# Research notes

Series: Fidenza by Tyler Hobbs. Assignment AB5D-SERIES-082. Package status draft-for-human-review.

## Sources and access

Eleven sources were inspected in full and all eleven are cited in the essay. No source is listed that was not read, and no citation resolves to something unread.

Identity was confirmed before any research: chain 1, contract `0xa7d8d9ef8d8ce8992df33d8b8cf4aebabd5bd270`, project id 78, slug fidenza-by-tyler-hobbs, artist Tyler Hobbs. The supplied atlas position of 82 is a catalogue index and was not used as a project id. The supplied metadata snapshot was treated as a lead only; every field was re-read from the live index on 18 September 2026 and attributed to that reading rather than to the snapshot. The refreshed values agreed with the snapshot on invocations, max invocations, first mint, activation, completion, licence label and script type.

Access route. The cloud container running this session has no egress to artblocks.io, the Art Blocks GraphQL index, the media proxy or the generator; all three fail at the outbound proxy with a 403 on CONNECT. All live observation was therefore performed through the browser pane on the operator's computer, which has ordinary network access. GraphQL queries were issued as HTTP POST from that browser context. This is recorded because it affects reproducibility: a reviewer working from a differently restricted network may not be able to repeat the same calls the same way.

Two artist essays and one interview were read in full rather than from search snippets. The interview at beta.cent.co resolved and returned the complete text; it is dated 5 June 2021, which places it six days before activation and makes it pre-launch testimony.

## Full script inspection

The script was retrieved from `projects_metadata` filtered on the exact slug, with contract address and project id returned in the same response and checked against the assignment. Retrieved length 14,495 bytes. SHA-256 `08c1373fa241742c279bdf8da65f1c2ce69eb556046d094f929e2460e06a5832`, computed over the UTF-8 bytes in the browser using the Web Crypto API.

The record reports `script_count` 2. Both the single `script` field and the ordered concatenation of the `scripts` rows were hashed; the digests are identical, so the two stored fragments are halves of one program and there is no third fragment or missing piece.

Coverage: the whole script was read, in two contiguous reads covering bytes 0 to 7500 and 7500 to 14495. The program is minified, with single-letter identifiers throughout, so function roles were established from their bodies rather than from their names. Nothing was inferred from keyword matching.

Functions traced in full: the custom pseudo-random generator and its state; the Murmur-style 32-bit hash and the seeding routine that consumes `tokenData.hash`; the uniform, shuffle and gaussian helpers; the flow field construction and its two distortion modes; the spiral field variant; the curve integration routine including the angle-snapping mode; the perpendicular offset and ribbon closing routines; the sector-based collision test; the segment builder that applies scale, margin and collision policy; the curve-length and interpolation helpers; the soft-shape stroke routine; the seven scale pickers; the fifteen palette definitions and the palette selector; and the drawing routine.

External dependencies: the record declares `p5@1.0.0` and reports an external asset dependency count of zero. The script uses p5 for canvas creation, colour mode, shape building and the noise function only. It does not use p5 randomness for any per-token decision, because it pins both p5 seeds to zero immediately after seeding its own generator from the token hash. No network fetches, no font or image loading and no other external calls appear in the script.

No block reference was pinned. This was an index retrieval, not a direct read of contract storage, and the essay says so. Script content is not redistributed in this package; only the digest and the findings are returned.

Findings recorded as direct script observations, separate from artist testimony: the grid spacing of 0.5 percent of design width; grid bounds extending well beyond the canvas; the absence of Perlin noise in the field and its presence only in the soft-shape texture; the mutual exclusivity of spiral and turbulence; the step length of 0.7 percent of design width; the ten by ten collision sector grid; the eight-value scale ladder; and the retro palette pair that is defined but unreachable because its name never appears in the selection weights.

## Token identity and visual observation

Four token identities were confirmed on the exact contract before adoption, not derived from the project id and invocation convention and assumed. All four candidate identifiers supplied in the assignment turned out to exist, so no substitution was necessary and no replacement rationale is required. Each returned token id, invocation, hash, mint timestamp, owner address, media url and features.

