docs: rewrite arborist-one-pager + arborist-two-pager for external readers — drop ticket refs, schema versions, and internal vocabulary; reframe around user value (verified answers, fabricated-citation prevention, replay)

Old drafts read as internal substrate notes. Rewrites lead with what the system does for a consumer or evaluator and what it costs to run, with no references to internal tickets, table names, schema-version strings, governance hash dimensions, or per-record audit-mode tokens. Appendix diagrams updated in lockstep: friendly labels ("grounded / partly grounded / not grounded") replace the schema-column trichotomy, layer names paraphrased away from SURFACE/CORE/PROVIDENCE.
2026-05-14 10:02:08 -04:00 · 2026-05-14 10:02:08 -04:00 · 75ae470581
commit 75ae470581
parent 0aada29a8e
8 changed files with 296 additions and 326 deletions
--- a/docs/_source/arborist-one-pager.rst
+++ b/docs/_source/arborist-one-pager.rst
@ -1,64 +1,61 @@
-arborist — a Q&A store with cryptographically verified answers
-================================================================
+arborist — answers your AI can prove
+=====================================

 .. class:: center

-*one-page summary · 2026-05-14 · AGPL-3.0-only · permacomputer.com*
+*one-page summary · permacomputer.com · AGPL-3.0*

 ----

-Most retrieval-augmented systems hand a language model some context, ask
-for an answer, and ship whatever comes back. There is no
-machine-checkable claim about what the answer actually quotes from its
-source. **arborist inverts that.** Every answer is verified against its
-source *after* generation by a deterministic, lexical check that produces
-three labels — **STRICT** (every claim verifies), **HYBRID** (mixed),
-**UNGROUNDED** (none). The verifier is binary. It does not negotiate.
-Answers and labels are written to a content-addressed SQLite cache, every
-state change appends a SHA-256 audit-chain row, and a single tampered
-record breaks the chain.
+Most retrieval-augmented question-answering systems hand a language
+model some context and ship whatever the model says. There is no way to
+tell whether the answer faithfully reflects the source or whether the
+model embroidered it. **arborist closes that gap.** Every answer is
+verified against its source *after* the language model finishes, by a
+mechanical checker — not by another AI grading the first one. The
+checker labels each answer **grounded** (every claim was found in the
+source), **partly grounded** (some claims, not all), or **not
+grounded**. The label travels with the answer and is stored in a
+tamper-evident chain back to the source bytes.

-**Three Merkle-bound layers.** Surface (ingested documents — Wikipedia,
-HTML, TeX, git, claim-packs), Core (distilled documents bound to their
-surfaces by per-chunk inclusion proofs), Providence cache (verified Q&A
-keyed on an 8-dimensional invariant that folds in the source-corpus root,
-the question hash, the model identity, the governance policy, the schema
-version, and three normalization-version dimensions). Bumping any one of
-the eight dimensions invalidates prior records on lookup. Cores never
-evict.
+**Fabricated citations become impossible.** When the model answers, it
+never types the quoted text. arborist tags each candidate source
+chunk with a short label — ``E1``, ``E2`` — and asks the model to
+answer using those labels. The model might write *"Jupiter is the
+largest planet [E1], with a radius of about 70 000 km [E2]"*; arborist
+renders the actual chunk text at display time. A model cannot fabricate
+a quote it never types.

-**Synthetic-elision-by-construction-impossible.** In claim-lattice-pointer
-mode, the model never types the quote string. The runtime hands it
-tokens — ``[E1]``, ``[E2]`` — and interpolates the literal source span at
-render time. The model proposes pointer-tagged clauses, the runtime
-composes the answer. A class of hallucinations (the frankenquote, the
-recombined-token claim) becomes impossible because the model cannot type
-a string it never sees.
+**The proof path is cheap and mechanical.** Verification is text
+comparison, not embeddings, not similarity, not another model. It runs
+on a laptop. Optional smart-ranking layers — cross-encoder rerankers,
+entailment models — exist on the side; they help arborist find better
+evidence, they do not influence whether an answer is certified.

-**Soft signals never enter the proof path.** Deflection, coherence,
-title-relevance, NLI vetoes, cross-encoder rerankers — all live as
-advisory sidecars or as demote-only governance signals folded into the
-cache key. They influence what the substrate caches; they do not
-influence what it certifies. The hard channel (SHA-256 commitments,
-audit chain, proof-of-inclusion) and the soft channel (embeddings,
-similarity scores, lexical heuristics) evolve on independent timelines.
+**Same question, same document, same answer.** Answers are content-
+addressed: the cache key folds in the document, the question, the
+model identity, and the policy under which the answer was checked.
+Two users asking the same question of the same document under the same
+policy hit the same record. Reproducible. Replayable. Shareable.

-**What it measures on real traffic.**
+**What it has measured on real traffic.**

- **100% mis-cite catch at 0% real-STRICT false positive** on the
-  808-cell pooled bench-qa run (relevance-reranker veto,
-  BAAI/bge-reranker-large, n=20 mis-cite fixtures; ticket #000052 §3.2).
- **92 / 92 (100%) warrant-chain coverage** on the claim-pack corpus —
-  18 public-domain or open-licensed textbook substrates plus curated
-  alias chains (ticket #000031).
- **0.54 STRICT-rate** on Hermes-3-Llama-3.1-8B quote mode (n=3 × 71
-  questions, real Wikipedia haystack); lattice modes carry 99%
-  directive-coverage on the seven-point program.
+- **100% misattribution catch at 0% false positives** — every answer
+  where the cited source was unrelated to the claim was flagged,
+  without a single grounded answer wrongly demoted, across the full
+  pooled test bed.
+- **55–65% topic-deflection catch at 0–0.4% false positives** — picks
+  off-topic answers out of the stream while leaving on-topic answers
+  untouched.
+- **100% citation coverage on the curated textbook corpus** — every
+  cited claim resolves to a chain of evidence ending at a public-domain
+  or open-licensed source.

-**What it runs on.** Python 3.12, SQLite (FTS5 + JSON), optional
-cross-encoder rerankers for the soft channel. No GPU required for the
-proof path. AGPL-3.0-only. Reference inference endpoint:
+**What it costs.** Python 3.12. SQLite, one file (~2 GB for a
+Wikipedia-sized corpus). No GPU for the proof path. Use any
+OpenAI-compatible inference endpoint; the free reference endpoint is
 `hermes.ai.unturf.com <https://hermes.ai.unturf.com>`_. Source:
-`git.unturf.com/engineering/unturf/arborist <https://git.unturf.com/engineering/unturf/arborist>`_.
-Full whitepaper: `unfirehose.com/merkle-providence-reverse-rag.html
+`git.unturf.com/engineering/unturf/arborist
+<https://git.unturf.com/engineering/unturf/arborist>`_. Full
+whitepaper: `unfirehose.com/merkle-providence-reverse-rag.html
 <https://unfirehose.com/merkle-providence-reverse-rag.html>`_.
--- a/docs/_source/arborist-two-pager.rst
+++ b/docs/_source/arborist-two-pager.rst
@ -12,7 +12,7 @@ arborist — Merkle-Providence Reverse RAG, reference implementation

 .. class:: center

-**License: AGPL-3.0-only**  ·  schema v9.8  ·  2026-05-14
+**License: AGPL-3.0-only**  ·  2026-05-14
 ·  *Russell Ballestrini <russell@unturf.com>*

 ----
@ -20,154 +20,129 @@ arborist — Merkle-Providence Reverse RAG, reference implementation
 Abstract
 --------

-**arborist** is a content-addressed Q&A store with cryptographically
-verified answers, the reference runtime for the *Merkle Providence
-Reverse RAG* whitepaper (`unfirehose.com/merkle-providence-reverse-rag.html
-<https://unfirehose.com/merkle-providence-reverse-rag.html>`_,
-Ballestrini · Wong · Morgan, April 2026). Every cached answer carries a Merkle
-proof of source origin and a *binary, lexical, no-LLM-in-the-loop*
-faithfulness label. The verifier never asks a model to grade itself.
-Soft signals — embeddings, NLI vetoes, similarity rerankers — exist
-as sidecars and demote-only governance signals; they never enter the
-proof path. The reference inference endpoint is Hermes-3-Llama-3.1-8B
-served freely at `hermes.ai.unturf.com <https://hermes.ai.unturf.com>`_;
-arborist runs against any OpenAI-compatible alternative with a matching
-``model_profile_hash``.
+**arborist** is an answer-cache for retrieval-augmented question
+answering that *verifies every answer against its source* with a
+mechanical, deterministic checker — no second AI judging the first.
+Every cached answer carries a cryptographic chain back to the bytes
+of the source it came from. It is the reference implementation of the
+*Merkle Providence Reverse RAG* whitepaper
+(`unfirehose.com/merkle-providence-reverse-rag.html
+<https://unfirehose.com/merkle-providence-reverse-rag.html>`_), with
+a free reference inference endpoint at `hermes.ai.unturf.com
+<https://hermes.ai.unturf.com>`_; arborist also runs against any
+OpenAI-compatible alternative.

