conflict-of-interest, correction, and appeals policies; two X Layer design partners.

Graduation metrics: ≥8 assets at production evidence standards; ≥30 active controls; 100% citation-bound claims; 99.5% surveillance SLO over 30 days; corrections within 24h; two independent testnet integrations; one external team using AssetGate or the API publicly; first paid commissioned adapter, rescan, or freshness subscription.

First revenue comes from a protocol, wallet, or infrastructure team — never from issuers paying for favorable state. Issuer-funded integrations are disclosed and organizationally separated from methodology.

Checkpoint: Touchstone is a credible X Layer project when another builder publicly consumes its state. Deployment alone is not adoption.

Phase 3 — Open standard and institutional readiness (Nov–Dec 2026)

Deliverables: Control Language Spec v0.1; Registry Interface Spec v0.1; Evidence Bundle Spec v0.1; reference compiler + offline verifier; reference clients; formal threat model; independent smart-contract audit; public compiler evaluation set; published methodology-change process; publisher/key-rotation spec; xStocks-readiness research package (contingent on confirmed ecosystem direction — never presented as guaranteed); corporate-action control prototypes (separate from core); first non-Touchstone publisher sandbox; a NIST AI RMF / Generative AI Profile mapping for compiler governance; and an SSDF-based software-development control baseline. ISO/IEC 27001 remains a later management- system and certification path, not a badge earned by adding documentation.

Graduation metrics: 25 production-grade assets; 5 external integrations (≥2 live); 3 external spec/adapter contributors; zero unresolved critical/high contract findings; public compiler evaluation with measured precision and abstention (≥90% precision on supported control families; coverage may stay lower through abstention); 3 paying organizations or $25k cumulative revenue; one external publisher accepted in a non-material test environment.

Monetization added: metered API plans, enterprise freshness SLAs, integration and adapter fees, dedicated evidence archives, paid on-demand scans, protocol support contracts. The public dossier, base verification endpoint, specs, and offline verifier stay open.

OKX / X Layer strategy: monthly ecosystem contribution reports; position as listing-diligence and collateral-admission infrastructure (never the listing decision-maker); request technical feedback from X Layer RWA/devrel; grants only with owner approval; demonstrate how tokenized equities could consume disclosure and corporate-action controls; never imply OKX/X Layer/issuer endorsement without written confirmation; registry and reference consumer stay native to X Layer even as observed assets span chains.

Phase 4 — Verification network (2027 H1)

Objective: remove Touchstone as the sole publisher.

Deliverables: curated multi-publisher network with identities and performance history; independent observation comparison; challenge and correction interface; publisher quorum policies; conflict disclosures; disaster-recovery regions; adapter certification tests; issuer-side machine-readable disclosure tooling; multi-chain observation with X Layer as the canonical registry home. No staking, slashing, or token merely to claim decentralization — accountable independent publishers with observable performance come first.

Graduation metrics: 100 covered assets; 5 independent publishers; 10 production consumers; 99.9% SLO over 90 days; ≥2 externally maintained publisher implementations; challenge acknowledgement <4h; confirmed corrections <24h; 10 paying organizations or ~$250k ARR; no single publisher required by every consumer policy.

Phase 5 — The independent RWA verification institution (2027+)

Public positioning is "the open verification and admission layer for tokenized real-world assets" — never "the Moody's of RWA" (that implies subjective ratings, regulated activity, and authority not yet earned).

Deliverables: independent methodology and appeals council; formal legal/compliance structure; publisher accreditation; public methodology and model-risk reports; annual audits; cross-jurisdiction source policies; coverage across major RWA categories and origin chains; exchange, wallet, custodian, and DeFi integrations; an issuer disclosure standard designed for machine verification; institutional-grade historical evidence archive; model/compiler reproducibility program.

Long-term metrics: 500 production-grade assets; 10+ RWA categories; 10 origin chains; 25+ production consumers; 10 independent publishers; 99.95% SLO; published annual transparency and model-risk reports; majority of registry reports produced or corroborated outside the founding organization; 50 paying organizations or >$1M ARR; no unresolved critical integrity incident; demonstrable use in listing, collateral, wallet, or settlement workflows.

Defensibility — what compounds

The Solidity registry and dossier UI are cloneable; the moat is everything that accumulates: (1) the content-addressed historical evidence graph; (2) the corpus of accepted/rejected/corrected/abstained controls tied to exact evidence; (3) compiler evaluations built on real financial-document failure cases; (4) hardened issuer-specific adapters with known cadence and failure semantics; (5) the public operational record — uptime, correction speed, incident transparency; (6) integration depth in production consumers; (7) publisher reputation over time; (8) issuer relationships that yield machine-readable feeds without surrendering independence; (9) open-standard leadership — contributors converge on Touchstone interfaces even with other publishers; (10) trust earned through restraint — a public history of abstaining instead of overclaiming. A well-funded team can reproduce the first demo. It cannot instantly reproduce years of evidence lineage, correction history, and integration trust.

Security, integrity, and governance ladder

Immediate (hackathon): separated deployer / publisher / Ed25519-reporting / operations keys; no secrets in code, logs, bundles, or clients; onchain key rotation; monotonic sequences + replay protection; allowlisted adapters; sandboxed parsing; prompt-injection tests; full model/prompt version recording; signed release manifests; daily backups + tested restore; public status and incident history.

