Reconcile: $ARGUMENTS

Reconcile: $ARGUMENTS

Cross-reference a paper (or all papers) against the collection, ensuring every citation link is bidirectional and accurate.

This is a notes-layer skill. It updates paper notes and collection cross-references only; it does not initialize or mutate propstore source branches.

Step 0: Determine Mode

Parse $ARGUMENTS:

• If --all: list all paper directories and process each one sequentially (Step 1 onward, looping)
• Otherwise: treat as a single paper directory path

if [[ "$ARGUMENTS" == "--all" ]]; then
  ls -d papers/*/ | grep -v "papers/pngs" | sort
  # Process each directory through Steps 1-5
else
  paper_dir="$ARGUMENTS"
fi

Step 1: Validate Paper Directory

ls "$paper_dir"/notes.md "$paper_dir"/citations.md 2>/dev/null

Required files:

• notes.md — must exist (contains the cross-references section we'll update)
• citations.md — must exist (contains the reference list we cross-reference from)

If either is missing, report and skip this paper:

SKIP: papers/[dirname]/ — missing [notes.md|citations.md], run paper-reader first

Also check: Does notes.md already have a ## Collection Cross-References section?

• If yes: this paper has been reconciled before. Read the existing section to understand current state, then update it.
• If no: this is a first-time reconciliation. Create the section from scratch.

---

Step 2: Forward Cross-Referencing (This Paper Cites → Collection)

Check which papers cited by this one are already in the collection.

2.1: Extract Citation Keys

Read citations.md — focus on both the full Reference List and the Key Citations for Follow-up section. Extract author last names and years for the key citations.

2.2: Search Collection Index

For each key citation, grep papers/index.md for the author's last name:

grep -i "AuthorName" ./papers/index.md

If found, the matching line contains the directory name. For more detail, read papers/<dirname>/description.md.

2.3: Write/Update Forward Cross-References

In notes.md, write or update the ## Collection Cross-References section:

## Collection Cross-References

### Already in Collection
- [[AuthorA_Year_ShortTitle]] — cited for [reason from citations.md context]

### New Leads (Not Yet in Collection)
- AuthorB (Year) — "Paper title" — relevant for [reason]

### Supersedes or Recontextualizes
- [If this paper extends/corrects/supersedes an existing collection paper, note it here]
- [Only genuine relationships — not every citation]

### Conceptual Links (not citation-based)
- [[PaperC_Year_Title]] — [specific topical connection: what claim/finding/method links these papers]

2.4: Backward Annotation (Supersedes Only)

If the "Supersedes or Recontextualizes" section is non-empty, append a see-also note to each affected paper's notes.md:

echo "" >> ./papers/AffectedPaper_Dir/notes.md
echo "---" >> ./papers/AffectedPaper_Dir/notes.md
echo "" >> ./papers/AffectedPaper_Dir/notes.md
echo "**See also:** [[NewPaper_Dir]] - [relationship description]" >> ./papers/AffectedPaper_Dir/notes.md

Check first whether a see-also note already exists (to avoid duplicates):

grep -c "NewPaper_Dir" ./papers/AffectedPaper_Dir/notes.md

---

Step 3: Reverse Citation Search (Collection Cites → This Paper)

Find papers in the collection that reference this paper.

3.1: Extract Search Keys

From the paper's notes.md, extract the first author's last name and year. Build a grep pattern:

# Search all collection markdown files for references to this paper
# Exclude the paper's own directory to avoid self-matches
grep -rl "AuthorLastName.*Year" ./papers/ --include="*.md" | grep -v "papers/AuthorLastName_Year_ShortTitle/"

Also try the directory name pattern:

grep -rl "AuthorLastName_Year" ./papers/ --include="*.md" | grep -v "papers/AuthorLastName_Year_ShortTitle/"

3.2: Add "Cited By" Section

If any collection papers cite this one, add or update a "Cited By" subsection in the Collection Cross-References:

### Cited By (in Collection)
- [[CitingPaper_Year_ShortTitle]] — cites this for [aspect, determined in Step 4]

If no papers cite this one, either omit the section or write:

### Cited By (in Collection)
- (none found)

---

Step 4: Conceptual Links (Topic-Based, Not Citation-Based)

Steps 2–3 find explicit citation links. This step finds conceptual connections: collection papers that address the same problems, whose findings interact with this paper's claims, or whose methods complement or contradict this paper — regardless of whether any citation relationship exists.

