docs: the support README and building-with-interfold describe APIs that no longer exist

[Toby1009]

Updated 2026-08-20. Re-checked against main at 3c7172c8. Everything below still
reproduces. The crates/compute-provider section changed shape rather than being fixed: #1821
deleted the API it described, so the README now documents parameters and signatures that no
longer exist.

Separate from the journal-layout drift, several documents in this area describe interfaces that have
moved. These are copy-and-run examples, so each fails at the point where someone tries to use it.

Two of them share the same two dead field names, threshold and e3ProgramParams, which suggests
both were written against the same earlier request API and neither was revisited.

crates/support/README.md

• The request example does not match E3RequestParams. It passes threshold: [M, N] and
  e3ProgramParams (:150-164). Neither field exists. The struct
  (packages/interfold-contracts/contracts/interfaces/IInterfold.sol:567-577) takes
  committeeSize, which is a CommitteeSize enum rather than an [M, N] pair, and paramSet
  instead of encoded program params. The example also omits paramSet, expectedFeeToken,
  expectedCryptoConfigId, and maxFee.

• The Boundless "defaults shown" are not the defaults, and every value differs:

  Key	README :101-107	Runtime host/src/lib.rs:121-171
  min_price_eth	0.001	0.00005
  max_price_eth	0.03	0.002
  timeout_secs	1200	600
  lock_timeout_secs	600	300
  ramp_up_secs	120	60
  lock_collateral_zkc	5.0	2.0

  The README's own later table (:222-233) matches the runtime values, so the document contradicts
  itself. Someone who copies the first block offers twenty times the intended minimum price.

• The journal description lists seven bound items and omits the parameter hash and the input root
  (:202-212: "the chain, Interfold contract, E3, encryption scheme, committee key, output, and SAFE
  commitment"). The journal has nine (crates/support/types/src/lib.rs:68-104), and both verifier
  layers check the two that are missing from the description.

• :247-257 names .github/workflows/support-docker.yml, which does not exist. The jobs are
  .github/workflows/ci.yml:284 and releases.yml:109. (Also noted in the build issue.)

• The /run_compute example sends the wrong committee-key field. Already fixed in open PR feat: add chunked C2 proving [skip-line-limit] #1772,
  which changes "committee_public_key" to "committee_public_key_hash". Listed only so it is not
  reported twice.

docs/pages/building-with-interfold.mdx

• getE3Quote is shown taking six positional parameters (:107-115). It takes a single
  E3RequestParams struct (contracts/Interfold.sol:1051-1053), as does request (:281-283).
• The JavaScript request object uses threshold: [3, 5] and e3ProgramParams (:271-289), the same
  two removed fields as the support README, and omits the four newer ones.
• The event is shown as E3Requested(uint256 e3Id, E3 e3, IE3Program indexed e3Program) (:239).
  The third parameter is now bytes32 indexed cryptoConfigId
  (contracts/interfaces/IInterfold.sol:137). Code that filters or decodes on the old shape does not
  match.

crates/compute-provider

• The crate README documents an API that fix: crisp security issues [skip-line-limit] #1821 deleted. Readme.md:82-88 lists use_parallel and
  batch_size as ComputeManager::new() parameters. The constructor now takes three arguments
  (src/compute_manager.rs:25-27), and use_parallel, batch_size, and the .chunks() path are
  gone from the crate along with its rayon dependency.
• The two examples call provider.start() with no argument and read one return value
  (Readme.md:40, :72). start now takes an InputPolicy and returns
  Result<(P::Output, Vec<u8>), ComputeError> (src/compute_manager.rs:57).
• The Risc0Provider example implements fn prove(&self, input: &ComputeInput) (Readme.md:65),
  which no longer satisfies the trait: prove gained a second parameter, policy: InputPolicy
  (src/ciphertext_output.rs:21).

Happy to open a PR for any of this if it is useful.

