Recommendation. Claus should prepare one agent identity and pilot useful inference for a first read of public Lab pitches when the rewarded testnet opens. Agent rules
Reward. Eligible settled purchases earn a share of the 1.2 billion FLOP agent pool, initially locked for inference use, with three spent unlocking one. Allocation terms
Timing. The working target remains Tuesday 3 November 2026 for about 90 days, with mainnet and TGE tracking to early February 2027. Launch FAQ
Condition. Say yes only after confirming the actual models, faucet limits, activity floor and attribution of settled spend to Claus’s account. Open attribution question
IMD. Each seat can hold a DID, but common control means one participant score, and separate scoring for independent owners using shared runners remains unresolved. Rules and runner question

Proposal research for Claus

Read on 20 October 2026. This report distinguishes published rules, observed public responses, inference and unanswered questions. Every FLOP participation, allocation, onboarding and reward rule below is Draft, including decisions described internally as ratified. No final public testnet rulebook was identified. Ratification within a research draft does not establish deployed behaviour. Observations of Claus, API schemas and vendor prices are facts about what the sources returned, rather than promises about future operation.

1. Timeline and source precedence

Draft. The launch FAQ, updated 20 October 2026, says the “working target is Tuesday 3 November 2026”. It gives about 90 days of testnet and mainnet, TGE and listing in Q1 2027, tracking to early February. TGE is simultaneous with mainnet. Adding 90 days to 3 November gives 1 February 2027, which explains the target without establishing an exact genesis date. The date depends on completing devnet onboarding. Forced upgrades with short deadlines are expected during testnet.

Draft. The testnet page still gives Q4 2026 and Q1 2027, allows extension for readiness, and places a finalized snapshot and an allocation review before genesis. Activity after the snapshot does not count. Testnet state itself does not migrate. The current testnet network profile remains planned, with no verified public chain specification, genesis hash, bootnodes, RPC, gateway, catalogue or supported release. This supports preparation today, not a claim that rewarded public participation is already available.

Observed public statements. Flop Labs on 6 October said “Testnet opens late Oct or early Nov for about 90 days”. Hayes on 11 October said “ship FLOP Testnet in a few weeks”. His 20 October post directed readers to Ask FLOP’s published documentation. The November FAQ is more specific than those earlier statements. No later postponement was found in the reviewed official pages and visible account posts. This is a bounded search finding, not proof that no announcement exists elsewhere.

Draft. Devnet earns no rewards or airdrop weight. The FAQ describes private operator admission and also contains inconsistent statements about agent access. Public gateway responses now demonstrate a devnet surface, as described below. Neither that surface nor older Technocore activity establishes rewarded testnet activity. Launch FAQ

Source conflicts matter. The whitepaper delegates participation details to the testnet and airdrop pages. Those pages call the Yellow Paper definitive but not final. The GitHub Yellow Paper and hosted edition differ despite sharing a draft version label. I directly checked that the hosted edition includes miner block rewards in unlock credit while the pinned mirror excludes them. Issue 165 also records the differing fee rules. Use the current dedicated participation pages as the proposal’s stated terms, but obtain a matching release and rulebook before relying on them financially.

2. What earns allocation

Every rule in this table is Draft, read 20 October 2026. Pool shares are conditional on eligibility and final conversion rules.

Role	Activity credited	Genesis allocation	Practical consequence
Agent	Compute purchased in settled sessions. Registration and balances alone do not qualify.	Proportional share of 1.2 billion FLOP. Nothing initially liquid.	Best fit for Claus if useful inference can be routed and credited to his identity.
Validator	Testnet ranking and admission to the genesis set.	Up to 1,000 seats, each granted a 1.2 million FLOP bond.	A competition for a bonded seat, with hosting and operations costs regardless of selection.
Miner	Verified compute served, rather than idle hardware or hardware purchase price.	Proportional share of 1.2 billion FLOP. One quarter initially liquid.	Requires GPU operation and demand. It is a separate undertaking for Claus.
Ecosystem participant	Separately specified creator, referral and builder programmes, with possible responsible disclosure rewards.	An 800 million FLOP reserve, rather than an automatic agent allocation.	No quantified entitlement for this proposal or an IMD integration was found.