-1. The system in one paragraph
-------------------------------
+What it solves
+--------------

-Three SQLite-backed layers, all Merkle-bound. **Surface** holds ingested
-documents (Wikipedia dumps, HTML pages, TeX textbooks, git histories,
-claim-packs), chunked, hash-rooted, and FTS5-indexed. **Core** holds
-distillations bound to their surfaces by per-chunk inclusion proofs in
-``derivations.proof_blob``. **Providence cache** holds verified Q&A
-records keyed on an 8-dimensional invariant. Every state-changing
-operation appends one row to ``audit_events`` carrying
-``event_hash = sha256(prev || canonical(body))``; a single tampered row
-breaks the chain. Re-ingest is idempotent (same content → same
-``document_root`` → no-op); same URI with different content yields a new
-doc plus a ``supersedes`` edge (lossless history).
+Standard retrieval-augmented systems hand a language model some
+context and accept its answer. The user has no way to tell whether the
+answer faithfully reflects the source or whether the model improvised.
+arborist replaces that trust gap with a check: **every answer is
+verified, every answer is labelled, every answer carries a proof chain
+back to the bytes of the source it cites.**

-2. What makes arborist different
---------------------------------
+How it works
+------------

-Eleven design rules govern the substrate. The four that matter most for
-a reviewer:
+arborist keeps three layers in a single SQLite file. **Ingested
+documents** are split into chunks, hashed, and indexed for full-text
+search. **Derived layers** — summaries, indices, distillations — are
+hash-bound to the documents they came from. **An answer cache** stores
+every verified answer, keyed by the document, the question, the model
+that produced it, and the policy under which it was checked. Every
+write appends a hash-chained audit row; a single tampered record
+breaks the chain.

-**The verifier stays binary.** Each evidence unit verifies or does not.
-No per-unit confidence scores, no fuzzy-match indicators, no soft
-labels. Once a soft signal contaminates the chain, every prior
-cryptographic claim becomes negotiable.
+When you ask a question, arborist retrieves candidate source chunks,
+tags each one with a short label (``E1``, ``E2``, …), and passes them
+to the language model. The model emits prose with those labels — for
+example, *"Jupiter is the largest planet [E1], with a radius of about
+70 000 km [E2]"*. **The model never types the quoted text.** arborist
+composes the answer at display time, interpolating the actual chunk
+text at each label. A class of fabrication — the synthetic quote that
+looks like a verbatim citation but came out of the model's prior —
+becomes impossible by construction.

-**The runtime owns quote text.** In ``claim_lattice_pointer`` mode the
-model emits prose with bracketed pointer tags (``[E1]``, ``[E2]``); the
-runtime interpolates the literal source span at render time. The model
-never types the quote string. *Synthetic-elision-by-construction-impossible*
-— a model cannot produce a frankenquote it cannot type.
+After generation, the verifier runs. Four mechanical strategies
+(exact-quote match, line-span match, named-entity proximity,
+paraphrase coverage) try in sequence to confirm each claim against
+the cited chunks. The output is a single label per answer: **grounded**
+(every claim verifies), **partly grounded**, or **not grounded**.
+There is no "high confidence" middle ground — confidence labels are
+soft signals in disguise. The verifier names what it can prove and
+what it cannot.

-**Soft hash never enters the proof path.** Embeddings, TF-IDF scores,
-cross-encoder rerankers, NLI judgments shape ranking and retrieval and
-inform demote-only governance. They never feed ``cache_key`` ,
-``document_root``, or ``audit_event_hash``. The hard channel and the
-soft channel evolve on independent timelines.
+What makes it different
+-----------------------

-**Labels name properties, not vibes.** ``POINTER-LINKED``,
-``ANCHOR-WARRANTED``, ``EVIDENCE-WARRANTED``, ``UNGROUNDED`` each name a
-property the verifier could lexically confirm or could not. There are
-no "high confidence" labels. A reviewer reads the label and immediately
-knows what was checked.
+**The verifier is mechanical, not neural.** Faithfulness is a textual
+property of (answer, evidence, source) and is computed deterministically.
+No language model in the proof path means no model drift in the proof
+path.

-3. The 8-dimensional cache key
-------------------------------
+**The model cannot fabricate citations.** arborist sits between the
+model's output and the rendered answer; the model proposes references,
+arborist renders the content of those references from the actual
+source bytes.

-A cache hit returns a previously computed answer with a Merkle proof of
-origin; a cache miss triggers fresh inference. The key folds together:
+**Soft signals stay outside the proof.** Embedding similarity,
+cross-encoder rerankers, natural-language-inference judgments — all
+useful, all used to find and rank evidence; none of them allowed to
+influence the verification label or the audit chain. The cheap,
+provable path ships by default; the expensive, semantic path is
+opt-in and additive.

-================================  ===================================================
-``source_root``                   Merkle root of the corpus the answer is about
-``question_hash``                 dedup-mode-aware hash of the question (strict | equivalence_class)
-``model_profile_hash``            model identity (name + sampling profile)
-``conversation_hash``             prior-turn dependency, if any
-``governance_policy_hash``        every policy field that alters generation
-``schema_version``                ``v9.8.0``
-``canonicalization_version``      ``norm-v1``
-``chunking_version``              ``tok-512-v1``
-================================  ===================================================
+**Replay is free and exact.** Answers are content-addressed. The same
+question on the same document under the same policy returns the same
+record, with the same proof chain. Different document, different
+model, or different policy yields a different cache entry. The cache
+partitions cleanly; the corpus stays untouched.

-Bumping any one dimension invalidates prior records on lookup —
-governance evolves, the corpus does not get rewritten when governance
-evolves. Every preflight policy decision is *bound twice*: once into
-``governance_policy_hash`` (cache identity) and once into ``run_dag_root``
-via the preflight stage (audit replay).
-
-4. The layered verifier
------------------------
-
-Four strategies are tried in order; first to find evidence classifies.
-**quote** (sequential pair-matching, never regex — prevents phantom
-inter-pair captures), **span** (verbatim line match), **entity**
-(proximity-clustered proper nouns), **paraphrase** (token-coverage,
-prose-shaped only). Trichotomy on the schema column: STRICT = every unit
-verifies, HYBRID = mixed, UNGROUNDED = none. The renderer maps that
-column to a four-rung claim-lattice ladder
-(POINTER-LINKED → ANCHOR-WARRANTED → EVIDENCE-WARRANTED;
-ENTAILMENT-VERIFIED reserved); programmatic callers see the trichotomy,
-human-facing surfaces see the ladder. The ENTAILMENT-VERIFIED rung is
-deliberately empty — committing NLI to the proof path would re-introduce
-the soft/hard boundary leak the architecture is built to prevent.
-Honesty about the ceiling is a feature.
-
----
-
-5. What we measure on real traffic
-----------------------------------
-
-The bench is the scoreboard; the live fixtures are the gates. Numbers
-that survive the 5pp signal floor on real-shard, real-LLM runs:
-
- **Mis-cite catch (claim about X cited to source about Y).** The
-  ``BAAI/bge-reranker-large`` veto at threshold ``θ ≤ -2.42`` catches
-  20/20 (100%) of mis-cite fixtures at strictly 0/808 (0%) real-STRICT
-  false positive on the pooled bench-qa, with a 1.4-point margin
-  between the worst mis-cite and the best STRICT (#000052 §3.2).
- **Deflection catch (off-topic answer to an on-topic question).** Same
-  reranker catches 11/20 (55%) of real-haystack deflection fixtures at
-  the same 0/808 FP threshold. The cost-pick alternate
-  (``ms-marco-MiniLM-L-6-v2``, 5× smaller) catches 13/20 (65%) at 0.4% FP.
- **Recombination catch (right tokens, wrong subject).** The NLI veto
-  (``facebook/bart-large-mnli``, top-k=12, θc=0.999) catches ~48% of
-  hand-crafted real-haystack recombination fixtures at 0/808 FP; the
-  ceiling is selector-bound, not entailment-bound (#000049 §7 #27).
- **Warrant-chain coverage.** 92/92 (100%) of the claim-pack corpus
-  resolves to a warrant chain (18 public-domain or open-licensed
-  textbook substrates + curated citation aliases + term aliases;
-  ticket #000031).
- **STRICT-rate, quote mode, real Wikipedia haystack.** 0.54 on
-  Hermes-3-Llama-3.1-8B (n=3 × 71 questions, sample-shuffled @ c=4;
-  see ``docs/qa-modes-bench.md``).
-
-Sample sizes are stated. Synthetic shortcuts are not. The five-point
-sweep-wide/sweep-deep discipline ("real-traffic-sized samples or it
-didn't happen") lives in ``docs/bench-maxing.md``.
-
-6. Closed substrate work
+What you would run it for
 -------------------------

