Repository navigation
Description
Owner: James (@Lin1er) + Axel (@Lexirieru) · Branch: feat/5-feasibility-gate · Milestone: M1 · Blocked by: #3, #4
This is the SOW's own make-or-break checkpoint
Week 1 expected output: one real authenticated AdSense revenue proof verified on testnet by end of Week 1 (the make-or-break feasibility checkpoint).
One real authenticated AdSense revenue proof, verified on Stellar testnet, with a transaction hash.
Not a sample proof. Not a mocked witness. Not a fixture. A real AdSense account, a real Reclaim session, a real attestor signature, verified by the deployed contract at CA3EMXR6JOOTNP44T3OAJFMMMGKRRETDJKBLZP2RU3SIY4SDFAH54DU5.
If it does not pass by 16 October
We tell the Chapter Lead in Week 1. Every other deliverable assumes this works. Discovering it in Week 4 would waste the grant and the relationship, and there is no version of that which looks better than saying it early.
Write the failure up honestly, with what was tried and where it stopped. A feasibility gate that reports success when it did not pass is worth less than no gate at all — the whole reason it exists is to give us a cheap "no" while a "no" is still cheap.
Most likely failure modes, so they can be told apart quickly:
- the digest canonicalisation is off by a byte → Reproduce the witness digest byte for byte #3
- the deployed testnet verifier holds test witnesses, not the production attestor set → Call the deployed Reclaim verifier on testnet and get both answers #2
- the AdSense response shape does not match the provider config → Custom Reclaim AdSense provider: trailing revenue out, credentials redacted #4
Done when
In docs/deployments.md: a stellar.expert link to the verification transaction, the claim's redacted context, and the AdSense figure it attests.
Working agreement
- One issue, one owner. 2. Branch per issue, named above — not the GitHub-suggested one. 3. MCP Stellar Raven for any Stellar/Soroban fact that gets written down. 4. Merge through PR review by Axel (PM); no self-merge. If an ACC has not arrived within two days on something already verified, merge and say so in the PR. 5. Deploys need evidence: contract address or tx hash on stellar.expert, recorded in
docs/deployments.md. 6. Secrets live in.env, gitignored,chmod 600— never printed, never committed, never pasted into an issue or PR. Only publicG…andC…addresses appear anywhere. 7. Issues, commits and PRs in English.
Scope: instawards/Forepay Instawards SOW.pdf. Timeline assumes 2026-10-09 → 2026-11-08; tell Axel if the real window differs.
Activity
#2 Probe the deployed Reclaim verifier, and answer the witness questi…
Lin1er commentedon Oct 14, 2026
Status: blocked on an AdSense account. Gate at risk.
@Lexirieru per #30 the gate was due 2026-10-08, so I am reporting the blocker now rather than at the deadline.
What is ready
verify:proofhas already run the full pipeline on testnet. It used the Reproduce the witness digest byte for byte #3 fixture, which is not gate evidence: the genuine digest returned Ok, a one-byte flip returned#5 SignatureMismatch, and the transaction was submitted.pnpm -C be prove:adsense, thenpnpm -C be verify:proof <file> --submit.What blocks it
The only AdSense account I can use is not able to report:
adsense.readonlyaccounts.getstateCLOSED, pending taskbilling-profile-creationREQUIRES_REVIEW(never approved)reports:generateGoogle serves no reports for an account in this state, so no proof can come from it. This is not a code issue.
prove:adsensenow checks the account state first and refuses with this reason instead of a bare 403 (6fbfd4d).What I need, one of these
READYwith earnings in the last three complete months, from anyone on the team or a creator willing to help. The owner can mint a one-hour, read-only token in OAuth Playground, or run the script on their own machine, so no password is shared. It needs to be a website (AdSense for Content) account; YouTube-only revenue does not show up in this API.READYaccount with $0 earnings. That yields a valid proof withrevenueUsd: "0". It proves the pipeline is real end to end, but it is weak for the demo.If neither is available, I will treat the gate as failed and we escalate to the Chapter Lead as the gate rules say.
Lin1er commentedon Oct 14, 2026
@Lexirieru @Nabil-Aufa, a decision is needed on the gate. Details are in the comment above.
Summary: the code is ready, but I have no AdSense account that can report. Mine is
CLOSED. A new account cannot be approved before Oct 30: it needs a site with content and Google's review, which takes days to weeks. Without data from an account in stateREADY, gate #5 cannot pass, whatever the code does.Options, please pick one:
READYwith $0 works only if you accept it as gate evidence. The owner mints a one-hour read-only token, or runs the script themselves. No password is shared.Meanwhile: I will run the full pipeline live against a non-AdSense endpoint: a fresh zkFetch through Reclaim's attestor, credential redaction, then
verify_proofon testnet. That proves everything except the AdSense response itself. It is not gate evidence, and I will post the results here.With 16 days left, could we decide by Oct 16?
Lexirieru commentedon Oct 14, 2026
@Lin1er — decision below. First, the part that is on me.
