AI Skill Report Card

Packaging Weaving Drafts Into Digital Products

A-83·Oct 29, 2026·Source: Web
13 / 15

Given an input draft, immediately classify it and start the pipeline:

Input type?
├── .wif file → parse directly (threading, tie-up, treadling, warp/weft colors already structured)
└── Scanned/photo image → requires reconstruction pass BEFORE anything else:
      1. Identify draft notation style (American vs. Scandinavian floor loom, point vs. straight threading)
      2. Transcribe threading (by shaft, left-to-right)
      3. Transcribe tie-up (which shafts tied to which treadles)
      4. Transcribe treadling sequence
      5. Flag any ambiguous/damaged cells for human confirmation BEFORE building the drawdown

Never generate a drawdown from a scanned source until transcription has been explicitly confirmed (by the user or by internal cross-check — see Workflow Step 2). A wrong drawdown propagated through the whole packet is the single most expensive failure mode.

Recommendation▾
Add a third example showing a fully ambiguous/low-resolution scan that gets rejected/returned for rescan, to illustrate that failure path concretely
15 / 15
Progress:
- [ ] Step 1: Ingest & classify source (WIF vs. scan)
- [ ] Step 2: Transcribe/parse draft elements + self-check consistency
- [ ] Step 3: Generate drawdown (woven simulation)
- [ ] Step 4: Derive specs (shafts, treadles, epi/ppi, yardage estimate, width/length)
- [ ] Step 5: Write pattern narrative (intro, materials, notes, troubleshooting)
- [ ] Step 6: Lay out packet (cover, diagrams, instructions, licensing/attribution)
- [ ] Step 7: Export print-ready PDF + QA pass

Step 1: Ingest & Classify

  • WIF: parse [WEAVING], [WARP], [WEFT], [THREADING], [TIEUP], [TREADLING], [COLOR TABLE] sections directly. Trust structured data; still sanity-check shaft/treadle counts match across sections.
  • Scanned image: note resolution and legibility. If resolution is too low to distinguish individual cells (common with phone photos of old drafts), say so explicitly and ask for a higher-res rescan or crop before proceeding — don't guess silently.

Step 2: Transcribe & Self-Check

For scans, convert the grid into the same structured form a WIF would have (threading sequence, tie-up matrix, treadling sequence, number of shafts/treadles).

Self-check before moving on:

  • Does every shaft used in threading appear in the tie-up?
  • Does every treadle used in treadling appear in the tie-up?
  • Is the threading repeat consistent with the stated pattern name/type (point twill, rosepath, overshot, etc.) if the historical label is legible?
  • Do selvedge threads look intentional (often a giveaway of transcription error if missing)?

If any check fails, flag it explicitly as a decision point rather than silently "fixing" it — historical drafts often have genuine irregularities (weaver's errors, pattern mends) that are part of the authentic charm and shouldn't be auto-corrected away.

Step 3: Generate Drawdown

Produce the woven simulation (warp/weft interlacement) from threading × tie-up × treadling. Render as a clean black/white or color grid matching standard weaving notation (warp vertical, weft horizontal, shaft numbers labeled, treadle numbers labeled, selvedges marked).

Step 4: Derive Specs

Calculate and state:

  • Number of shafts and treadles required
  • Suggested EPI/PPI based on yarn weight (ask or infer from context; default to stating a range, e.g., "10/2 cotton at 24 EPI" rather than a single unverified number)
  • Warp length and yardage estimate for a stated finished size (give the formula, not just the number, so the buyer can adjust)
  • Loom waste allowance (standard default: 18–24 inches)

Step 5: Write Pattern Narrative

Sections to include, in order:

  1. Pattern name & historical note (if sourced from a named archive/collection, credit it)
  2. Skill level
  3. Materials (warp, weft, reed/heddle count, loom width needed)
  4. Threading draft (diagram + written shaft sequence)
  5. Tie-up (diagram)
  6. Treadling (diagram + written sequence, direct tie-up vs. table loom note if applicable)
  7. Drawdown (full visual)
  8. Weaving notes (selvedge handling, beat, finishing/wet-finishing instructions)
  9. Troubleshooting (common draft-specific issues, e.g., floats catching, tension for long floats in overshot)
  10. Licensing/usage note (personal use, attribution to original historical source, no resale of the draft data itself if it's public domain transcription vs. original adaptation)

Step 6: Layout

  • Cover page with a rendering of the finished cloth (drawdown as fabric-like image) or a photo if available
  • Consistent typography, numbered pages, table of contents for multi-pattern packets
  • Diagrams large enough to read when printed at letter/A4 size — never shrink a tie-up box below readable threshold
  • Include both US and metric measurements

Step 7: Export & QA

  • Export as print-ready PDF (embedded fonts, CMYK-safe colors if color yarn swatches included, 300dpi diagrams minimum)
  • Final QA pass: re-check drawdown matches threading/tie-up/treadling one more time after layout (layout tools can silently corrupt tables/images)
  • Verify page numbers, TOC links, and that no diagram got cropped by the page margin
Recommendation▾
Include a short template/skeleton of the actual PDF table of contents or section headers as a literal reusable artifact
14 / 20

Example 1: Input: Clean .wif file, "Snail's Trail and Cat's Paw," 4 shafts, 4 treadles, cotton warp. Output: Packet with drawdown rendered directly from WIF data (no transcription step needed), materials section defaulting to 10/2 cotton at 24 EPI, full narrative, 8-page PDF with cover showing simulated cloth swatch.

Example 2: Input: Phone photo of an 1880s handwritten draft, faded ink, point-twill threading, treadling partially illegible in one corner. Output: Transcription with the illegible treadling section explicitly marked [UNCERTAIN: repeats pattern established in rows 1–12, assumed symmetric — confirm before weaving], drawdown generated only for the confirmed portion with the uncertain section visually flagged (hatched or grey-shaded) in the diagram, and a note in the packet's historical-note section describing the draft's condition and the transcription assumption made.

Recommendation▾
Show a concrete sample drawdown grid (even as ASCII/text) rather than only describing it, since this is the core deliverable and currently stays abstract
  • Always separate "what the source says" from "what the agent assumes" — use explicit bracketed flags in working notes, resolved (or left visibly flagged) before final export.
  • Default to conservative yarn/sett recommendations with a stated rationale, not a bare number.
  • Preserve historical irregularities rather than "cleaning up" a draft into modern symmetry unless the user asks for a corrected/modernized version — offer both as separate outputs if unsure which is wanted.
  • Keep diagrams vector-based (SVG/PDF-native) rather than rasterized screenshots whenever the pipeline allows it — this keeps printed output crisp at any size.
  • Name files and internal sections consistently (draftname_threading.svg, draftname_drawdown.pdf) to make revision passes traceable.
  • Generating a drawdown before transcription is confirmed — this is the #1 cause of a packet that looks finished but is simply wrong.
  • Silently "fixing" asymmetric or irregular historical drafts, erasing the authentic character that makes historical sourcing valuable.
  • Shrinking tie-up/threading diagrams to fit page layout until they're unreadable when printed.
  • Omitting the yardage/warp-length formula and giving only a single number with no way for the buyer to adjust for their own project width.
  • Forgetting attribution/licensing language when a draft is sourced from a named historical archive or collection — this is both an ethical and legal exposure.
  • Treating a low-resolution scan as "good enough" and guessing at ambiguous cells instead of flagging them or requesting a better source image.
0
Grade A-AI Skill Framework
Scorecard
Criteria Breakdown
Quick Start
13/15
Workflow
15/15
Examples
14/20
Completeness
17/20
Format
14/15
Conciseness
13/15