-Recent substrate work that is *landed and pinned*: ticket #000005
-(four-rung ladder display layer), #000008 (broad-quantifier preflight
-guard with six-level disable hierarchy), #000010 (metacognition
-preflight — false-premise, contradiction, stale-risk, out-of-corpus),
-#000015 (π* canonical projections — arithmetic, logic-kernel,
-algebra-symbolic), #000027 (canonical-projection persistence in
-providence cache), #000031 (claim-pack warrant resolver, 100%
-coverage), #000049 §3 (NLI veto, partial closure), #000052 §3.1
-(coherence sidecar — circular, vacuous, phrase-component-reuse) and
-§3.2 (relevance reranker, viable runtime soft-veto). The synthesis
-brief for the independent review thread is
-``docs/relevance-and-veto-synthesis-for-dav1d.md``.
+You hold a corpus — internal documents, public scientific literature,
+a legal archive, a textbook collection — and you want reliable Q&A
+over it without paying a closed-source vendor or wiring up a vector
+database. You care that the answers track your sources. You want users
+to be able to verify, replay, and share the proof. You want it to keep
+working when the model changes, when the policy changes, when the
+corpus version-bumps.

----
+What it costs
+-------------
+
+==================  ============================================================
+Language            Python 3.12
+Storage             SQLite — one file, ~2 GB for a Wikipedia-sized corpus
+Compute             No GPU required for the proof path. Inference runs wherever
+                    you want — locally, or via an OpenAI-compatible endpoint.
+Optional            Cross-encoder rerankers (80–560 MB) for harder relevance
+                    ranking. They never enter the proof.
+License             AGPL-3.0-only
+==================  ============================================================
+
+What it has measured on real traffic
+-------------------------------------
+
+- **Misattribution detection.** Catches 100% of answers where the cited
+  source is unrelated to the claim — at strictly zero false positives
+  across the full pooled test bed (cross-encoder reranker, hard
+  threshold, real Wikipedia-haystack questions).
+
+- **Topic-deflection detection.** Catches 55–65% of off-topic answers
+  at 0–0.4% false positives, depending on the reranker model chosen
+  (a small 80 MB model and a larger 560 MB model both clear the bar).
+
+- **Citation coverage.** On the curated textbook corpus, 100% of cited
+  claims resolve to a chain of evidence ending at a public-domain or
+  open-licensed source — eighteen textbooks plus curated cross-citation
+  aliases.
+
+Sample sizes are stated, sample sources are real, and the gates are
+reported against the same numbers an external reviewer can reproduce
+from the repository.

 License — Permacomputer Preamble
 ---------------------------------
@ -188,24 +163,25 @@ License — Permacomputer Preamble

 ----

-Appendix A — Three-layer architecture
--------------------------------------
+Appendix A — The three layers
+------------------------------

 .. figure:: diagrams/pager-arch-stack.png
   :width: 5.5in
   :align: center

-   Surface, Core, Providence cache — each layer hash-binds to the next;
-   every state change appends a row to the audit chain.
+   Ingested documents, derived layers, and the answer cache — each
+   layer hash-binds to the next. Every state change appends one row
+   to the audit chain; a single tampered record breaks the chain.

-Appendix B — Verifier flow
---------------------------
+Appendix B — How an answer is verified
+---------------------------------------

 .. figure:: diagrams/pager-verifier-flow.png
   :width: 6.5in
   :align: center

-   Question → retrieval → pointer-tagged context → LLM → binary
-   verifier → trichotomy label → audit chain & providence cache. The
-   LLM never types the quote string.
-
+   The model sees labelled source chunks and emits prose tagged with
+   those labels — it never types the quote text. A mechanical
+   verifier then checks every cited claim against the actual source
+   and assigns the answer one of three labels.
--- a/docs/diagrams/pager-arch-stack.dot
+++ b/docs/diagrams/pager-arch-stack.dot
@ -1,4 +1,5 @@
 // Compact 3-layer architecture for the 2-pager appendix.
+// External-facing labels — no schema-column or table-name vocabulary.
 // Render: dot -Tsvg pager-arch-stack.dot -o pager-arch-stack.svg

 digraph arch_stack {
@ -9,11 +10,11 @@ digraph arch_stack {
    edge [fontname="Helvetica", fontsize=9]
    bgcolor="white"

-    surface [label="SURFACE — ingested documents\nWikipedia · HTML · TeX · git · claim-packs\nchunked → Merkle-rooted → FTS5-indexed", fillcolor="#e9f3ff"]
-    core    [label="CORE — distilled documents\nMerkle-bound to surfaces by per-chunk\ninclusion proofs (derivations.proof_blob)", fillcolor="#dff5e7"]
-    cache   [label="PROVIDENCE CACHE — verified Q&A\n8-dim cache_key · audit_mode label ∈\n{STRICT, HYBRID, UNGROUNDED}", fillcolor="#fff3d6"]
+    surface [label="Ingested documents\n(Wikipedia · HTML · TeX · git · claim-packs)\nchunked → hashed → full-text-indexed", fillcolor="#e9f3ff"]
+    core    [label="Derived layers\n(summaries, indices, distillations)\nhash-bound back to source chunks", fillcolor="#dff5e7"]
+    cache   [label="Answer cache\nverified Q&A · grounded / partly / not grounded\ncryptographic chain back to source bytes", fillcolor="#fff3d6"]

-    audit [label="AUDIT CHAIN — every state change\nappends sha256(prev || canonical(body))\nto audit_events (single SQLite file)", fillcolor="#fde7e9", shape=note]
+    audit [label="Audit chain\nevery write appends a hash-linked row;\na single tampered record breaks the chain", fillcolor="#fde7e9", shape=note]