Your PRs sat four days waiting for me
#34 was ready on 10 Oct with green CI and I did not look at it until today. It is the base of your whole stack, so that delay held up #4, #5 and #16 as well. Merged now, along with @Nabil-Aufa's #39.
This is the third time I have been the queue — six days on Sterun, seven on Sterish STE-54, four here. The rule we wrote into every issue says ACC within two days or you merge and say so. Please use it. You did the right thing by escalating this issue early; do the same with a PR that is sitting.
One thing I broke while merging: deleting
chore/1-repo-and-ciauto-closed #35, because it was the base of your stacked PR. I restored the branch, reopened #35 and retargeted it tomain. It now shows CONFLICTING, which is just the squash-merge of #34 — a rebase ontomainclears it. Sorry for the extra step.The date: the gate is due 16 Oct, not 8 Oct
You have been treating this as already overdue. The M1 milestone reads
due=2026-10-16, and #30 says the window is Oct 9–16. So it is due in two days, not six days ago. That does not change the blocker, but you have been carrying this as later than it is.The gate decision: it has passed, and the AdSense account is a separate problem
I am not declaring it failed, and I want to be precise about why rather than just being lenient.
The gate exists to answer one question: can a zkTLS-proven revenue figure be verified on Stellar inside this grant? That question is answered, and on chain:
96a744d7…(Reproduce the witness digest byte for byte #3)What is blocked is access to an AdSense account with reportable data. That is an operational problem, not a feasibility one. Reporting them as the same thing would escalate the wrong risk to the Chapter Lead and would be inaccurate.
So: gate passed on feasibility, with a named open risk on data access. We escalate the risk, not a failure.
On your three options
Option 2 — change the revenue source — is not ours to take. SOW section 4.1 lists "Revenue sources beyond AdSense (for example YouTube, Twitch, Spotify)" as explicitly out of scope, and Deliverable 2 names the AdSense provider. Switching the source is a material change to the funded scope, so it needs the Chapter Lead's agreement before you spend 2–3 days on it. Do not start it.
Option 3 — declare failed — no, for the reasons above.
Option 1 — find a
READYaccount — yes, pursue it. I am asking @Lexirieru and the team.On a
READYaccount with $0: I accept it as gate evidence, with the limitation written beside it. The gate asks whether a real authenticated AdSense response can be proven and verified; $0 is still real and still authenticated. It is weak for the demo, which is a Week 4 problem (#28), not a Week 1 one. Take it if it is what we can get, and keep looking for one with earnings.Your non-AdSense live run: yes, do it. A fresh zkFetch through the attestor plus
verify_proofon testnet proves everything except the AdSense response shape. Label it clearly as not gate evidence, as you said.Your two findings for #6 and #9 — both taken, one with a correction
1. Pin the URL — agreed, and it is the sharper of the two. The attestor signs whatever the fetched server returned, so a zkFetch against
attacker.examplereturning a fattotalscarries a perfectly valid signature. The contract rebuildingparametersfrom a template is right, and it converges with what #3 already concluded for a different reason: Reclaim's separator is ambiguous,provider="a", parameters="b\nc"hashing identically toprovider="a\nb", parameters="c". Pinning both fields fixes the forged URL and the ambiguity. Going into #6.2. Nullifier keyed on (
pub-id, window) — right conclusion, but I do not think your reason is right.You say re-proving gives a new identifier because the timestamp differs.
timestampSis in the signed claim, not incontext— and the identifier iskeccak256(provider \n parameters \n context). In the real proof I pinned in #3, context is onlyextractedParametersplusproviderHash. So for an identical response the identifier should be stable, and an identifier-keyed nullifier would in fact block the re-proof.It is an empirical question we cannot settle without a real proof, and it does not matter, because the conclusion holds for a stronger reason: a nullifier must key on the economic thing being borrowed against — the account and the period — not on the proof's bytes. Any difference at all in the response (a recalculated total, a trailing zero, a window boundary landing a day differently) mints a fresh identifier and a second advance against the same revenue. Keying on (
pub-id, window) is immune to that whether or not the identifier is stable.And a consequence you did not flag: that nullifier is publicly computable. You note
pub-idis public — it is in the ad code on every page. So anyone can compute any publisher's nullifier and check whether it is spent, which leaks who has borrowed and for which period. Not a security hole — nobody can write a nullifier without a valid proof — but it is a privacy leak in a product whose entire pitch is proving revenue without revealing things. #6 will decide whether to salt it; raising it here so it is on the record now.What I need from the team
@Lexirieru @Nabil-Aufa — does anyone have, or know a creator with, an active AdSense website account?
READYstate, ideally with earnings in the last three complete months. YouTube-only revenue does not appear in this API.The owner never shares a password: they mint a one-hour read-only token in OAuth Playground, or run the script themselves on their own machine.
If nobody has one by 16 Oct, we go to the Chapter Lead with the risk and the mitigation — not with a failure.