[Toby1009]

The list above turned out to be incomplete. Working through it surfaced roughly three times as many
stale statements in the same three files, so recording them here rather than leaving the issue
describing a smaller problem than the PR fixes.

PR #1813 covers all of it. Two items moved out.

Also found in crates/support/README.md

• e3_id was numeric in five places, not one: both webhook examples, the /run_compute body, and
  both response examples. The field is a string in ComputeRequest, WebhookPayload, and
  ProcessingResponse.
• Step 5 said the server falls back to dev mode when Boundless configuration is absent. The backend
  comes from risc0_dev_mode alone, whose default is 1, and a Boundless request with missing
  credentials fails rather than falling back. The README already documented the field correctly in
  Step 1, so this was another self-contradiction.
• The CRISP request example passed customParams: "". CRISPProgram._initRound decodes seven values
  from it, so the call reverts.
• The request flow said "Fee payment (1 USDC)". request transfers the quoted fee in the configured
  fee token.
• The directory list named guest/. The guest is at methods/guest/.
• fixtures/payload.json, which the Testing section tells you to run through ./curl_test.sh, sends
  a numeric e3_id and omits four required fields, so the request fails to deserialize.
• The six Boundless offer fields are not usable overrides at all. Filed separately as fix(support-scripts): program start forwards Boundless flags that its launcher rejects #1812.

Also found in docs/pages/building-with-interfold.mdx

• The intro and the Core Responsibilities list assigned input submission and input-tree management to
  Interfold. Both belong to each E3 program.
• The E3 struct described seed as a committee-selection seed and requestBlock as a block number.
  requestCommittee ignores the passed seed and the registry derives its own; requestBlock is
  assigned block.timestamp. The struct also omitted ciphertextCommitment.
• maxDuration was described as an input-window limit. It bounds the worst-case
  request-to-decryption duration.
• The KeyPublished row said inputs are accepted at that stage, and the Failed row said refunds are
  initiated on entry. Input acceptance is up to the E3 program, and processE3Failure calculates
  refunds.
• request was shown with positional parameters and a single return value. It takes one
  E3RequestParams and returns (uint256, E3).
• The request-flow narrative said the stored E3 seed is shared with deterministic ticket scoring. The
  registry derives a separate committee seed after the request is final.
• The input-window paragraph said the protocol gates submission. Each E3 program does, and the default
  template checks only the closing bound.
• The Merkle-tree section described a Lean tree owned by Interfold. The default template uses
  InternalLazyIMT and inserts each SAFE ciphertext commitment directly as a leaf.
• CiphertextOutputPublished omitted ciphertextCommitment, and the program's InputPublished was
  shown with an inputHash it does not have.
• The result-publication flow mentioned only the E3 program verifier. The protocol verifier runs
  first, and both must pass.
• The JavaScript request snippet reverted as written: request pulls the fee with transferFrom, and
  it rejects a quote above maxFee.

Also found in crates/compute-provider/Readme.md

• The installation line combined git and path, which cargo rejects as ambiguous.
• The snippets imported voting_core::fhe_processor and methods::VOTING_ELF. Neither exists.
• The dependency URL still pointed at the pre-rename gnosisguild organisation.
• The Usage text described what ComputeManager::start returns incorrectly.

Moved out of this issue

• fix(support-scripts): program start forwards Boundless flags that its launcher rejects #1812 — interfold program start forwards six Boundless auction flags that its own launcher
  rejects, so setting any of those config fields stops the container from starting. That is a code
  defect, not a documentation one.
• fix(compute-provider): ComputeManager accepts batch_size Some(0) and then panics #1815 — ComputeManager::new accepts batch_size: Some(0) and start() then panics. The
  batch_size bullet above conflated the documentation mismatch with the missing validation; fix(docs): correct stale API references in support and building guides #1813
  corrects the text and fix(compute-provider): ComputeManager accepts batch_size Some(0) and then panics #1815 tracks the code.

templates/default/interfold.config.yaml is also in the PR, because its comment described the same
six Boundless fields as usable overrides.