    surface -> core   [label="distill", fontsize=9]
    core    -> cache  [label="verify → cache", fontsize=9]
--- a/docs/diagrams/pager-arch-stack.png
+++ b/docs/diagrams/pager-arch-stack.png
--- a/docs/diagrams/pager-arch-stack.svg
+++ b/docs/diagrams/pager-arch-stack.svg
@ -4,77 +4,77 @@
 <!-- Generated by graphviz version 2.43.0 (0)
 -->
 <!-- Title: arch_stack Pages: 1 -->
-<svg width="379pt" height="268pt"
- viewBox="0.00 0.00 379.00 268.00" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink">
+<svg width="434pt" height="268pt"
+ viewBox="0.00 0.00 433.50 268.00" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink">
 <g id="graph0" class="graph" transform="scale(1 1) rotate(0) translate(4 264)">
 <title>arch_stack</title>
-<polygon fill="white" stroke="transparent" points="-4,4 -4,-264 375,-264 375,4 -4,4"/>
+<polygon fill="white" stroke="transparent" points="-4,4 -4,-264 429.5,-264 429.5,4 -4,4"/>
 <!-- surface -->
 <g id="node1" class="node">
 <title>surface</title>
-<path fill="#e9f3ff" stroke="black" d="M319,-44C319,-44 98,-44 98,-44 92,-44 86,-38 86,-32 86,-32 86,-12 86,-12 86,-6 92,0 98,0 98,0 319,0 319,0 325,0 331,-6 331,-12 331,-12 331,-32 331,-32 331,-38 325,-44 319,-44"/>
-<text text-anchor="middle" x="208.5" y="-31.2" font-family="Helvetica,sans-Serif" font-size="11.00">SURFACE — ingested documents</text>
-<text text-anchor="middle" x="208.5" y="-19.2" font-family="Helvetica,sans-Serif" font-size="11.00">Wikipedia · HTML · TeX · git · claim&#45;packs</text>
-<text text-anchor="middle" x="208.5" y="-7.2" font-family="Helvetica,sans-Serif" font-size="11.00">chunked → Merkle&#45;rooted → FTS5&#45;indexed</text>
+<path fill="#e9f3ff" stroke="black" d="M380.5,-44C380.5,-44 153.5,-44 153.5,-44 147.5,-44 141.5,-38 141.5,-32 141.5,-32 141.5,-12 141.5,-12 141.5,-6 147.5,0 153.5,0 153.5,0 380.5,0 380.5,0 386.5,0 392.5,-6 392.5,-12 392.5,-12 392.5,-32 392.5,-32 392.5,-38 386.5,-44 380.5,-44"/>
+<text text-anchor="middle" x="267" y="-31.2" font-family="Helvetica,sans-Serif" font-size="11.00">Ingested documents</text>
+<text text-anchor="middle" x="267" y="-19.2" font-family="Helvetica,sans-Serif" font-size="11.00">(Wikipedia · HTML · TeX · git · claim&#45;packs)</text>
+<text text-anchor="middle" x="267" y="-7.2" font-family="Helvetica,sans-Serif" font-size="11.00">chunked → hashed → full&#45;text&#45;indexed</text>
 </g>
 <!-- core -->
 <g id="node2" class="node">
 <title>core</title>
-<path fill="#dff5e7" stroke="black" d="M245.5,-116C245.5,-116 31.5,-116 31.5,-116 25.5,-116 19.5,-110 19.5,-104 19.5,-104 19.5,-84 19.5,-84 19.5,-78 25.5,-72 31.5,-72 31.5,-72 245.5,-72 245.5,-72 251.5,-72 257.5,-78 257.5,-84 257.5,-84 257.5,-104 257.5,-104 257.5,-110 251.5,-116 245.5,-116"/>
-<text text-anchor="middle" x="138.5" y="-103.2" font-family="Helvetica,sans-Serif" font-size="11.00">CORE — distilled documents</text>
-<text text-anchor="middle" x="138.5" y="-91.2" font-family="Helvetica,sans-Serif" font-size="11.00">Merkle&#45;bound to surfaces by per&#45;chunk</text>
-<text text-anchor="middle" x="138.5" y="-79.2" font-family="Helvetica,sans-Serif" font-size="11.00">inclusion proofs (derivations.proof_blob)</text>
+<path fill="#dff5e7" stroke="black" d="M295.5,-116C295.5,-116 112.5,-116 112.5,-116 106.5,-116 100.5,-110 100.5,-104 100.5,-104 100.5,-84 100.5,-84 100.5,-78 106.5,-72 112.5,-72 112.5,-72 295.5,-72 295.5,-72 301.5,-72 307.5,-78 307.5,-84 307.5,-84 307.5,-104 307.5,-104 307.5,-110 301.5,-116 295.5,-116"/>
+<text text-anchor="middle" x="204" y="-103.2" font-family="Helvetica,sans-Serif" font-size="11.00">Derived layers</text>
+<text text-anchor="middle" x="204" y="-91.2" font-family="Helvetica,sans-Serif" font-size="11.00">(summaries, indices, distillations)</text>
+<text text-anchor="middle" x="204" y="-79.2" font-family="Helvetica,sans-Serif" font-size="11.00">hash&#45;bound back to source chunks</text>
 </g>
 <!-- surface&#45;&gt;core -->
 <g id="edge1" class="edge">
 <title>surface&#45;&gt;core</title>
-<path fill="none" stroke="black" d="M187.5,-44C181.11,-50.39 173.98,-57.52 167.17,-64.33"/>
-<polygon fill="black" stroke="black" points="164.57,-61.98 159.97,-71.53 169.52,-66.93 164.57,-61.98"/>
-<text text-anchor="middle" x="190.5" y="-55.8" font-family="Helvetica,sans-Serif" font-size="9.00">distill</text>
+<path fill="none" stroke="black" d="M248.1,-44C242.47,-50.25 236.2,-57.22 230.2,-63.89"/>
+<polygon fill="black" stroke="black" points="227.41,-61.75 223.32,-71.53 232.62,-66.44 227.41,-61.75"/>
+<text text-anchor="middle" x="252" y="-55.8" font-family="Helvetica,sans-Serif" font-size="9.00">distill</text>
 </g>
 <!-- audit -->
 <g id="node4" class="node">
 <title>audit</title>
-<polygon fill="#fde7e9" stroke="black" points="365,-260 128,-260 128,-216 371,-216 371,-254 365,-260"/>
-<polyline fill="none" stroke="black" points="365,-260 365,-254 "/>
-<polyline fill="none" stroke="black" points="371,-254 365,-254 "/>
-<text text-anchor="middle" x="249.5" y="-247.2" font-family="Helvetica,sans-Serif" font-size="11.00">AUDIT CHAIN — every state change</text>
-<text text-anchor="middle" x="249.5" y="-235.2" font-family="Helvetica,sans-Serif" font-size="11.00">appends sha256(prev || canonical(body))</text>
-<text text-anchor="middle" x="249.5" y="-223.2" font-family="Helvetica,sans-Serif" font-size="11.00">to audit_events (single SQLite file)</text>
+<polygon fill="#fde7e9" stroke="black" points="419.5,-260 174.5,-260 174.5,-216 425.5,-216 425.5,-254 419.5,-260"/>
+<polyline fill="none" stroke="black" points="419.5,-260 419.5,-254 "/>
+<polyline fill="none" stroke="black" points="425.5,-254 419.5,-254 "/>
+<text text-anchor="middle" x="300" y="-247.2" font-family="Helvetica,sans-Serif" font-size="11.00">Audit chain</text>
+<text text-anchor="middle" x="300" y="-235.2" font-family="Helvetica,sans-Serif" font-size="11.00">every write appends a hash&#45;linked row;</text>
+<text text-anchor="middle" x="300" y="-223.2" font-family="Helvetica,sans-Serif" font-size="11.00">a single tampered record breaks the chain</text>
 </g>
 <!-- surface&#45;&gt;audit -->
 <g id="edge5" class="edge">
 <title>surface&#45;&gt;audit</title>
-<path fill="none" stroke="#888888" stroke-dasharray="5,2" d="M241.95,-44.24C250.98,-51.8 259.59,-61.18 264.5,-72 270.08,-84.3 269.38,-158.79 260.95,-206.05"/>
-<polygon fill="#888888" stroke="#888888" points="257.49,-205.49 258.96,-215.98 264.35,-206.86 257.49,-205.49"/>
+<path fill="none" stroke="#888888" stroke-dasharray="5,2" d="M294.75,-44.09C302.85,-51.88 310.68,-61.42 315,-72 320.38,-85.18 319.5,-160.34 311.42,-206.13"/>
+<polygon fill="#888888" stroke="#888888" points="307.98,-205.49 309.46,-215.98 314.84,-206.85 307.98,-205.49"/>
 </g>
 <!-- cache -->
 <g id="node3" class="node">
 <title>cache</title>
-<path fill="#fff3d6" stroke="black" d="M215,-188C215,-188 12,-188 12,-188 6,-188 0,-182 0,-176 0,-176 0,-156 0,-156 0,-150 6,-144 12,-144 12,-144 215,-144 215,-144 221,-144 227,-150 227,-156 227,-156 227,-176 227,-176 227,-182 221,-188 215,-188"/>
-<text text-anchor="middle" x="113.5" y="-175.2" font-family="Helvetica,sans-Serif" font-size="11.00">PROVIDENCE CACHE — verified Q&amp;A</text>
-<text text-anchor="middle" x="113.5" y="-163.2" font-family="Helvetica,sans-Serif" font-size="11.00">8&#45;dim cache_key · audit_mode label ∈</text>
-<text text-anchor="middle" x="113.5" y="-151.2" font-family="Helvetica,sans-Serif" font-size="11.00">{STRICT, HYBRID, UNGROUNDED}</text>
+<path fill="#fff3d6" stroke="black" d="M266,-188C266,-188 12,-188 12,-188 6,-188 0,-182 0,-176 0,-176 0,-156 0,-156 0,-150 6,-144 12,-144 12,-144 266,-144 266,-144 272,-144 278,-150 278,-156 278,-156 278,-176 278,-176 278,-182 272,-188 266,-188"/>
+<text text-anchor="middle" x="139" y="-175.2" font-family="Helvetica,sans-Serif" font-size="11.00">Answer cache</text>
+<text text-anchor="middle" x="139" y="-163.2" font-family="Helvetica,sans-Serif" font-size="11.00">verified Q&amp;A · grounded / partly / not grounded</text>
+<text text-anchor="middle" x="139" y="-151.2" font-family="Helvetica,sans-Serif" font-size="11.00">cryptographic chain back to source bytes</text>
 </g>
 <!-- core&#45;&gt;cache -->
 <g id="edge2" class="edge">
 <title>core&#45;&gt;cache</title>
-<path fill="none" stroke="black" d="M131,-116C129.01,-121.57 126.82,-127.71 124.68,-133.71"/>
-<polygon fill="black" stroke="black" points="121.24,-132.93 121.17,-143.53 127.83,-135.29 121.24,-132.93"/>
-<text text-anchor="middle" x="160.5" y="-127.8" font-family="Helvetica,sans-Serif" font-size="9.00">verify → cache</text>
+<path fill="none" stroke="black" d="M184.5,-116C178.69,-122.25 172.22,-129.22 166.03,-135.89"/>
+<polygon fill="black" stroke="black" points="163.18,-133.82 158.94,-143.53 168.31,-138.58 163.18,-133.82"/>
+<text text-anchor="middle" x="209" y="-127.8" font-family="Helvetica,sans-Serif" font-size="9.00">verify → cache</text>
 </g>
 <!-- core&#45;&gt;audit -->
 <g id="edge4" class="edge">
 <title>core&#45;&gt;audit</title>
-<path fill="none" stroke="#888888" stroke-dasharray="5,2" d="M200.73,-116.01C213.64,-123.04 225.91,-132.2 234.5,-144 247.33,-161.63 250.87,-186.11 251.31,-205.54"/>
-<polygon fill="#888888" stroke="#888888" points="247.81,-205.72 251.28,-215.73 254.81,-205.74 247.81,-205.72"/>
+<path fill="none" stroke="#888888" stroke-dasharray="5,2" d="M254.11,-116.13C265.91,-123.34 277.28,-132.55 285,-144 297.19,-162.08 300.75,-186.55 301.37,-205.85"/>
+<polygon fill="#888888" stroke="#888888" points="297.87,-205.99 301.44,-215.97 304.87,-205.94 297.87,-205.99"/>
 </g>
 <!-- cache&#45;&gt;audit -->
 <g id="edge3" class="edge">
 <title>cache&#45;&gt;audit</title>
-<path fill="none" stroke="#888888" stroke-dasharray="5,2" d="M151.73,-188.09C163.18,-194.29 175.8,-201.01 187.5,-207 190.37,-208.47 193.33,-209.97 196.32,-211.46"/>
-<polygon fill="#888888" stroke="#888888" points="194.83,-214.63 205.35,-215.94 197.94,-208.36 194.83,-214.63"/>
-<text text-anchor="middle" x="213" y="-199.8" font-family="Helvetica,sans-Serif" font-size="9.00">every write</text>
+<path fill="none" stroke="#888888" stroke-dasharray="5,2" d="M187.31,-188C204.58,-195.51 224.22,-204.05 242.21,-211.88"/>
+<polygon fill="#888888" stroke="#888888" points="240.88,-215.11 251.45,-215.89 243.68,-208.69 240.88,-215.11"/>
+<text text-anchor="middle" x="255.5" y="-199.8" font-family="Helvetica,sans-Serif" font-size="9.00">every write</text>
 </g>
 </g>
 </svg>
--- a/docs/diagrams/pager-verifier-flow.dot
+++ b/docs/diagrams/pager-verifier-flow.dot
@ -1,5 +1,5 @@
 // Compact verifier flow for the 2-pager appendix.
-// (Question, source-chunks) → model → (text, pointers) → verifier → label.
+// External-facing labels — no internal vocabulary.
 // Render: dot -Tsvg pager-verifier-flow.dot -o pager-verifier-flow.svg