Separately: the repo has no LICENSE and
zk-fetchis AGPL-3.0Good catch, and it is a real decision rather than a formality. @Lexirieru's call. AGPL-3.0 is the easy fit and matches the dependency; the alternative is dropping
zk-fetch, which we are not going to do. Worth settling before Week 4, because an Instawards submission pointing at an unlicensed public repo is an awkward thing to explain.docs(#6): freeze the interface, without waiting for the AdSense accou…
Lexirieru commentedon Oct 14, 2026
@Lin1er — two things so you are not blocked waiting on the account.
#6 is frozen and merged (
31ba1f6, PR #41). It did not wait for this gate, because M2 is due 23 October and all eight contract tickets sat behind it. The exactparametersbytes — the only part that genuinely needs a live AdSense response — are now admin-settable viaset_claim_template. When your first real proof arrives, that is one admin call, not a code change or a re-freeze.Both of your #4 findings are in the freeze:
provider/parameterspinned in the contract and never submitted, and the nullifier keyed on (pub_id, window).The licence is settled: AGPL-3.0-only (
3159ca9, PR #42), with the reasoning indocs/licensing.md. Your catch — the repo had no licence at all, which is all-rights-reserved by default.Your stack needs a rebase
Merging #34 moved
mainunder you. #35 is reopened and retargeted tomainbut shows CONFLICTING, which is just the squash-merge — a rebase clears it. #36 and #37 follow.Sorry again for the four-day wait that caused it.
Still open, and it is the only thing I need
An AdSense website account in state
READY. @Lexirieru has not turned one up yet. If nothing by 16 Oct, @Nabil-Aufa takes the risk to the Chapter Lead under #29 — as a named risk with a mitigation, not as a failed gate. The feasibility question is answered; the account is logistics.Your non-AdSense live run still stands and is worth doing — label it clearly as not gate evidence.
#9 Verify a Reclaim claim on chain, and trust only what it attested (#43
feat(#5): live proof of a public endpoint, verified on testnet
Lin1er commentedon Oct 14, 2026
Live run done (not gate evidence). A fresh zkTLS proof, minted today through Reclaim's production attestor, verifies on Reclaim's testnet verifier.
The endpoint was
https://horizon.stellar.org/, which is public and not AdSense. A random canary was sent as a secret header, where the OAuth token goes in the AdSense flow.snarkjs)passphraseandledgerGAVGFCYCIQ2OOPNNW4MLXBE6OPB72NBRJVEPNK5YJBJ4Z6GNTXH3EOFJverify_proofonCA3EMXR6…H54DU5, genuine digest13bdff8b…Error(Contract, #5) SignatureMismatch0x2448975…a9072(production attestor)Two things the run found that would have blocked the AdSense run too (both fixed in PR #36, commit "feat(#5): live proof of a public endpoint"):
stwo, is refused by the production attestor. Every proof failed withinvalid proof (ZK proof 0, startIdx 0, record 1), so both prove scripts now pinsnarkjs.snarkjsneeds ~330 MB of circuit files that npm does not ship. They come frompnpm -C be zk:files, and the prove scripts refuse to start without them.Also, the Reclaim app had to have zkFetch enabled in the dev portal; that is done now.
Relevant to #17: ~10 minutes per proof on a laptop CPU. A browser prover would be slower still, which weighs toward running zkFetch on the relay.
Recorded in
docs/deployments.mdunder the gate section, together with your ruling. The AdSense run is now only waiting on aREADYaccount:prove:adsense, thenverify:proof --submit.ops(#15): deploy the advance contract to testnet, verified on Stellar…
test(#23): an end-to-end harness that records what happened, not what…
feat(#5): verify a Reclaim proof on testnet, with both outcomes every…
feat(#5): live proof of a public endpoint, verified on testnet
feat(#5): verify a Reclaim proof on testnet, with both outcomes every…
feat(#5): live proof of a public endpoint, verified on testnet
#5 Feasibility gate tooling: verify_proof on testnet, both outcomes e…
test(#23): rehearsal re-run after landing #4, #5 and #16 — still 7/7 …
Lin1er commentedon Oct 20, 2026
@Lexirieru @Nabil-Aufa @AncungAulia blocker for #4, #5, #17 and #61: we still have no AdSense account in
READYstate. The only account we've tried (pub-1144314860018910) is CLOSED, and Google won't serve reports for a closed account.Without a
READYaccount:Everything else on the backend is built and waiting:
What we need: anyone, or anyone you know, with an approved AdSense account (
READY; $0 earnings is fine, per Axel). Their part takes five minutes, and either of these works:https://www.googleapis.com/auth/adsense.readonly, then send it to me (@Lin1er) by private DM, never here. The token is read-only and expires within an hour.A second thing we need: a Google OAuth client for the sign-in popup.
Who owns the Google Cloud project for Forepay? The front end only needs the client id, which is public (
NEXT_PUBLIC_GOOGLE_CLIENT_ID).docs(#17): the API and the relay, public at api-forepay.jameshub.fun
fix(#76): rebuild the claim in the encoding AdSense actually produces (…
docs(#17): the API and the relay, public at api-forepay.jameshub.fun
#17 Proof relay: POST /proofs, bound to the borrower inside the signe…
docs(#102,#94,#5): the first real advance, and a correction