Source: Airdrop sections 1 to 6. Its current miner summary permits the remainder to unlock against mainnet inference fees and block rewards. The mirror conflict noted above prevents treating that as a settled implementation fact.

Draft. Agent grants are “Spendable only on inference”. The same page says “Every 3 $FLOP of the locked balance spent in settled sessions unlocks 1 $FLOP”. There is no calendar vesting or expiry for unspent principal. The intended full use of a grant spends three quarters on compute and releases one quarter as liquid FLOP. It is not a promise of a liquid payment equal to all testnet spending, nor a fixed conversion rate from play tokens. Airdrop section 5

3. Agent route and join steps

The following steps describe the Draft route, read 20 October 2026. They are preparation and acceptance steps, not a claim that a complete public testnet runbook is available.

1. Create an Ed25519 key locally and retain its private key securely. Technocore’s signed identity format is did:key:z6Mk…. The ellipsis describes the format, not an identity to use. Its official repository documents signed room messages. A room signature is not automatically a chain transaction signature. Creating or posting a DID earns nothing by itself.
2. Establish the FLOP account and confirm how the DID binds to that account and the eventual genesis beneficiary. The simple participation page requires a DID and wallet; the newer FAQ says a DID key and HTTP requests suffice without installing wallet software. The underlying chain uses Substrate accounts. Claus’s existing Ethereum wallet is not automatically a FLOP signing account, and the advertised EVM account linkage remains unbuilt. Ask for an explicit binding and recovery procedure. Launch FAQ, chain FAQ and open identity question 123
3. Wait for the verified public testnet profile and obtain faucet tokens through the official route. The published design imposes limits per identity and says inference volume is bounded by that faucet allowance. The testnet grant amount is not published in the reviewed rules. The reset period is undecided, including whether claims are hourly or daily. Do not budget requests per day from a guessed allowance. Launch FAQ
4. Discover the actual served models and select a registered model and precision that meet the task’s quality and latency needs. Miners choose which models they load. Validators are intended to retain roughly 20 to 30 leading open models, but that is not a guaranteed launch menu. Proprietary frontier models are excluded at launch. The buyer selects on model, price, latency and region, and pins the weight hash. Demand FAQ
5. Submit a signed inference request through the supported client or gateway. The launch FAQ says the request is signed with the identity key; the unresolved DID binding must be handled by the supported procedure. The agent integration page lists model weight hash, maximum latency, compute in FLOPs, confidentiality flag and fee. The Yellow Paper Appendix C and section 12 instead describe selecting a miner and opening a funded compute channel. These flows need an implementation mapping; issue 142 asks for it. Set a budget and bounded signing authority using the deployed controls.
6. Complete useful work and retain the request identity, model hash, output, channel identity, receipt, settled amount and finalized chain evidence. Verify that the scoring account is Claus’s account. A successful HTTP response or estimated token usage alone does not prove eligible spend. The unresolved gateway attribution issue is material. Issue 172
7. Meet the published activity floor once released. Check the allocation snapshot during the review window and contest an incorrect entry through the announced procedure. Preserve keys for the genesis account. Use the locked mainnet grant for further inference; claim only amounts that have actually unlocked. Testnet rules and airdrop terms

Draft. What counts. The public participation rule is settled compute purchased. Yellow Paper R8.4 is more specific about exclusions: “completed job count and active days MUST NOT be scoring terms”. A faucet claim, sponsored bond, transfer or held balance does not create allocation rights. A useful task merely submitted, a failed request, an unsettled channel or chat activity is insufficient evidence. The precise minimum activity floors, conversion caps and aggregation method remain open in E.38. Hosted Yellow Paper section 8.2 and E.38

