PRISM v2.0.1

PRISM v2.0.1 — release-hygiene patch

A focused patch pass on v2.0.0 addressing the converged ChatGPT 5.5 Pro review. No architectural changes. No reopening of the v2 design. The patches are concentrated in contract precision, prompt-output hygiene, schema fidelity, and adding what was missing from the framework body itself: discoverability of the project, license, feedback channels, and update path.

Substantive changes

Singleton Lens story made consistent (P0-1)

v2.0.0 embedded the Lens Library in Appendix G but the front matter, §1.2, and §7.1 still required a separate Lens attachment as a normal default. v2.0.1 makes embedded Lens the default; standalone is an optional override only when the project explicitly pins to a newer Library version than the embedded copy. A singleton PRISM.md attachment is now sufficient for normal operation.

Prompt hash → Prompt digest (P0-3)

Prompt hash asked vendor chats to compute a SHA-256 prefix of the prompt body. They can't, in general — and a model may invent a plausible-looking hash. v2.0.1 replaces Prompt hash with Prompt digest: orchestration generates a short identifier and writes it into the Envelope; execution copies it verbatim into the Output; orchestration verifies copy-through, not cryptographic computation. Catches wrong-prompt / wrong-attachment delivery without depending on vendor hash computation.

Verification status restored as explicit axis (P0-5)

v1.x's [UNVERIFIED] tagging was folded into low-confidence in v2.0. That conflated two orthogonal axes: a high-confidence inference may be unverified (no source confirmed it directly), and a verified finding may be low-confidence because the source itself is ambiguous. v2.0.1 restores the separation explicitly. The §4.11 finding structure now requires:

• Verification status: [verified | partially verified | unverified | not found]
• Null results: (surfaces checked where nothing material was found; optional)
• Could not verify: (claims, sources, attachments, or tools unavailable/inconclusive; optional)

…alongside the existing Confidence: axis.

last_verified gating via verification_basis (P0-4)

Lens v0.10 introduces the verification_basis: field. PRISM §7.4 freshness logic now gates on it: a schema-introduction-only basis produces unverified-anchor disposition regardless of date age, triggering point refresh before depth use. Closes the silent false-fresh bypass.

Master emission half-step (P1-2)

v2.0.0 emitted a fresh Master at every turn-close regardless of whether material state changed — heavyweight on mobile but operationally safe (operator never picks the wrong Master because the latest is always the most recent emission). v2.0.1 preserves the always-emit safety property and adds a _no_change filename suffix marker for non-content-bearing turns. Operator can defer cloud-save and attachment-swap on _no_change emissions.

Smaller hygiene patches

• Eleven Monitors → Twelve Monitor slots (P0-2). M1–M12 now stated correctly throughout.
• Success criteria block added to §12.1 atomic prompt template (P0-6). Names what a good execution output must accomplish.
• §17 E.5 → MO-5 (P1-3). Mobile-guide subsections renamed to MO-1 through MO-5 to disambiguate from Appendix E.5 What's next template.

New surfaces

§18 Project, feedback, updates

The framework file no longer hides the project from anyone reading it as a singleton attachment. §18 carries:

• Project identity: repository URL, maintainer, license summary (CC BY 4.0 docs / MIT code / CC BY-SA 4.0 CoC), version, release date.
• Resource fetch convention: stable raw URLs for the framework, Lens, releases page, and version stamps. New resources follow the same path pattern.
• §18.3 Currency check at session open: when the substrate supports web access, orchestration may GET the VERSION endpoints and surface a soft flag if newer versions are available. Opportunistic, not mandatory.
• Feedback channels: bugs → Issues; ideas/observations → Discussions Ideas; show-and-tell → Discussions Show-and-tell; questions → Discussions Q&A; security → private vulnerability reporting.
• Citation snippet: short attribution suitable for inline use; full BibTeX-equivalent in CITATION.cff.

VERSION files

VERSION at repo root and lens/VERSION are single-line files that adopting tooling can fetch cheaply to check current published versions without parsing the full framework or Lens body.

What's not in v2.0.1

• Vendor Selection cache (P1-1): deferred to backlog. Reasonable refinement; not load-bearing.
• §9 Setup posture (lean | standard | deep): rejected. v1 Scope Flags by another name; v2's Setup quality system is intact.
• Acceptance harness: kept as a working artifact in the workshop, not folded into the framework body. Lighter-touch than codifying it in the doc.
• Cowork/Computer-Use automation, Claude Plugins shortlists, full domain-practitioner Lens accretion: belong in backlog/futures, per the review's own guidance.

Calibration

Still rev. 1 draft thresholds on M5 bands, Update session triggers, and probe iteration ceilings. v2.0.1 is dogfood-worthy; first real v2 audit is the post-release shakedown.

