Skip to content

docs(research): publish verbatim-vs-derivative axis essay (#47) - #139

Closed
jphein wants to merge 2 commits into
mainfrom
docs/verbatim-derivative-essay-47
Closed

docs(research): publish verbatim-vs-derivative axis essay (#47)#139
jphein wants to merge 2 commits into
mainfrom
docs/verbatim-derivative-essay-47

Conversation

@jphein

@jphein jphein commented May 23, 2026

Copy link
Copy Markdown
Collaborator

Summary

Publishes docs/research/verbatim-vs-derivative-axis.md — a standalone essay developing the foundational architectural call behind MemPalace's storage layer: store verbatim, derive lazily, treat every derivative as replaceable. Distinct surface from the README's compressed thesis section; the README's "What's next" already promised this writeup.

Seven sections:

  1. The architectural choice — verbatim canonical, derivative-replaceable; three claims taken as one.
  2. Why derive-on-write accumulates failure modes — lost nuance, classifier ceiling, irreversibility, locked-in derivation assumptions. Compares Mem0 / Letta / Cognee / Hindsight head-on.
  3. Empirical signal — the 2026-04-25 → 2026-05-05 recovery-collection episode. 210× token-budget gap (3 → 1,267 tokens/Q on kind=content) closed by a single structural change at the write layer.
  4. Vendor-API validation — Anthropic's Managed Agents Dreams API (beta header dreaming-2026-04-21) as independent ratification of the axis. Input read-only, output a separate store, idempotent on inputs.
  5. Operational implications — backup collapses to one concern; audit + re-derivation become routine; model-upgrade resilience; privacy and shelving are architectural.
  6. Limits — five named exceptions where the axis does NOT apply (explicit shelving, per-domain redaction, regulated-data classes, adversarial inputs, high-cardinality machine output).
  7. Conclusion — three-piece case (structural, empirical, convergence).

~3300 words. Fork-changes.yaml entry + FORK_CHANGELOG.md + README's fork-change-queue table all regenerated.

Closes #47.

Test plan

  • scripts/render-docs.py regenerates FORK_CHANGELOG.md + README.md cleanly
  • scripts/check-docs.sh — all four checks green (commit hashes resolve, FORK_CHANGELOG matches YAML, upstream PR states match, fork hash references resolve). Pre-existing 2538 vs 2641 test-count drift in README body text is out of scope for this PR.
  • Essay cross-references resolve (three-patterns-for-agent-memory.md, three-mempalace-consumers.md, adaptmem-orthogonal-layers.md, convergent-findings-kostadis-comparison.md, 2026-05-05-verbatim-only-design.md, RFC 001, Dreams API canonical URL)
  • No code changes; docs-only PR

Copilot AI review requested due to automatic review settings May 23, 2026 00:57
@gemini-code-assist

Copy link
Copy Markdown

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

@jphein
jphein force-pushed the docs/verbatim-derivative-essay-47 branch from 3743b3a to e0892a1 Compare May 23, 2026 01:03
@jphein jphein mentioned this pull request May 23, 2026
1 task
jphein added 2 commits May 22, 2026 18:10
Develops MemPalace's foundational architectural call — store verbatim,
derive lazily, treat every derivative as replaceable — into a
standalone reference piece, distinct from the README's compressed
thesis section.

Seven sections covering the architectural choice, why derive-on-write
accumulates failure modes (lost nuance, classifier ceiling,
irreversibility, locked-in derivation assumptions), the empirical
2026-04-25 → 2026-05-05 recovery-collection episode (210× token gap
closed via structural change at the write layer), Anthropic's
Managed Agents Dreams API (beta header dreaming-2026-04-21) as
vendor-API ratification, operational implications, five named
exceptions where the axis does NOT apply (explicit shelving,
per-domain redaction, regulated-data classes, adversarial inputs,
high-cardinality machine output), and a three-piece conclusion.

Closes #47.
@jphein
jphein force-pushed the docs/verbatim-derivative-essay-47 branch from e0892a1 to 84a6fec Compare May 23, 2026 01:10
@jphein

jphein commented May 23, 2026

Copy link
Copy Markdown
Collaborator Author

Superseded by clean rebased PR.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Publish standalone essay on the verbatim-vs-derivative axis

2 participants