Draft. Settlement precision. Reserved channel escrow is the payment commitment, and R12.1a states “Under-use MUST NOT be refunded.” That makes the distinction between settled reservation spending and verified inference actually delivered important. Issue 32 asks which quantity receives agent credit. It discusses the ongoing agent reward leg and implications for conversion; it is an unanswered interpretation question, not an adopted loophole. For mainnet spend to unlock, R8.7 requires eligible locked principal in the session’s gross payable, finalized settlement, expiry of the seven day dispute window and resolution of disputes without an upheld fraud finding. Refunds, penalties and transfers do not earn that unlock credit. Hosted Yellow Paper sections 8.2 and 12.1

Draft. Common control. The testnet page says “wallets under common control are scored as a single participant”. It excludes demand paid to a miner under the buyer’s control and funds circulated among controlled wallets. Manufactured activity forfeits eligibility unless successfully challenged during review. More wallets do not establish more participants. These are eligibility rules even though E.38 and E.49 acknowledge missing runtime common control enforcement. Successful settlement does not prove economically independent demand. Testnet section 6 and Yellow Paper E.38 and E.49

Draft. Agent payments. The testnet overview advertises transactions between agents on Technocore, but the newer FAQ expressly leaves transfers of testnet play balances unfinalized. It separately describes transferable work vouchers carrying their unlock obligation. Neither statement makes arbitrary agent payments eligible inference purchases. Technocore chat and deal coordination are distinct from value settlement. Its repository and the linked TCLK project do not establish a live testnet payment rail. Obtain confirmation before proposing paid subcontracting between testnet agents. Launch FAQ, Technocore repository and open transaction question 133

Does an OpenAI compatible endpoint exist?

Observed, 20 October 2026. Yes, a FLOP branded gateway publishes an OpenAPI 2.0.0 schema, with base URLs https://api.flop.network/api/v1 and https://api.flop.network/v1. It advertises chat completions, Responses for Codex and Messages for Claude Code. This is an API claim and an observed schema, not a completed integration test or endorsement of any model’s tool competence.

Observed. Its health response identifies devnet. A read of the alias catalogue returned one model, qwen/qwen2.5-0.5b-instruct, mapped to Qwen/Qwen2.5-0.5B-Instruct. It advertised text input and output and max_tokens, not an embedding service. The bare /models route returned a commercial agreement gate error. Examples naming Llama in the schema are examples, not evidence that Llama is currently served.

Draft interface. The schema describes funded bearer credentials and signed wallet challenge enrollment on devnet. Its default new wallet allowance is 25 FLOP, scoped to a model and expiring within 24 hours. Reenrollment does not replenish it. That allowance is not a published recurring testnet faucet amount. The official testnet profile still lists its gateway as unavailable. Verify the endpoint’s official testnet identity, enrollment policy and end account credit before supplying funds or relying on allocation. No wallet was enrolled and no paid or free inference was submitted for this report.

4. Validator route and join steps

Draft selection rules. The current launch FAQ names uptime, block production, accuracy, latency and geographic spread across the testnet. It does not publish weights. Infrastructure diversity includes hosting providers and autonomous systems. The complete cutover rubric is unfinished. Keeping up with required software upgrades is also measured. Open participation during testnet does not guarantee a genesis seat.

Draft economics and limits. The network sponsors testnet stake. The top 1,000 at cutover receive a 1.2 million FLOP bond each, rather than a purchase requirement or foundation delegation. The grant is locked until the first halving, approximately two years after genesis, then released one validator per day under an order still to be fixed. Release does not unbond an active validator’s required stake. The specification states five slots per operating entity; the FAQ says this and a proposed 10 percent provider cap are not yet enforced and need reconciliation with institutional guidance. Treat five as the planning limit. Launch FAQ, airdrop terms and Yellow Paper section 15

Draft distinction. Testnet cutover scoring is not the same as continuing mainnet selection. The validator introduction describes a ratified target of stake ordering above availability and verification liveness floors, with roughly 50 validators rotating monthly. It also explicitly says the running rotation and cap enforcement are incomplete. Do not promise permanent membership or extrapolate an equal monthly income from a 1,000 seat illustration.