Files in this release

• PRISM.md — canonical (always-current).
• PRISM_v2_0_1.md — version-pinned snapshot, byte-identical to canonical at this tag.
• lens/PRISM_lens_library.md — Lens v0.10 (canonical).
• lens/PRISM_lens_library_v0_10.md — Lens v0.10 snapshot.
• VERSION — 2.0.1.
• lens/VERSION — 0.10.
• SKILL.md, CITATION.cff, README.md — refreshed to v2.0.1.

Prior MINOR snapshots (PRISM_v2_0_0.md, lens/PRISM_lens_library_v0_9.md) are removed from main per RELEASING.md retention; recoverable via tags v2.0.0 and prism-lens-v0.9 respectively.

Provenance

• Converged review: ChatGPT 5.5 Pro, 2026-04-25 (workshop synthesis/ — private repo, available on request).
• Patches qualified by maintainer per workshop design/PRISM_v2_0_1_patch_spec.md.

Companion release

PRISM Lens Library v0.10 ships in the same commit as this framework release; the embedded Appendix G mirrors the standalone Lens byte-for-byte at this tag.

PRISM v2.0.0

PRISM v2.0.0 — first release on the v2 line.

v2 is a major rebuild around the Lens Library (now embedded as Appendix G of PRISM.md), a triple execution contract (Envelope · Self-check · Output), continuous Master + What's next state, and a telemetric context-pressure framework. Per the design document, v2 is not a patch on v1.10.4; v1.10.4 stays terminal on the v1.x line and remains in this repo for projects pinned there.

Headline changes from v1.10.4

• Two session types named explicitly. Orchestration (Claude) owns Master state, Monitor fires, Setup probes, convergence reasoning. Execution (any vendor) owns single-shot work. Crossing the boundary is the triple contract.
• Triple contract — Envelope · Self-check · Output. Vendor-agnostic by construction; survives future automation modes unchanged. Self-check operationalizes substrate verification (SP-13) inside execution sessions.
• Lens Library v0.9 is required and embedded. PRISM.md singleton includes the v0.9 catalog as Appendix G — one file is enough to run an orchestration session. The standalone lens/PRISM_lens_library.md remains canonical for catalog evolution (Update sessions, version bumps).
• Continuous-state mechanic. Master and What's next written at every orchestration turn-close, regardless of band. Misreads of context pressure are recoverable by construction.
• Telemetric M5. Seven signals (volumetric + behavioral), four bands (🟢🟡🟠🔴), qualitative compounding. Old M5 (Attachment Pressure) and old M12 (Conversation Pressure) both absorbed.
• Migration handoff as a defined artifact at 🔴 (mandatory) and 🟠 (operator-elective).
• Vendor Selection at every dispatch with live web-search currency check. SP-10 promoted to a named Standing Principle.
• Setup is iterative, not waterfall. P0→P1 boundary clears on three-layer readiness: Structural completeness + Library coverage saturation + Operator ratification.

Probes (seven, Setup-only)

Probe	Status	Notes
P1 Coverage grading	🆕	Library-driven; tri-state-with-maybe disposition per Lens
P2 Adversarial Scope	reshape	Setup-only now; Vendor Triangulation lives separately at convergence
P3 Decision Framing	new explicit	Outputs Decision brief + Stakeholder register
P4 Pre-mortem	🆕	Imagines execution complete; surfaces failure modes
P5 Falsifier	new explicit	Falsifier criteria for Layer 2 synthesis
P6 Domain Reconnaissance	reshape	From v1.x Discovery role; outputs Jurisdiction map
P7 User Voice	promoted	From v1.x Phase 2 enrichment role; informs strategy before execution

Monitors (M1–M12, M12 slot reused)

• M5 absorbs old M5 + old M12 (Conversation Pressure retired).
• M7 renamed Claim Conflict (was Assumption Conflict; v2 has no Assumption Register).
• M12 slot reused for Result Completeness Check — within-domain coverage gap detection at convergence (v1.x Coverage role lineage).
• M8 surface narrowed to evidence inside the audit; framework-anchor staleness handled by point refresh.

Standing Principles

• New: SP-12 Bounded-Search Disclosure, SP-13 Substrate Declaration, SP-14 Filename Discipline (extracted from SP-8).
• Extended: SP-1 (covers offers to reconstruct), SP-10 (named principle).
• Narrowed: SP-8 (Canonical Authority only).
• Dissolved: SP-3 (incompatible with orchestration/execution split).

v1.x constructs dissolved or relocated

• All five Scope Flags (SF1–SF5) dissolved.
• Runtime Profile dissolved into telemetric framework.
• GATE-0/1/2 folded into Monitors + What's next + M11.
• Adaptations lighter — strategy revision at convergence + Master version bump.
• Discrepancy Check absorbed into M7 + Layer-1 reconciliation.