This is what makes the collection a knowledge graph, not just a citation graph. This step is not optional. Citation-based cross-referencing alone misses the most valuable connections: papers from different research traditions that converge on the same empirical observation, or later papers that provide mechanisms for earlier observations.

4.1: Identify Key Claims and Topics

Read this paper's notes.md and extract the 3–6 most important claims, methods, or findings. These are the search axes.

What to extract — be specific about the phenomenon, not just the topic:

• A specific empirical observation (e.g., "formant transitions hold at 65ms while steady states stretch 1.5x")
• A model or framework (e.g., "multi-stream parallel representation with synchronized tiers")
• A mechanism or explanation (e.g., "aspiration overlays the CV transition as an independent stream")
• A critique of another approach (e.g., "linear phoneme models conflate phonological and phonetic units")
• An open problem the paper identifies (e.g., "minimum/maximum duration constraints not specified")

4.2: Search Collection for Topic Matches

For each claim/topic, think about what kinds of papers would connect:

Same phenomenon, different framework:

• Does another tradition (articulatory phonology, acoustic phonetics, perceptual studies) observe the same thing this paper describes? Search for the phenomenon's acoustic/articulatory/perceptual terms.
• Example: "stable transition duration" → search for "stiffness", "gesture duration", "transition.*ms", "formant transition"

Mechanism for observation (or observation for mechanism):

• Does this paper observe something that another paper explains, or vice versa?
• Example: paper observes "only steady states lengthen" → search for "boundary lengthening", "phrase-final", "π-gesture", "prosodic.*slowing"

Data that grounds or challenges claims:

• Does another paper provide the empirical measurements this paper's model requires?
• Does another paper's data contradict this paper's predictions?

Cross-level connections:

• Articulatory ↔ Acoustic ↔ Perceptual papers often address the same phenomenon at different levels of description

Use targeted grep searches across papers/*/notes.md:

# Search for papers discussing the same phenomenon
grep -rl "relevant_term" ./papers/ --include="notes.md" | grep -v "papers/ThisPaper/"

Read matching sections (not full files) to assess connection strength.

4.3: Classify Connection Strength

For each match, classify as:

• Strong — directly addresses the same problem, provides data this paper uses, contradicts/confirms a key finding, or provides a mechanism for this paper's observations (or vice versa). Different formalisms converging on the same empirical fact is always Strong.
• Moderate — related methodology or overlapping problem space, but not directly interacting
• Weak — tangential overlap; omit from cross-references

Only surface Strong and Moderate connections. Weak connections create noise.

4.4: Write Conceptual Links Section

Add a ### Conceptual Links (not citation-based) subsection to the Collection Cross-References:

### Conceptual Links (not citation-based)
- [[PaperA_Year_Title]] — [specific connection: what claim/finding/method connects these papers and how they relate — convergence, tension, mechanism↔observation, etc.]
- [[PaperB_Year_Title]] — [specific connection]

Each entry must state the specific relationship, not just "related to duration modeling." Good: "Hertz's 'stable transition phenomenon' (CV transitions hold at ~65ms while steady states stretch) is exactly what AP predicts for a high-stiffness gesture — different formalisms, same empirical convergence." Bad: "Also about formant transitions."

Group entries by theme when there are 3+ connections (use bold subheadings like **Duration modeling:**).

4.5: Bidirectional Annotation

For Strong connections, check if the connected paper's notes.md already mentions this paper:

• If not, add a reciprocal entry in that paper's ### Conceptual Links (not citation-based) section (create the subsection if needed)
• Check for duplicates before writing

For Moderate connections, only annotate this paper (not the connected paper) — the connected paper's own reconciliation pass will pick it up if the connection is genuinely bidirectional.

---

Step 5: Reconcile Citing Papers

For each paper found in Step 3, read its notes.md (specifically the Collection Cross-References, Related Work, and Open Questions sections). Check for and fix:

5.1: Leads Listing This Paper

Leads can appear in two places depending on whether the citing paper has been reconciled before:

1. ### New Leads (Not Yet in Collection) — inside a ## Collection Cross-References section (reconciled papers)
2. ## Related Work Worth Reading — a flat list at the end of notes.md (unreconciled papers, the paper-reader default)

Search both sections for an entry matching this paper (by author name and year).