Draft hardware. The testnet page recommends at least eight CPU cores, 64 GB RAM, 4 TB NVMe, an absolute storage minimum of 2 TB and a redundant 1 Gbps connection. No GPU is required. The operator economics FAQ says eight physical cores, 32 GB ECC with 64 GB comfortable, 4 TB preferred and symmetric unmetered 1 Gbps. Yellow Paper section 15.3 adds at least 3.4 GHz, SMT disabled and enterprise NVMe. These are inconsistent provisional profiles, not interchangeable certified minimums. Issue 128 remains open on the discrepancy.

All join rules below are Draft, read 20 October 2026.

1. Register interest through the official validator interest form linked from the testnet page. The FAQ says the form is an optional contact list, not approval or a reserved seat. Private devnet rehearsal currently needs operator coordination and earns nothing.
2. Confirm the final hardware profile, scoring rubric, entity declaration, provider limits and sponsored stake process before committing to rent. Choose a region and provider with useful diversity, sufficient usable disk and measured connectivity.
3. Obtain pinned node and DA images, a chain specification, independently published expected genesis hash, bootnodes and actual testnet admission instructions. The validator manual lists these inputs. Its existing procedure is explicitly disposable devnet, with administrator funding and promotion. It is not a working public testnet installer.
4. Generate separate account, node and consensus session keys under the supported release. Keep backups, persistent volumes and private RPC access. Register the economic account and session keys through the actual release’s procedure. The specification’s validators.register queues the candidate; it does not immediately activate it. Do not copy development keys into a public network. Validator manual and Yellow Paper section 15.2
5. Verify actual activation, BABE authoring, AlephBFT finality, DA custody and serving, plus accepted verification duties. Monitor the ranking and upgrade deadlines throughout the testnet. Saving a container log is not proof that the validator was active or scored. Validator manual
6. Review the cutover result, account and granted bond. Accept the mainnet slashing, ongoing stake and operations obligations only after inspecting the final runtime and release terms. Airdrop terms and validator introduction

Draft key answer. Consensus operating keys can differ from the economic account: the manual separately handles account and DA credentials, node identity and BABE and Aleph session keys. That does not establish a selectable, unrelated reward recipient. The economics FAQ says a separate payout address is being added in its miner fleet discussion. I found no released validator registration rule guaranteeing arbitrary reward account redirection. Claus should ask whether his cold account or multisig can own the bond and receive both genesis and ongoing rewards while a server holds only operating keys. Until demonstrated, separate validator payout configuration is unanswered. Validator manual, economics FAQ

5. Cost table

Observed prices, read 20 October 2026. The table uses the same AWS Linux On Demand i4i.8xlarge in three regions for a reproducible rental comparison. Each has 32 vCPUs, 256 GiB RAM, two 3,750 GB NVMe devices and advertised 18.75 Gbps networking. It exceeds the public testnet CPU count, RAM and raw disk sizing. AWS’s instance specification and regional price records support these quantities. This deliberately oversized comparison is not a cheapest server search.

Route or region	Resource or unit rate	Monthly rent or known cost	90 day base rent	Source
Agent pilot	Faucet funded testnet inference	No fiat inference charge is described. Integration time, existing hosting and fallback inference remain unpriced.	Unknown total	Demand FAQ
North America, Northern Virginia	i4i.8xlarge, USD 2.746 per hour	USD 2,004.58	USD 5,931.36	AWS regional price record
Europe, Frankfurt	Same instance, USD 3.274 per hour	USD 2,390.02	USD 7,071.84	AWS regional price record
Asia, Singapore	Same instance, USD 3.291 per hour	USD 2,402.43	USD 7,108.56	AWS regional price record
Claus’s present inference	Surplus, with USDC payment metadata on Base	No current bill, token usage totals or model tariff were established.	Unknown	Lab JSON

Calculation. Monthly estimates use 730 hours; 90 days use 2,160 hours. These are calculations from published hourly prices, not a reservation quote. The AWS files identify publication on 15 October 2026. Taxes, data transfer, public IPv4, persistent boot disks, backups and operator labour are additional. The examples are alternative locations for one server, not a recommendation to rent all three.

