← Project Log
Decision

DEC-0017: Interpretive Drift Tracking — Concept Evolution as Content

Date: 2026-03-30 Status: Accepted Scope: KB metadata, project log, site (potential), content pipeline Origin: Observation from Claude Opus (cloud session) after KB deepening batch

The Observation

When new source clusters enter the KB, the interpretive frame of concept entries shifts — not just in volume (more text) but in kind (different understanding). This evolution is currently invisible.

Concrete example: CON-0004 (Participation)

Before 2026-03-29: Participation was Owen Barfield's concept. Four tradition sections (Archaic/Indigenous, Ancient Greek, Hermetic/Neoplatonic, Romantic/Idealist). The traditions field listed ["Anthropological", "Romantic-Idealist", "Neoplatonic", "Indigenous/Primal"]. Source references pointed to Barfield and related Western thinkers.

After 2026-03-29 deepening batch: Three new tradition-specific subsections were added — Daoist (Tao Te Ching), Hindu (Bhagavad Gita), and Homeric (Iliad). The concept is no longer "a Western Romantic-philosophical idea with some cross-cultural analogues." It is now "a cross-cultural pattern with independent evidence in at least six traditions." The traditions field was not updated to reflect this — the metadata lags behind the content.

This is not a data quality bug. It is a category shift: a concept that was defined by one tradition's vocabulary now has independent attestation in traditions that never read Barfield. The meaning of the concept in the project's argument changed. Nothing in the system recorded that shift.

Other examples from the same batch

  • CON-0002 (Katabasis): Accumulated 5 source-specific treatments. The concept moved from "the descent phase of initiation" to "a structural pattern that different traditions stress-test differently" — Euripides shows katabasis when the container fails, Aristophanes shows katabasis as comedy, Virgil shows teleological katabasis, Homer shows katabasis before the Mysteries formalized it.
  • CON-0020 (Metanoia): Received material from 4 sources (Barfield, Plato, Eliade, Upanishads). The concept shifted from a primarily Christian/Greek framing to one with Vedantic and phenomenological attestation.
  • CON-0043 (Catharsis): Gained an Euripidean subsection that reframes catharsis as potentially pathological — "catharsis without resolution" — a reading that contradicts the Aristotelian standard.

What Exists Now

The KB tracks operational metadata about changes:

Field What it records What it misses
last_updated / last_updated_by When and who Why, and what shifted
agent_notes What an agent did Not used consistently; no structured schema for interpretive claims
traditions Which traditions are represented Not updated when new traditions are added by deepening
essay_appearances Which essays reference this concept Not whether the essay changed how the concept is understood
Git history Full diff of every change Requires manual reading to extract interpretive shifts
Prompt relay results What was done in each PR Operational summary, not interpretive analysis

None of these answer: "When did Participation become a cross-cultural concept rather than a Western one?" or "When did the project's understanding of katabasis gain its comedic dimension?"

What's Worth Tracking

Not every edit is an interpretive shift. The system should distinguish:

Change type Example Worth tracking?
Volume A concept gets a longer body No
Coverage A new tradition-specific subsection is added Maybe — depends on whether it changes the concept's meaning
Reframing A concept's role in the project argument shifts Yes
Tension New material contradicts or complicates existing material Yes
Convergence Independent traditions arrive at the same structure Yes — this is the project's core method
Promotion A claim moves up the epistemic spectrum (speculative → defensible) Yes
Demotion A claim moves down (consensus → disputed) Yes

The last two are particularly interesting because the project's epistemic spectrum (documented fact → scholarly consensus → defensible interpretation → philosophical commitment → speculative extension → open question) already provides the vocabulary. A concept entry could track which level its central claims occupy, and shifts between levels could be logged.