If found in ### New Leads (Not Yet in Collection):

• Move the entry out of "New Leads"
• Add it to ### Now in Collection (previously listed as leads) (create subsection if needed)

If found in ## Related Work Worth Reading:

• Do NOT delete the entry (that section is the paper-reader's historical output)
• Annotate inline by appending → NOW IN COLLECTION: [[Author_Year_ShortTitle]] to the entry
• Add the entry to ### Now in Collection (previously listed as leads) in the ## Collection Cross-References section (create section and subsection if needed)

In both cases, write the "Now in Collection" entry with:

• Correct description of what this paper actually contributes
• Key finding summary
• Any tensions or confirmations between the two papers' findings

Example:

### Now in Collection (previously listed as leads)
- [[Groth_2010_AnatomyNanopublication]] — Defines nanopublication model (concept→triple→statement→annotation→nanopublication) with RDF Named Graph serialization. Structurally analogous to the micropublication model but focused on Semantic Web interoperability rather than argumentation structure.

5.2: Inaccurate Descriptions

If the citing paper describes this paper inaccurately (wrong method, wrong finding, wrong scope):

• Edit the description inline to be correct
• Common errors: confusing which variables were manipulated, attributing findings from a different paper by the same author group

5.3: Open Questions Answered

If the citing paper has open questions (## Open Questions) that this paper addresses:

• Annotate the question: append [Addressed by Author_Year_ShortTitle — finding summary]
• Do NOT check the box — that's for the user to decide

5.4: Interesting Tensions

If the new paper's findings conflict with or nuance the citing paper's conclusions:

• Document in the citing paper's notes (in the cross-references section or as an inline note)
• Document in this paper's notes (in the cross-references section)
• Be specific: what differs, why (different methodology? different controls? different population?)

---

Step 6: Report

Output a summary:

Reconciled: papers/[dirname]/
  Forward: N already in collection, M new leads, K supersedes
  Reverse: J collection papers cite this one
  Conceptual: S topic-based connections surfaced (T strong, U moderate)
  Updated:
    - papers/CitingPaper1/ — moved lead to "Now in Collection", corrected description
    - papers/CitingPaper2/ — added to "Already in Collection"
    - papers/AffectedPaper/ — added see-also backward annotation
    - papers/ConnectedPaper/ — added conceptual link (bidirectional)
  Tensions found:
    - [brief description of any finding conflicts, or "none"]

For --all mode, output a final summary after all papers are processed:

Reconciliation complete: X papers processed
  - Y papers had citing papers in collection
  - Z leads marked as fulfilled
  - S conceptual links surfaced
  - W tensions documented
  - V papers skipped (missing notes.md or citations.md)

---

The Reconciliation Principle

After reconciliation, every cross-reference in the collection should be bidirectional and accurate:

1. If Paper A cites Paper B, and both are in the collection:

   • A's notes mention B in "Already in Collection" (with correct description)
   • B's notes mention A in "Cited By (in Collection)"

2. If Paper A was listed as a "New Lead" in Paper B, and A is now in the collection:

   • The lead entry is moved to "Now in Collection" with accurate summary
   • Any inaccurate description from when the lead was first noted is corrected

3. If Paper A supersedes/extends Paper B:

   • A's notes say so in "Supersedes or Recontextualizes"
   • B's notes have a "See also" annotation pointing to A

4. If Papers A and B have conflicting findings:

   • Both papers' notes document the tension with specifics

5. If Papers A and B address the same problem or their findings substantively interact (even without citation):

   • At least one paper's notes mention the other in "Conceptual Links (not citation-based)"
   • Strong connections are annotated bidirectionally; moderate connections at minimum on the paper being reconciled

---

Running on All Papers

When invoked with --all, process papers in alphabetical order. For each paper:

1. Run Steps 1-6
2. Do not re-read papers that were already updated as citing papers — their own turn will come in the alphabetical sweep
3. Be idempotent: running --all twice should produce the same result (no duplicate annotations)

Idempotency Checks

Before every write operation, check if the content already exists:

• Before adding "Cited By" entry: grep -c "PaperDirName" notes.md
• Before adding "See also": grep -c "PaperDirName" affected_notes.md
• Before moving a lead: check if "Now in Collection" subsection already lists it

---

Do NOT:

• Create or modify papers/index.md entries (that's paper-reader's job)
• Delete or overwrite existing notes content (only append/update cross-reference sections)
• Modify the paper's core notes sections (Summary, Parameters, Equations, etc.)
• Output full notes content to conversation (just the reconciliation summary)