 digraph verifier_flow {
@ -10,27 +10,27 @@ digraph verifier_flow {
    edge [fontname="Helvetica", fontsize=11]
    bgcolor="white"

-    question [label="QUESTION", fillcolor="#fff7e6", shape=note]
-    retrieval [label="retrieval\n(FTS5 + phrase route\n+ rivalry exclusion)", fillcolor="#e9f3ff"]
-    pointers [label="POINTER MAP\nE1 → span\nE2 → span\n…", fillcolor="#dff5e7", shape=note]
-    model [label="LLM\nsees: question +\npointer-tagged chunks\nemits: prose with [E1] [E2] tags\n(never types the quote string)", fillcolor="#f0e6ff"]
-    verifier [label="verifier\n(quote · span · entity\n· paraphrase — binary,\nno LLM call)", fillcolor="#fff3d6"]
+    question  [label="Question", fillcolor="#fff7e6", shape=note]
+    retrieval [label="Retrieve\ncandidate source chunks", fillcolor="#e9f3ff"]
+    pointers  [label="Labelled chunks\nE1 → text\nE2 → text\n…", fillcolor="#dff5e7", shape=note]
+    model     [label="Language model\nsees: question + labelled chunks\nemits: prose tagged [E1] [E2]\n(never types the quote text)", fillcolor="#f0e6ff"]
+    verifier  [label="Verifier\nfour mechanical strategies,\nno model in the loop", fillcolor="#fff3d6"]

-    label_strict [label="STRICT\nevery claim verifies", fillcolor="#28a745", fontcolor="white"]
-    label_hybrid [label="HYBRID\nsome verify, some don't", fillcolor="#ffc107"]
-    label_ungrounded [label="UNGROUNDED\nnone verify", fillcolor="#dc3545", fontcolor="white"]
+    label_grounded    [label="grounded\nevery claim verifies", fillcolor="#28a745", fontcolor="white"]
+    label_partial     [label="partly grounded\nsome verify, some don't", fillcolor="#ffc107"]
+    label_ungrounded  [label="not grounded\nnone verify", fillcolor="#dc3545", fontcolor="white"]

-    audit [label="audit chain\n+ providence cache\n(8-dim key, governance-aware)", fillcolor="#fde7e9", shape=note]
+    audit [label="Audit chain\n+ answer cache\n(reproducible, replayable)", fillcolor="#fde7e9", shape=note]

    question -> retrieval -> pointers
    pointers -> model
-    model -> verifier [label="answer\n+ pointer IDs"]
-    verifier -> label_strict
-    verifier -> label_hybrid
+    model -> verifier [label="answer with [E1] [E2] tags"]
+    verifier -> label_grounded
+    verifier -> label_partial
    verifier -> label_ungrounded
-    label_strict     -> audit [style=dashed, color="#888888"]
-    label_hybrid     -> audit [style=dashed, color="#888888"]
+    label_grounded   -> audit [style=dashed, color="#888888"]
+    label_partial    -> audit [style=dashed, color="#888888"]
    label_ungrounded -> audit [style=dashed, color="#888888"]