Before external production dependence: hardware-backed or multisig admin roles; independent contract audit; external pipeline security review; incident-response runbook; 99.9% SLO architecture; multi-region deployment; supply-chain controls; reproducible deployment; upgrade/rollback policy; independent legal review; published correction and challenge policy.

Before multi-publisher operation: admission criteria; conflict disclosures; key-compromise procedure; quorum and disagreement semantics; suspension rules; appeals; independent methodology governance; no token or slashing until real adversarial and legal requirements justify them.

Adopted from the incumbents (design audit, 2026-08-17/18)

Two rounds against Chainlink and RedStone. Round 1 audited their price-feed surface, which was the wrong opponent; round 2 found the products in this category. Ordered by what each closes, not by how large it sounds.

1. Root / derivative publisher hierarchy. From RedStone's TSSO, which co-signs single-source NAV with a cold root key for material change and a bounded hot key for routine updates. Closes routine-key compromise without giving the hot key authority over the control set. touchstone/publish.py, touchstone/keyring.py, TouchstoneRegistry.sol.
2. Chained publication. A report must commit to its predecessor's digest. Today report.py carries sequence and correction_of but no parent, and the registry enforces sequence monotonicity only — so an authorized publisher can place an unrelated report at the next sequence and nothing detects the splice. Sequence binds position; it does not bind content. Requires a registry struct change, so it lands in the next deployed version rather than being bolted onto the current one.
3. Explicit SOURCE_ERROR publication. The state exists in the transition model and has no runtime producer: a fetch failure ends the epoch instead. touchstone/epoch.py, touchstone/schedule.py, touchstone/controls.py.
4. Scheduled republication of unchanged state. TSSO re-signs a static value to keep it from reading as stale. Adopt the shape, not the numeric deviation threshold — a 0.1% band is meaningless over CONFIRMED / UNVERIFIABLE / STALE / INCONSISTENT.
5. Bind the evidence span to the adapter field it justifies (residual R-1). The span is presence-only today.

Deferred, deliberately: threshold co-signing. RedStone ship a quorum product with getUniqueSignersThreshold and an onchain median, and still concluded that a valuation with one authoritative source cannot be established by aggregation. Three signers reading the same issuer API agree on the same wrong answer. Quorum earns its place when there are publishers whose disagreement carries information, not before.

Watched, not adoptable: zkTLS (Chainlink DECO) is a public sandbox with no production product. It would prove transport — that these bytes came from that server — which is strictly stronger than the unauthenticated fetch this project records today. It would not prove typed interpretation, predicate evaluation, approval binding, or abstention.

Public-backing milestone arc (every item drafted, owner approves)

The owner's personal account carries the convictions; the project account carries technical progress. Deliberate, not promotional.

1. The commitment (after Phase 0 gates + provisional brand clearance): the problem, the evidence-integrity principles, the decision to build publicly. Never name issuers as partners or imply endorsement.
2. Evidence standard (after control-language v0): one exact citation becoming a typed control, plus rejected examples. Message: the hard part is not extracting a number — it is refusing to claim more than the source proves.
3. First X Layer enforcement (after the testnet full loop): source artifact → signed report → registry transaction → AssetGate decision → explorer link.
4. Living dossier (Aug 18–20): the public dossier and the verified / not-verified distinction.
5. Testnet release candidate (Aug 21): hero demo, testnet addresses, open verification endpoint, limitations, the @XLayerOfficial mention. Separate approvals for post and submission. (Amended 2026-08-16: mainnet is unscheduled and conditional on a proven testnet loop, so it cannot be a launch item on a fixed date.)
6. Reliability proof (Aug 28): scheduled vs completed epochs, evidence changes, incidents and recovery, uptime. Observation counts are never presented as adoption.
7. Snapshot report (Sept 1): full judging-window history, signed archive, roadmap.
8. External adoption (only when real): the first protocol consuming Touchstone state — the milestone that proves infrastructure status.

Ambition theater — banned phrasing and mechanics

Cut everywhere, permanently: "the Moody's/S&P of tokenized finance" (use "machine-verification and admission layer"); "regulatory-grade" before formal legal and audit work; stake/slashing before independent publishers and real dispute data; any governance token; issuer-paid ratings (Touchstone does not rate, and payment never affects status); asset-count vanity expansion; daily social posting as proof of life (the heartbeat and signed epochs are the proof); guaranteed xStocks alignment (say "ready to help if confirmed"); "every major RWA chain" before production consumers; automated legal conclusions.

Priority stack and never-cut list

Hackathon priority: 1) evidence integrity, 2) one flawless hero loop, 3) contract-enforced consumption, 4) autonomous reliability, 5) public transparency, 6) second/third asset coverage, 7) x402 and richer dossier interaction. If time slips, cut in order: x402 rescans → third adapter → semantic-diff visualization → rich Q&A. Never cut: exact citations, abstention, signed reports, the transparency log, AssetGate, end-to-end surveillance, public incident visibility, key separation, the Sept 1 operations plan.