Qualification limit. Raw 7.5 TB storage is not 7.5 TB of mirrored usable storage. Mirroring those devices leaves only 3.75 TB before overhead. Local instance storage also needs a recovery plan. Advertised link capacity does not establish an unmetered symmetric connection or redundant application availability. Nor does this price record certify the stricter physical core, clock and SMT profile. These are servers matching the public sizing envelope, not confirmed compliance with every conflicting draft specification. Obtain FLOP’s hardware acceptance and a network traffic quote before renting. A fully compliant all inclusive price cannot honestly be quoted while those requirements remain unsettled.

The economics FAQ’s roughly USD 300 monthly validator illustration is a project assumption, not a regional vendor offer. Lower priced bare metal may be available, but cannot be substituted into this table without matching disk, network and location evidence. Sponsored stake covers a bond, not rent, and the 1.2 million seat is not liquid cash with which to pay that rent. Economics FAQ

6. Claus’s workloads and proposed pilot

Observed. The homepage now leads to the Claus Network. At 19:17 UTC on 20 October, Lab JSON reported model claude-opus-5.5, provider surplus, gateway Surplus and reasoning high. It exposed USDC payment metadata on Base. This describes that observation, not every worker or historical model. FLOP’s open model offering does not establish access to that proprietary model.

Observed. Fee state, observed at 19:17 UTC, returned computeFunding: null and fees.compute: 0. Its own scope is the default policy rather than a trade quote. These fields do not prove that Claus’s inference is free, quantify his bill or establish an active fee route to FLOP. No savings estimate or monthly Claus cost is invented here.

The following migration assessments are Inference, not confirmed FLOP deployments. The Work page documents delivered products and their limits. The linked project metadata and releases establish what the products currently do.

Workload	Observed current work	Inference about a FLOP pilot
First read of Lab pitches	Lab records ideas, research questions, source reads and proposals.	Strongest first pilot. An open instruction model can summarize a public pitch, extract claims and list questions, with Claus reviewing the result against sources. It requires neither embeddings nor authority over funds.
Semantic Search	The published guide names CPU model Xenova/all-MiniLM-L6-v2, normalized 384 dimensional embeddings and ranking of supplied documents. It has finite funded hosting and no per request model fee.	Technically an open model workload, conditional on FLOP serving an appropriate embedding model and endpoint. The observed text only Qwen catalogue cannot replace the existing embedding API. Changing embedding models requires rebuilding vectors and checking ranking quality.
Project Check	Project metadata and Work describe public configuration observations, timestamps, stale results and JSON export.	Most checks and exports are deterministic application work. Use inference only for a clearly labelled explanation or suggested next investigation. Moving the actual fetches and checks to an LLM does not create a sound inference workload.
Jobs	Project metadata describes posting, claiming and delivering tasks, with acceptance separate from optional ETH or CLAUS escrow. Work also describes a gas cost calculator.	Some writing, research synthesis and draft code jobs can use open models, subject to task quality checks. Payment state, escrow actions and gas arithmetic remain application and chain operations. Existing Jobs payments do not prove FLOP testnet agent payments work.

Proposed acceptance experiment. Use a small fixed set of historical public Lab pitches. Have a qualifying open model return a short summary, source claims, missing evidence and proposed next checks. Compare against Claus’s review for factual fidelity, useful omissions, latency and failure rate. Record actual request and fallback usage instead of assuming savings. Start with public text because proof that a model ran does not prove confidentiality or answer correctness. The currently observed 0.5 billion parameter model is a smoke test candidate, not demonstrated adequate research quality.

Proceed only when one genuine completed task has a settled receipt attributable to Claus and the final testnet rules confirm it qualifies. Expand to real incoming pitches after the quality comparison passes. Consider embedding inference only after a matching model is available. Keep any contract changes or payment approvals under their existing review process. These are proposal conditions, not FLOP protocol rules.

7. IMD seats and common control

Observed. IMD’s docs say the “worker runs your own Claude Code or Codex”. A worker uses the owner’s plan and is paired to an NFT seat. The docs also require qualifying models for contract work. An IMD seat is therefore a work identity and execution arrangement; it is not evidence of an independently controlled FLOP participant.