-    {rank=same; label_strict; label_hybrid; label_ungrounded}
+    {rank=same; label_grounded; label_partial; label_ungrounded}
 }
--- a/docs/diagrams/pager-verifier-flow.png
+++ b/docs/diagrams/pager-verifier-flow.png
--- a/docs/diagrams/pager-verifier-flow.svg
+++ b/docs/diagrams/pager-verifier-flow.svg
@ -4,149 +4,145 @@
 <!-- Generated by graphviz version 2.43.0 (0)
 -->
 <!-- Title: verifier_flow Pages: 1 -->
-<svg width="1288pt" height="152pt"
- viewBox="0.00 0.00 1288.00 152.00" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink">
+<svg width="1389pt" height="152pt"
+ viewBox="0.00 0.00 1389.00 152.00" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink">
 <g id="graph0" class="graph" transform="scale(1 1) rotate(0) translate(4 148)">
 <title>verifier_flow</title>
-<polygon fill="white" stroke="transparent" points="-4,4 -4,-148 1284,-148 1284,4 -4,4"/>
+<polygon fill="white" stroke="transparent" points="-4,4 -4,-148 1385,-148 1385,4 -4,4"/>
 <!-- question -->
 <g id="node1" class="node">
 <title>question</title>
-<polygon fill="#fff7e6" stroke="black" points="74,-90 0,-90 0,-54 80,-54 80,-84 74,-90"/>
-<polyline fill="none" stroke="black" points="74,-90 74,-84 "/>
-<polyline fill="none" stroke="black" points="80,-84 74,-84 "/>
-<text text-anchor="middle" x="40" y="-68.9" font-family="Helvetica,sans-Serif" font-size="12.00">QUESTION</text>
+<polygon fill="#fff7e6" stroke="black" points="64,-90 0,-90 0,-54 70,-54 70,-84 64,-90"/>
+<polyline fill="none" stroke="black" points="64,-90 64,-84 "/>
+<polyline fill="none" stroke="black" points="70,-84 64,-84 "/>
+<text text-anchor="middle" x="35" y="-68.9" font-family="Helvetica,sans-Serif" font-size="12.00">Question</text>
 </g>
 <!-- retrieval -->
 <g id="node2" class="node">
 <title>retrieval</title>
-<path fill="#e9f3ff" stroke="black" d="M239,-95.5C239,-95.5 119,-95.5 119,-95.5 113,-95.5 107,-89.5 107,-83.5 107,-83.5 107,-60.5 107,-60.5 107,-54.5 113,-48.5 119,-48.5 119,-48.5 239,-48.5 239,-48.5 245,-48.5 251,-54.5 251,-60.5 251,-60.5 251,-83.5 251,-83.5 251,-89.5 245,-95.5 239,-95.5"/>
-<text text-anchor="middle" x="179" y="-81.9" font-family="Helvetica,sans-Serif" font-size="12.00">retrieval</text>
-<text text-anchor="middle" x="179" y="-68.9" font-family="Helvetica,sans-Serif" font-size="12.00">(FTS5 + phrase route</text>
-<text text-anchor="middle" x="179" y="-55.9" font-family="Helvetica,sans-Serif" font-size="12.00">+ rivalry exclusion)</text>
+<path fill="#e9f3ff" stroke="black" d="M250,-90C250,-90 109,-90 109,-90 103,-90 97,-84 97,-78 97,-78 97,-66 97,-66 97,-60 103,-54 109,-54 109,-54 250,-54 250,-54 256,-54 262,-60 262,-66 262,-66 262,-78 262,-78 262,-84 256,-90 250,-90"/>
+<text text-anchor="middle" x="179.5" y="-75.4" font-family="Helvetica,sans-Serif" font-size="12.00">Retrieve</text>
+<text text-anchor="middle" x="179.5" y="-62.4" font-family="Helvetica,sans-Serif" font-size="12.00">candidate source chunks</text>
 </g>
 <!-- question&#45;&gt;retrieval -->
 <g id="edge1" class="edge">
 <title>question&#45;&gt;retrieval</title>
-<path fill="none" stroke="black" d="M80.26,-72C85.5,-72 91.02,-72 96.66,-72"/>
-<polygon fill="black" stroke="black" points="96.9,-75.5 106.9,-72 96.9,-68.5 96.9,-75.5"/>
+<path fill="none" stroke="black" d="M70.13,-72C75.41,-72 81.06,-72 86.91,-72"/>
+<polygon fill="black" stroke="black" points="86.91,-75.5 96.91,-72 86.91,-68.5 86.91,-75.5"/>
 </g>
 <!-- pointers -->
 <g id="node3" class="node">
 <title>pointers</title>
-<polygon fill="#dff5e7" stroke="black" points="372,-102 278,-102 278,-42 378,-42 378,-96 372,-102"/>
-<polyline fill="none" stroke="black" points="372,-102 372,-96 "/>
-<polyline fill="none" stroke="black" points="378,-96 372,-96 "/>
-<text text-anchor="middle" x="328" y="-88.4" font-family="Helvetica,sans-Serif" font-size="12.00">POINTER MAP</text>
-<text text-anchor="middle" x="328" y="-75.4" font-family="Helvetica,sans-Serif" font-size="12.00">E1 → span</text>
-<text text-anchor="middle" x="328" y="-62.4" font-family="Helvetica,sans-Serif" font-size="12.00">E2 → span</text>
-<text text-anchor="middle" x="328" y="-49.4" font-family="Helvetica,sans-Serif" font-size="12.00">…</text>
+<polygon fill="#dff5e7" stroke="black" points="396,-102 289,-102 289,-42 402,-42 402,-96 396,-102"/>
+<polyline fill="none" stroke="black" points="396,-102 396,-96 "/>
+<polyline fill="none" stroke="black" points="402,-96 396,-96 "/>
+<text text-anchor="middle" x="345.5" y="-88.4" font-family="Helvetica,sans-Serif" font-size="12.00">Labelled chunks</text>
+<text text-anchor="middle" x="345.5" y="-75.4" font-family="Helvetica,sans-Serif" font-size="12.00">E1 → text</text>
+<text text-anchor="middle" x="345.5" y="-62.4" font-family="Helvetica,sans-Serif" font-size="12.00">E2 → text</text>
+<text text-anchor="middle" x="345.5" y="-49.4" font-family="Helvetica,sans-Serif" font-size="12.00">…</text>
 </g>
 <!-- retrieval&#45;&gt;pointers -->
 <g id="edge2" class="edge">
 <title>retrieval&#45;&gt;pointers</title>
-<path fill="none" stroke="black" d="M251.35,-72C256.79,-72 262.24,-72 267.59,-72"/>
-<polygon fill="black" stroke="black" points="267.72,-75.5 277.72,-72 267.72,-68.5 267.72,-75.5"/>
+<path fill="none" stroke="black" d="M262.02,-72C267.51,-72 273,-72 278.39,-72"/>
+<polygon fill="black" stroke="black" points="278.62,-75.5 288.62,-72 278.62,-68.5 278.62,-75.5"/>
 </g>
 <!-- model -->
 <g id="node4" class="node">
 <title>model</title>
-<path fill="#f0e6ff" stroke="black" d="M596,-108.5C596,-108.5 417,-108.5 417,-108.5 411,-108.5 405,-102.5 405,-96.5 405,-96.5 405,-47.5 405,-47.5 405,-41.5 411,-35.5 417,-35.5 417,-35.5 596,-35.5 596,-35.5 602,-35.5 608,-41.5 608,-47.5 608,-47.5 608,-96.5 608,-96.5 608,-102.5 602,-108.5 596,-108.5"/>
-<text text-anchor="middle" x="506.5" y="-94.9" font-family="Helvetica,sans-Serif" font-size="12.00">LLM</text>
-<text text-anchor="middle" x="506.5" y="-81.9" font-family="Helvetica,sans-Serif" font-size="12.00">sees: question +</text>
-<text text-anchor="middle" x="506.5" y="-68.9" font-family="Helvetica,sans-Serif" font-size="12.00">pointer&#45;tagged chunks</text>
-<text text-anchor="middle" x="506.5" y="-55.9" font-family="Helvetica,sans-Serif" font-size="12.00">emits: prose with [E1] [E2] tags</text>