Proposed System: Drift Annotations

1. drift_log field on concept entries

A structured YAML array in concept frontmatter recording interpretive shifts:

drift_log:
  - date: "2026-03-29"
    agent: "claude-code"
    type: "convergence"
    description: >-
      Participation gained independent Daoist (wu wei) and Hindu (nishkama
      karma) attestation. No longer a Western-only concept.
    sources_added: ["LIB-0249", "LIB-0288"]
    epistemic_shift: null
  - date: "2026-03-29"
    agent: "claude-code"
    type: "coverage"
    description: >-
      Homeric material (Iliad) added as primary literary evidence for
      original participation in the Western tradition.
    sources_added: ["LIB-0182"]
    epistemic_shift: null

Fields:

  • date: when the shift occurred
  • agent: who made the change
  • type: one of convergence, reframing, tension, coverage, promotion, demotion
  • description: 1–2 sentences naming what shifted (written by the agent that performed the deepening, or backfilled by Cursor from git history)
  • sources_added: which LIB entries drove the shift
  • epistemic_shift: if a central claim moved on the epistemic spectrum, record from and to (e.g., {from: "speculative extension", to: "defensible interpretation"})

2. make drift-report script

A script that:

  • Reads all drift_log entries across concept files
  • Groups by date, type, and tradition cluster
  • Outputs a narrative summary of how the KB's interpretive frame has evolved over time
  • Flags concepts whose traditions field is out of sync with their actual tradition-specific subsections

3. Drift annotations in the deepening workflow

When Claude Code executes a KB deepening prompt, the acceptance criteria would include: "If the new material changes the concept's interpretive frame — adds a new tradition, introduces a tension, or shifts a claim's epistemic level — add a drift_log entry."

This is a small addition to the existing deepening prompt template.

Where This Could Surface

Option A: Project log only (internal)

Drift reports live in project-log/chronicle/ as periodic summaries. No public-facing component. The system serves as an internal record of how the project's understanding evolved, useful for the human's editorial decisions and for future agents reading context.

Option B: Site project log (public, curated)

A public-facing page (e.g., /project-log or /changelog) that surfaces curated drift summaries as editorial meta-commentary. Not raw metadata — written prose that says "This week, the concept of Participation gained independent Daoist and Hindu attestation, which means..."

This is the channel DEC-0016 proposed as "curated editorial meta-commentary" (the counter-proposal to raw agent logs). It would be Claude Code's work to write these, reviewed by the human before publishing.

Option C: Integrated into concept pages (public, ambient)

Each concept page on the site could show a "History" or "Evolution" section derived from its drift_log. Visitors would see not just what the project currently says about Participation, but how that understanding developed — which sources arrived when, what they changed.

Option D: Its own series (content)

As Opus suggested: the meta-evolution of the KB's understanding becomes a content stream. Episodes or essays that narrate how the project's interpretive frame shifted as new material arrived. "The week we realized Participation wasn't Western" as content, not metadata.

These options are not mutually exclusive. A likely progression: A (instrument it) → B (surface it publicly) → D (make it content) with C as an ambient layer throughout.

What This Is Not

  • Not a version control viewer. Git already tracks every diff. The drift log tracks meaning, not text.
  • Not an automated system that detects shifts from diffs. Interpretive shifts require judgment — an agent or human must name what changed and why it matters. The system provides the schema; the agent provides the interpretation.
  • Not a replacement for the existing metadata. last_updated, agent_notes, and prompt relay results continue to serve their operational purpose. The drift log adds an interpretive layer on top.

Risks

  • Overhead: Adding drift annotations to every deepening prompt increases Claude Code's per-entry work. Mitigated by making it optional (only when something actually shifted) and brief (1–2 sentences).
  • Subjectivity: Different agents may disagree on whether a change constitutes a "reframing" or just "coverage." Mitigated by the type taxonomy and by human review of drift reports.
  • Retroactive backfill: The KB has already undergone significant interpretive shifts (Waves 1–3, today's deepening batch) with no drift log. A backfill pass would require reading git history and reconstructing what shifted. This is feasible but time-consuming.

Decision Needed

  1. Accept the schema — add drift_log to concept entry frontmatter and update the deepening prompt template
  2. Build the report — make drift-report script
  3. Choose a surface — which of Options A–D to pursue first
  4. Backfill scope — whether to retroactively annotate the shifts that already happened (today's batch at minimum)
  5. Defer — carry the idea without implementing it yet

References

  • CON-0004 (Participation): concrete example of interpretive drift
  • CON-0002 (Katabasis): 5-source accumulation example
  • DEC-0016 §Documentarian Feed: curated editorial meta-commentary proposal
  • docs/editorial-guidance.md §II: epistemic spectrum vocabulary
  • PR-0074 through PR-0118: the deepening batch that triggered this observation
0:00
0:00