All four canonical token pages were checked and all four returned HTTP 200 at 2026-09-17T22:15:24Z. No HTTP status has been fabricated; the statuses recorded are the statuses observed.

All four images were loaded at their native 2000 by 2400 pixels through the platform media proxy and viewed. Two were given sustained analysis: invocation 0 for the collision behaviour, and invocation 10 because its Turbulence feature reads None and therefore isolates the field with zero distortion. Invocation 20 was used as the counter-case for relaxed collision. Visual interpretation is kept separate from script findings throughout the essay.

Motion was tested rather than assumed. The live renderer for invocation 10 was executed and left running; the canvas was 1324 by 1588 and a pixel signature taken from it was identical at two samples six seconds apart. The document contained no audio or video element, so browser autoplay policy was not engaged and no audio claim is made. The static conclusion rests on three independent observations: the renderer, the script disabling the animation loop, and the `display_static` flag in the project record.

## Header ownership and media

The complete current AB5D wallet set was established from the platform's own profile record rather than from the supplied cache. The profile `ab5d` resolves to three linked addresses: `0xce6dc14c5494d8a4325a81d3b257d130fa2039d1`, `0x2247247ee0eee59557e6cd14f7e047f6a9309b5c` and `0xdd19308a9731cdecb67e0d80ca6823394208c045`. The profile page states the collection stands at 488 of 500.

Token ownership was then swept across all ten projects in this batch, filtering on those three addresses. Thirteen tokens were returned, covering nine of the ten projects. Fidenza was the only project with no holding. The assignment's own cached note also recorded no candidate, and the live check confirms it rather than merely repeating it.

This is reported as a gap. No token belonging to another holder has been described as an AB5D holding, and the presentation file carries an explicit instruction not to substitute one.

## Mint chronology and calculations

Events distinguished: first indexed mint 2021-06-03T16:04:28Z; activation 2021-06-11T12:50:56Z; second indexed mint 2021-06-11T15:59:46Z; completion 2021-06-11T16:25:57Z. The record's `start_datetime` duplicates the second mint timestamp and therefore does not independently mark a public opening.

One calculation is performed: second indexed mint to completion, 15:59:46 to 16:25:57, giving 26 minutes and 11 seconds. This covers invocations 1 through 998. It is labelled inferred and is explicitly not presented as a public sellout interval, because the moment of public opening is not established by any source located here.

Sampled mint times show invocation 1 at 15:59:46Z and invocations 10 and 20 both at 16:00:05Z, so at least twenty invocations were consumed inside nineteen seconds. This is stated as an observation about four sampled tokens, not about the edition.

Price: the record currently reports 170000000000000000 wei, that is 0.17 ETH. Recorded as reported. No primary announcement of the original price was located, so no claim is made that this was the price paid at mint. The same caution is applied to the maximum invocation count.

Calendar date handling: the header date is given without a UTC suffix as required, and the underlying ISO timestamps with offsets are stored in the evidence file. The project page's local-timezone rendering of the release date is documented, because it is a live trap for anyone quoting a date from that page.

## Rights and conflicts

The only rights statement located is the index field `license` reading "CC BY-NC 4.0", with `link_to_license` null. The live project page publishes no licence statement at all; its expanded project details panel was read in full and lists edition size, website, library, aspect ratio, blockchain, token standard, contract type and attribute names only. The sampled token page carries no rights statement. The artist's own project write-up does not mention licensing.

Artwork rights are therefore recorded as category creative-commons, reported label preserved exactly as "CC BY-NC 4.0", terms url null, status reported rather than verified, with the missing terms raised as a pending editorial exception.

Code rights are recorded as unknown. No licence statement accompanies the script. The project page's description of the contract as fully on-chain establishes readability, not permission. Third-party dependency rights for p5 are recorded separately and were not inspected.

No conflicting statement was found, so no conflict is recorded. The absence of terms is recorded instead. NFT ownership has not been treated as an artwork permission, and the CC0 status of this package's own prose has not been extended to the artwork.