-<text text-anchor="middle" x="506.5" y="-42.9" font-family="Helvetica,sans-Serif" font-size="12.00">(never types the quote string)</text>
+<path fill="#f0e6ff" stroke="black" d="M629,-102C629,-102 441,-102 441,-102 435,-102 429,-96 429,-90 429,-90 429,-54 429,-54 429,-48 435,-42 441,-42 441,-42 629,-42 629,-42 635,-42 641,-48 641,-54 641,-54 641,-90 641,-90 641,-96 635,-102 629,-102"/>
+<text text-anchor="middle" x="535" y="-88.4" font-family="Helvetica,sans-Serif" font-size="12.00">Language model</text>
+<text text-anchor="middle" x="535" y="-75.4" font-family="Helvetica,sans-Serif" font-size="12.00">sees: question + labelled chunks</text>
+<text text-anchor="middle" x="535" y="-62.4" font-family="Helvetica,sans-Serif" font-size="12.00">emits: prose tagged [E1] [E2]</text>
+<text text-anchor="middle" x="535" y="-49.4" font-family="Helvetica,sans-Serif" font-size="12.00">(never types the quote text)</text>
 </g>
 <!-- pointers&#45;&gt;model -->
 <g id="edge3" class="edge">
 <title>pointers&#45;&gt;model</title>
-<path fill="none" stroke="black" d="M378.17,-72C383.46,-72 388.98,-72 394.64,-72"/>
-<polygon fill="black" stroke="black" points="394.95,-75.5 404.95,-72 394.95,-68.5 394.95,-75.5"/>
+<path fill="none" stroke="black" d="M402.25,-72C407.58,-72 413.11,-72 418.74,-72"/>
+<polygon fill="black" stroke="black" points="418.99,-75.5 428.99,-72 418.99,-68.5 418.99,-75.5"/>
 </g>
 <!-- verifier -->
 <g id="node5" class="node">
 <title>verifier</title>
-<path fill="#fff3d6" stroke="black" d="M849,-102C849,-102 719,-102 719,-102 713,-102 707,-96 707,-90 707,-90 707,-54 707,-54 707,-48 713,-42 719,-42 719,-42 849,-42 849,-42 855,-42 861,-48 861,-54 861,-54 861,-90 861,-90 861,-96 855,-102 849,-102"/>
-<text text-anchor="middle" x="784" y="-88.4" font-family="Helvetica,sans-Serif" font-size="12.00">verifier</text>
-<text text-anchor="middle" x="784" y="-75.4" font-family="Helvetica,sans-Serif" font-size="12.00">(quote · span · entity</text>
-<text text-anchor="middle" x="784" y="-62.4" font-family="Helvetica,sans-Serif" font-size="12.00">· paraphrase — binary,</text>
-<text text-anchor="middle" x="784" y="-49.4" font-family="Helvetica,sans-Serif" font-size="12.00">no LLM call)</text>
+<path fill="#fff3d6" stroke="black" d="M984,-95.5C984,-95.5 827,-95.5 827,-95.5 821,-95.5 815,-89.5 815,-83.5 815,-83.5 815,-60.5 815,-60.5 815,-54.5 821,-48.5 827,-48.5 827,-48.5 984,-48.5 984,-48.5 990,-48.5 996,-54.5 996,-60.5 996,-60.5 996,-83.5 996,-83.5 996,-89.5 990,-95.5 984,-95.5"/>
+<text text-anchor="middle" x="905.5" y="-81.9" font-family="Helvetica,sans-Serif" font-size="12.00">Verifier</text>
+<text text-anchor="middle" x="905.5" y="-68.9" font-family="Helvetica,sans-Serif" font-size="12.00">four mechanical strategies,</text>
+<text text-anchor="middle" x="905.5" y="-55.9" font-family="Helvetica,sans-Serif" font-size="12.00">no model in the loop</text>
 </g>
 <!-- model&#45;&gt;verifier -->
 <g id="edge4" class="edge">
 <title>model&#45;&gt;verifier</title>
-<path fill="none" stroke="black" d="M608.13,-72C637.15,-72 668.55,-72 696.63,-72"/>
-<polygon fill="black" stroke="black" points="696.81,-75.5 706.81,-72 696.81,-68.5 696.81,-75.5"/>
-<text text-anchor="middle" x="657.5" y="-87.2" font-family="Helvetica,sans-Serif" font-size="11.00">answer</text>
-<text text-anchor="middle" x="657.5" y="-75.2" font-family="Helvetica,sans-Serif" font-size="11.00">+ pointer IDs</text>
+<path fill="none" stroke="black" d="M641.1,-72C692.45,-72 753.92,-72 804.56,-72"/>
+<polygon fill="black" stroke="black" points="804.87,-75.5 814.87,-72 804.87,-68.5 804.87,-75.5"/>
+<text text-anchor="middle" x="728" y="-75.2" font-family="Helvetica,sans-Serif" font-size="11.00">answer with [E1] [E2] tags</text>
 </g>
-<!-- label_strict -->
+<!-- label_grounded -->
 <g id="node6" class="node">
-<title>label_strict</title>
-<path fill="#28a745" stroke="black" d="M1023.5,-144C1023.5,-144 913.5,-144 913.5,-144 907.5,-144 901.5,-138 901.5,-132 901.5,-132 901.5,-120 901.5,-120 901.5,-114 907.5,-108 913.5,-108 913.5,-108 1023.5,-108 1023.5,-108 1029.5,-108 1035.5,-114 1035.5,-120 1035.5,-120 1035.5,-132 1035.5,-132 1035.5,-138 1029.5,-144 1023.5,-144"/>
-<text text-anchor="middle" x="968.5" y="-129.4" font-family="Helvetica,sans-Serif" font-size="12.00" fill="white">STRICT</text>
-<text text-anchor="middle" x="968.5" y="-116.4" font-family="Helvetica,sans-Serif" font-size="12.00" fill="white">every claim verifies</text>
+<title>label_grounded</title>
+<path fill="#28a745" stroke="black" d="M1158.5,-144C1158.5,-144 1048.5,-144 1048.5,-144 1042.5,-144 1036.5,-138 1036.5,-132 1036.5,-132 1036.5,-120 1036.5,-120 1036.5,-114 1042.5,-108 1048.5,-108 1048.5,-108 1158.5,-108 1158.5,-108 1164.5,-108 1170.5,-114 1170.5,-120 1170.5,-120 1170.5,-132 1170.5,-132 1170.5,-138 1164.5,-144 1158.5,-144"/>
+<text text-anchor="middle" x="1103.5" y="-129.4" font-family="Helvetica,sans-Serif" font-size="12.00" fill="white">grounded</text>
+<text text-anchor="middle" x="1103.5" y="-116.4" font-family="Helvetica,sans-Serif" font-size="12.00" fill="white">every claim verifies</text>
 </g>
-<!-- verifier&#45;&gt;label_strict -->
+<!-- verifier&#45;&gt;label_grounded -->
 <g id="edge5" class="edge">
-<title>verifier&#45;&gt;label_strict</title>
-<path fill="none" stroke="black" d="M861.07,-94.51C872.92,-98.01 885.11,-101.62 896.83,-105.09"/>
-<polygon fill="black" stroke="black" points="895.96,-108.48 906.54,-107.96 897.95,-101.77 895.96,-108.48"/>
+<title>verifier&#45;&gt;label_grounded</title>
+<path fill="none" stroke="black" d="M991.9,-95.53C1003.75,-98.79 1015.84,-102.13 1027.48,-105.33"/>
+<polygon fill="black" stroke="black" points="1026.56,-108.71 1037.13,-107.99 1028.42,-101.96 1026.56,-108.71"/>
 </g>
-<!-- label_hybrid -->
+<!-- label_partial -->
 <g id="node7" class="node">
-<title>label_hybrid</title>
-<path fill="#ffc107" stroke="black" d="M1037,-90C1037,-90 900,-90 900,-90 894,-90 888,-84 888,-78 888,-78 888,-66 888,-66 888,-60 894,-54 900,-54 900,-54 1037,-54 1037,-54 1043,-54 1049,-60 1049,-66 1049,-66 1049,-78 1049,-78 1049,-84 1043,-90 1037,-90"/>
-<text text-anchor="middle" x="968.5" y="-75.4" font-family="Helvetica,sans-Serif" font-size="12.00">HYBRID</text>