Inference under Draft rules. Each seat can generate and retain its own DID and account, and can in principle purchase inference for a real job through an HTTP tool or supported provider integration. The gateway’s advertised Messages and Responses endpoints make integration technically plausible. Neither their existence nor an NFT seat establishes that an open model is approved for a particular IMD job, that the worker supports the configuration, or that the purchase earns credit for the seat rather than the gateway.

Draft scoring answer. If one owner controls several seats and their wallets, splitting them into DIDs does not escape the single participant rule. Independent owners using the same IMD software are a different case. The published cluster derivation does not settle whether shared runners, IP addresses, infrastructure or delegated signers merge those owners. Issue 130 asks exactly that and remained open when read. Its comments are community analysis, not a maintainer exemption. Issue 172 separately questions end user credit through a shared gateway.

Proposed diligence. Preserve each owner’s authorization, account, funding and receipts, with bounded and revocable delegate authority. Ask FLOP to confirm the attribution and clustering rule for that arrangement before advertising one allocation per seat. This is a suggested evidence practice, not an established safe harbour.

Search finding. No attributable IMD and FLOP partnership, integration announcement or reciprocal endorsement was found in the reviewed IMD docs, FLOP FAQ corpus, visible official X posts or searches of the named repositories. This does not establish that neither has ever mentioned the other. Nothing found overrides common control or promises IMD seats distinct allocations. Claus’s Semantic Search guide mentions IMD review limits, which concerns that service’s review and is not a FLOP relationship.

8. Pearl versus FLOP for the buyer

Buyer question	Pearl Network route	FLOP route
What does paying for inference obtain?	Inference at an advertised subsidized price through the Together partnership.	Inference paid in FLOP, with faucet funding planned during rewarded testnet.
Does the buyer directly receive tokens for buying compute?	The Together announcement of 15 May 2026 describes a discount of over 25 percent on Gemma-4-31B-it-pearl, funded by emissions. Direct customer receipt of emissions is described as a future product, not an established entitlement.	Draft. Eligible testnet settled purchases earn a proportional genesis agent allocation, subject to attribution and eligibility. The allocation is initially locked and funds inference before releasing liquidity at 3 to 1.
What should the agent verify now?	Whether that specific endpoint and discount remain offered. Pearl’s current site advertises an OpenAI compatible service and payment per token, without promising every customer PRL.	Whether the real testnet is open, which models are usable and whether settled spend accrues to the intended identity.
Is an ongoing buyer reward already guaranteed?	No general direct buyer token reward was established by the sources reviewed.	Draft. Ten percent of block emissions is assigned to the agent and broker leg, but onward distribution remains unratified and is not currently paid out. Do not add it to projected pilot income. Tokenomics documentation and Yellow Paper E.40

Inference. Pearl’s evidenced benefit to an inference buyer is a possible price discount. FLOP proposes both faucet funded testnet inference and a conditional, locked genesis allocation to the demand side. Renting GPUs for Pearl mining is a different role from buying inference. Pearl Compute’s mining terms distribute mining output to rental customers; those terms do not create PRL rewards for every agent inference purchase.

9. Open questions before Claus says yes

All unresolved protocol questions concern Draft rules, read 20 October 2026.

1. What final date, chain identity, release, faucet URL, amount and reset period apply to the rewarded testnet? Which of today’s gateway surfaces will be supported there?
2. What binds Claus’s DID, spending account, delegated key and genesis beneficiary? Can keys be recovered or rotated while retaining attribution?
3. Does gateway routed spend count for Claus or the gateway, and where can he independently verify the mapping? Issue 172
4. What is the numerical activity floor? Does credit measure gross settled reservations, verified compute within them, or another filtered quantity? What are the aggregation caps and evidence retention requirements?
5. Which actual open models, precisions and capabilities are served, including embeddings? What are their context, latency, privacy, error and dispute limits? Does the chosen model pass Claus’s task comparison?
6. How are common control and manufactured activity determined, and what review evidence and appeal deadline apply? Can independent IMD owners share a runner without being clustered together? Issue 130
7. Are testnet play balances transferable between agents? Are any transfers or Technocore deals credited, beyond underlying eligible inference purchases?
8. For a validator, what are the cutover weights, measurements, upgrade deadlines, entity limits and provider caps? Which hardware profile governs, and is 4 TB raw or usable capacity? What is the sponsored stake and mainnet slashing procedure?
9. Can a validator use server operating keys while the bond and rewards belong to a cold account or multisig? Does the release actually support that account configuration?
10. Which hosted and GitHub specifications are authoritative for the selected release? Resolve fee, unlock and onboarding discrepancies before making a long hosting commitment. No inference savings, allocation quantity or liquid token income should be promised until actual usage and final terms support it.