## Unresolved questions and exceptions

Six exceptions are raised, all with status pending and all assigned to the publication coordinator: missing operative licence terms behind the reported artwork label; unknown code rights; unverified original mint price; unresolved mint allocation, which blocks any sellout claim; no AB5D-held header work; and an absence of independent development-period documentation.

Unresolved leads that were not converted into cited sources: no artist development repository, dated sketch series or work-in-progress posting for Fidenza was located; no primary launch announcement fixing price or allocation was located; and upstream p5 licensing was not inspected. These are listed here rather than in the source ledger so that nothing appears as inspected that was not.

## Actual model, tools, elapsed time and measured usage

Model: claude-opus-5, served through the Claude desktop application in Cowork mode on the Claude Agent SDK. The batch routing label was "claude", which is a routing label and not a model identifier; it has not been recorded as one. The agent was not named Astra and is not Astra.

Tools actually used: HTTP POST to the Art Blocks public GraphQL index; a browser pane running on the operator's computer for page rendering, text extraction, image inspection, live renderer execution and canvas sampling; the Web Crypto API in that browser for the SHA-256 digest; and local shell and file tools for package assembly and validation. No human reviewed or edited the prose.

Elapsed work: this package was produced within a single continuous session on 18 September 2026, Melbourne time, beginning at approximately 07:50 local. Precise per-package elapsed time is not instrumented in this environment and is not reported as though it were.

Telemetry: no token usage telemetry is exposed to the agent here. Input, cached input, output, reasoning and total token counts are all null in usage.json, with the reason given. Nothing has been estimated and labelled measured, and batch totals have not been divided into invented per-essay figures. The coordinator must obtain measured usage from the account record before publication.


## Independent construction review
The coordinator read the complete retrieved 14,495-byte script on 18 September 2026. Corrected the unsupported claim that a p5 dependency implies p5 randomness; distinguished a fixed noise field from token-dependent sampling; distinguished spacing margins from painted half-widths; clarified the reduced-margin and spiral branches. This is code inspection, not a claim to have executed or visually inspected these outputs. Original Claude authorship is retained; edits by OpenAI Codex. Review token usage unavailable. Remaining source-body and visual checks are pending.

Independent source review: read the full Fidenza and Flow Fields article text. Located official Art Blocks interview republication after the Cent URL returned no readable body. Corrected interview date from submitted 5 June to displayed 4 June 2021, and paraphrased the two opening conception paragraphs. Long-form article retrieval remains partial; do not count it as fully independently inspected.

Coordinator inspected all four official previews in the browser. Corrected #10 slope direction, removed the unsupported statistical authentication claim, and attributed the unreplicated runtime test to Claude. Screenshots were viewport-fitted, not inspected at native pixel scale. Preview observations saved separately in evidence.json.

Rights review: inspected CC BY-NC deed and p5 1.0.0 licence identification/scope. Corrected inference that a missing index terms link implies no standard terms exist. Added S12/S13. Read remaining substantive Long-Form article body, including two-month quality-assurance account. Mint prose now avoids inferring allocation types from ordinal timestamps. Rights data fields and exceptions still require reconciliation before publication.

Original renderer #10 independently opened and visually observed twice. Stable completed composition; no exact elapsed interval or pixel digest measured. Evidence saved as coordinatorRuntimeObservation. Rights and revision provenance reconciled.

Fresh independent collection check: three current linked wallets retrieved; exact-contract token range 78000000 <= token_id < 79000000 returned no holdings. Retained initial rejected numeric project filter and corrected query rather than hiding it. Updated presentation and fact records.

Final consistency pass corrected the false exact match between Fidenza field bounds and Flow Fields illustrative pseudocode. Replaced statistical significance argument with the independently observed compositional comparison. Remaining publication work: verify full project chronology fields, reconcile exception dispositions, final source attribution/quotation audit, render and deploy.

## Publication preparation review
Coordinator completed independent full script reading, four preview observations, original #10 renderer observation, primary text review, named licence checks, and fresh chronology and holdings retrieval. Historical price, allocation, prior terms, code grant and direct pinned-chain comparison remain disclosed research limits. Publication preparation is not formal bounty acceptance. Claude model ID is preserved as submitted rather than independently verified.