-<text text-anchor="middle" x="968.5" y="-62.4" font-family="Helvetica,sans-Serif" font-size="12.00">some verify, some don&#39;t</text>
+<title>label_partial</title>
+<path fill="#ffc107" stroke="black" d="M1172,-90C1172,-90 1035,-90 1035,-90 1029,-90 1023,-84 1023,-78 1023,-78 1023,-66 1023,-66 1023,-60 1029,-54 1035,-54 1035,-54 1172,-54 1172,-54 1178,-54 1184,-60 1184,-66 1184,-66 1184,-78 1184,-78 1184,-84 1178,-90 1172,-90"/>
+<text text-anchor="middle" x="1103.5" y="-75.4" font-family="Helvetica,sans-Serif" font-size="12.00">partly grounded</text>
+<text text-anchor="middle" x="1103.5" y="-62.4" font-family="Helvetica,sans-Serif" font-size="12.00">some verify, some don&#39;t</text>
 </g>
-<!-- verifier&#45;&gt;label_hybrid -->
+<!-- verifier&#45;&gt;label_partial -->
 <g id="edge6" class="edge">
-<title>verifier&#45;&gt;label_hybrid</title>
-<path fill="none" stroke="black" d="M861.07,-72C866.44,-72 871.87,-72 877.31,-72"/>
-<polygon fill="black" stroke="black" points="877.72,-75.5 887.72,-72 877.72,-68.5 877.72,-75.5"/>
+<title>verifier&#45;&gt;label_partial</title>
+<path fill="none" stroke="black" d="M996.18,-72C1001.59,-72 1007.03,-72 1012.43,-72"/>
+<polygon fill="black" stroke="black" points="1012.77,-75.5 1022.77,-72 1012.77,-68.5 1012.77,-75.5"/>
 </g>
 <!-- label_ungrounded -->
 <g id="node8" class="node">
 <title>label_ungrounded</title>
-<path fill="#dc3545" stroke="black" d="M1009,-36C1009,-36 928,-36 928,-36 922,-36 916,-30 916,-24 916,-24 916,-12 916,-12 916,-6 922,0 928,0 928,0 1009,0 1009,0 1015,0 1021,-6 1021,-12 1021,-12 1021,-24 1021,-24 1021,-30 1015,-36 1009,-36"/>
-<text text-anchor="middle" x="968.5" y="-21.4" font-family="Helvetica,sans-Serif" font-size="12.00" fill="white">UNGROUNDED</text>
-<text text-anchor="middle" x="968.5" y="-8.4" font-family="Helvetica,sans-Serif" font-size="12.00" fill="white">none verify</text>
+<path fill="#dc3545" stroke="black" d="M1140,-36C1140,-36 1067,-36 1067,-36 1061,-36 1055,-30 1055,-24 1055,-24 1055,-12 1055,-12 1055,-6 1061,0 1067,0 1067,0 1140,0 1140,0 1146,0 1152,-6 1152,-12 1152,-12 1152,-24 1152,-24 1152,-30 1146,-36 1140,-36"/>
+<text text-anchor="middle" x="1103.5" y="-21.4" font-family="Helvetica,sans-Serif" font-size="12.00" fill="white">not grounded</text>
+<text text-anchor="middle" x="1103.5" y="-8.4" font-family="Helvetica,sans-Serif" font-size="12.00" fill="white">none verify</text>
 </g>
 <!-- verifier&#45;&gt;label_ungrounded -->
 <g id="edge7" class="edge">
 <title>verifier&#45;&gt;label_ungrounded</title>
-<path fill="none" stroke="black" d="M861.07,-49.49C876.09,-45.05 891.68,-40.44 906.14,-36.16"/>
-<polygon fill="black" stroke="black" points="907.35,-39.45 915.94,-33.25 905.36,-32.74 907.35,-39.45"/>
+<path fill="none" stroke="black" d="M991.9,-48.47C1009.75,-43.55 1028.18,-38.48 1044.74,-33.91"/>
+<polygon fill="black" stroke="black" points="1046.08,-37.17 1054.79,-31.14 1044.22,-30.43 1046.08,-37.17"/>
 </g>
 <!-- audit -->
 <g id="node9" class="node">
 <title>audit</title>
-<polygon fill="#fde7e9" stroke="black" points="1274,-95.5 1076,-95.5 1076,-48.5 1280,-48.5 1280,-89.5 1274,-95.5"/>
-<polyline fill="none" stroke="black" points="1274,-95.5 1274,-89.5 "/>
-<polyline fill="none" stroke="black" points="1280,-89.5 1274,-89.5 "/>
-<text text-anchor="middle" x="1178" y="-81.9" font-family="Helvetica,sans-Serif" font-size="12.00">audit chain</text>
-<text text-anchor="middle" x="1178" y="-68.9" font-family="Helvetica,sans-Serif" font-size="12.00">+ providence cache</text>
-<text text-anchor="middle" x="1178" y="-55.9" font-family="Helvetica,sans-Serif" font-size="12.00">(8&#45;dim key, governance&#45;aware)</text>
+<polygon fill="#fde7e9" stroke="black" points="1375,-95.5 1211,-95.5 1211,-48.5 1381,-48.5 1381,-89.5 1375,-95.5"/>
+<polyline fill="none" stroke="black" points="1375,-95.5 1375,-89.5 "/>
+<polyline fill="none" stroke="black" points="1381,-89.5 1375,-89.5 "/>
+<text text-anchor="middle" x="1296" y="-81.9" font-family="Helvetica,sans-Serif" font-size="12.00">Audit chain</text>
+<text text-anchor="middle" x="1296" y="-68.9" font-family="Helvetica,sans-Serif" font-size="12.00">+ answer cache</text>
+<text text-anchor="middle" x="1296" y="-55.9" font-family="Helvetica,sans-Serif" font-size="12.00">(reproducible, replayable)</text>
 </g>
-<!-- label_strict&#45;&gt;audit -->
+<!-- label_grounded&#45;&gt;audit -->
 <g id="edge8" class="edge">
-<title>label_strict&#45;&gt;audit</title>
-<path fill="none" stroke="#888888" stroke-dasharray="5,2" d="M1035.72,-108.77C1048.7,-105.39 1062.59,-101.77 1076.4,-98.18"/>
-<polygon fill="#888888" stroke="#888888" points="1077.63,-101.48 1086.43,-95.57 1075.87,-94.7 1077.63,-101.48"/>
+<title>label_grounded&#45;&gt;audit</title>
+<path fill="none" stroke="#888888" stroke-dasharray="5,2" d="M1168.2,-107.94C1179.18,-104.83 1190.8,-101.54 1202.37,-98.26"/>
+<polygon fill="#888888" stroke="#888888" points="1203.39,-101.61 1212.05,-95.51 1201.48,-94.87 1203.39,-101.61"/>
 </g>
-<!-- label_hybrid&#45;&gt;audit -->
+<!-- label_partial&#45;&gt;audit -->
 <g id="edge9" class="edge">
-<title>label_hybrid&#45;&gt;audit</title>
-<path fill="none" stroke="#888888" stroke-dasharray="5,2" d="M1049.13,-72C1054.63,-72 1060.23,-72 1065.85,-72"/>
-<polygon fill="#888888" stroke="#888888" points="1065.99,-75.5 1075.99,-72 1065.99,-68.5 1065.99,-75.5"/>
+<title>label_partial&#45;&gt;audit</title>
+<path fill="none" stroke="#888888" stroke-dasharray="5,2" d="M1184.19,-72C1189.53,-72 1194.94,-72 1200.35,-72"/>
+<polygon fill="#888888" stroke="#888888" points="1200.72,-75.5 1210.72,-72 1200.72,-68.5 1200.72,-75.5"/>
 </g>
 <!-- label_ungrounded&#45;&gt;audit -->
 <g id="edge10" class="edge">
 <title>label_ungrounded&#45;&gt;audit</title>
-<path fill="none" stroke="#888888" stroke-dasharray="5,2" d="M1021.06,-31.42C1038,-35.83 1057.45,-40.89 1076.75,-45.91"/>
-<polygon fill="#888888" stroke="#888888" points="1075.94,-49.32 1086.5,-48.45 1077.7,-42.54 1075.94,-49.32"/>
+<path fill="none" stroke="#888888" stroke-dasharray="5,2" d="M1152.08,-31.49C1167.19,-35.77 1184.45,-40.66 1201.63,-45.53"/>
+<polygon fill="#888888" stroke="#888888" points="1201.15,-49.03 1211.73,-48.39 1203.06,-42.3 1201.15,-49.03"/>
 </g>
 </g>
 </svg>