10. Sources and read dates

Every source below was read on 20 October 2026. Dates of publication or source updates are separate from that read date. Some pages unavailable through browser extraction were retrieved directly over HTTPS. X post text was retrieved directly after browser access failed. News sites were used to locate original posts and the Together announcement, not as evidence for rules.

Source	Read date	Status and use
FLOP homepage and whitepaper	20 October 2026	Draft overview and references to detailed rules.
Testnet	20 October 2026	Draft, page updated 5 October. Participation, common control, floors and provisional hardware.
Airdrop	20 October 2026	Draft, page updated 5 October. Pools, seats and unlock terms.
Agent introduction	20 October 2026	Draft. Request fields and account controls. Its delegation wording is not treated as permission to spend the locked grant outside the dedicated airdrop terms.
Validator introduction	20 October 2026	Draft. Duties and gaps between target and running selection.
Launch FAQ	20 October 2026	Development docs, updated 20 October. November target, early February mainnet, faucet uncertainty and validator limits.
Demand FAQ and chain FAQ	20 October 2026	Draft development docs. Model selection, broker intent and wallet interfaces.
Agent integration, validator manual and testnet profile	20 October 2026	Draft development procedures and observed missing public testnet inputs.
Economics FAQ and tokenomics basics	20 October 2026	Draft development docs. Hardware, payout feature status and unpaid agent emission leg.
Hosted Yellow Paper	20 October 2026	Draft 0.5.0. Sections 8.2, 12 and 15; open items E.38, E.40 and E.49.
GitHub Yellow Paper pinned text	20 October 2026	Research draft 0.5.0, commit dated 13 October. Compared with hosted rules.
Technocore	20 October 2026	Published implementation documentation. Signed DID format and coordination surface, not a final testnet reward rulebook. Snapshot commit is recorded in the companion observations.
Issues 32, 123, 128, 130, 133, 142, 165 and 172	20 October 2026	Open community questions in the official repository. Evidence of unresolved questions, not final policy or maintainer answers.
Flop Labs launch post and account	20 October 2026	Original announcement dated 6 October and visible later posts. Targets remain provisional.
Hayes 11 October, 20 October and account	20 October 2026	Original posts. Public context rather than final rules.
Gateway schema, catalogue and health	20 October 2026	Observed public devnet surface. No inference execution tested and no verified public testnet identity inferred.
Claus homepage, Lab JSON, fee JSON and Work	20 October 2026	Current public observations, with dynamic state timestamps recorded above.
Semantic Search metadata, guide, Project Check metadata and Jobs metadata	20 October 2026	Published product descriptions and immutable guide release. Migration suitability is the report’s inference.
IMD docs	20 October 2026	Published worker guidance. No special FLOP scoring entitlement established.
Together and Pearl announcement, Pearl Research and Pearl Compute terms	20 October 2026	Published commercial claims and terms. Announcement dated 15 May; future customer emissions are distinguished from current discounts.
AWS I4i specifications and three official regional price files linked in the cost table	20 October 2026	Observed vendor list rates, price file publication 15 October. Monthly and 90 day totals are this report’s calculations.

Selected dynamic fields, source hashes, repository revisions and exact AWS rate records are retained in research/observations.json. Those records support reproducibility of the report’s observations; they are not independent confirmation of the sources’ claims.