See Appendix D in PRISM.md for the full surface-drift map.

Repository changes in this release

• PRISM.md is the singleton: framework body + skill frontmatter + Lens Library v0.9 embedded as Appendix G.
• PRISM_v2_0_0.md — version-pinned snapshot of PRISM.md at this tag (for git-tag recovery).
• PRISM_v1_10_4.md retained at root for projects pinned to v1.10.4.
• SKILL.md — standalone skill-loader (frontmatter only) for operators who prefer the decoupled loader / body layout.
• design/ added — v2 specification (PRISM_v2_spec_rev2.md) + design document (PRISM_v2_rev8_design_document.md) as provenance for the framework as shipped.
• README.md updated.

Beta posture and empirical calibration

v2.0 ships with several thresholds at rev. 1 draft estimates. Calibration items for post-release real-use signal:

• M5 band thresholds (volumetric values).
• Probe iteration ceilings (current floor 2, soft ceiling 4).
• Probe 1 fires-maybe operator-fatigue.
• Update session trigger threshold (currently 3 stale-pattern accumulations across 6+ months).
• Multi-vendor Self-check verification (Claude Opus 4.6/4.7 and Sonnet 4.6 verified; Gemini, ChatGPT, Perplexity behavior is report-worthy).

Open Discussions thread for feedback, divergences, and adaptation reports: https://github.com/Ronkupper/PRISM/discussions

How to use

1. Grab PRISM.md from this repo. Singleton: framework + Lens Library v0.9 (embedded as Appendix G) + skill frontmatter.
2. Attach to a Claude session, or install as a Claude Skill.
3. Hand Claude your subject and goal.

Optional / advanced: install SKILL.md (standalone loader) and attach PRISM.md separately if you prefer the decoupled layout. Pin to PRISM_v2_0_0.md if you want the version visible in the filename.

Substrate: Claude Opus 4.6 / 4.7 in orchestration; multi-vendor on the execution side.

PRISM Lens Library v0.10

PRISM Lens Library v0.10 — schema-fidelity bump

v0.10 is a schema-fidelity bump on top of v0.9: same 23 lenses, same content, same triggers. The change is structural — the addition of a verification_basis: field on the two rubric-anchored entries — and it exists to fix a real logic bug surfaced by the converged ChatGPT review of v2.0.0.

What changed

• verification_basis: field on entries carrying last_verified:. One of {schema-introduction-only, independent-review}.
• LL-D-002 (WCAG 2.2) and LL-D-005 (OWASP ASVS 5.0.0) now both carry verification_basis: schema-introduction-only — making explicit that their last_verified: 2026-04-24 reflects schema-introduction date, not an independently performed currency check.
• Limitation #4 reframed to reference the new field.
• Schema documentation updated to describe the field's role in adopting frameworks' freshness logic.

Why this matters

Adopting frameworks (PRISM included) compute lens-anchor freshness from last_verified:. Without verification_basis:, an anchor's schema-introduction date silently passed the freshness check on the strength of date alone — a false-fresh result that suppressed the currency check that should have fired. v0.10 closes that gap. PRISM v2.0.1's §7.4 now gates freshness on verification_basis:: a schema-introduction-only basis triggers unverified-anchor disposition regardless of date age.

The field flips to independent-review after a real currency review is performed.

Calibration status

Still pre-release. Awaiting at least one real-world calibration application before promotion to v1.0 stable.

Files

• lens/PRISM_lens_library.md — canonical (always-current).
• lens/PRISM_lens_library_v0_10.md — version-pinned snapshot, byte-identical at this tag.
• lens/VERSION — single-line version stamp (0.10) for cheap currency-check fetches per PRISM v2.0.1 §18.

Compatibility

PRISM v2.0.1 embeds Lens v0.10 in Appendix G as the default Library source. Operators on a project pinned to standalone v0.10 attach it explicitly to override Appendix G.

Provenance

Surfaced by the converged ChatGPT 5.5 Pro review of PRISM v2.0.0 (P0-4: last_verified semantics can suppress the needed currency check). Patch shape qualified by maintainer to use the lighter-touch verification_basis field rather than nulling last_verified outright.

PRISM Lens Library v0.9 (pre-release)

First public release of the PRISM Lens Library — a standalone, framework-neutral reference catalog of 23 audit-scope lenses (5 universal + 18 domain across 6 packs). Each entry names a specific class of silent omission an audit can plausibly miss, for use as a coverage map at scope-definition time.

Status

v0.9 pre-release. Awaiting at least one real-world calibration application before promotion to v1.0 stable. Integration into PRISM itself is planned for v2 (PX-SI-01 Phase B) as a separate effort; this release does not bump PRISM's own version (still v1.10.4).

Files

• lens/PRISM_lens_library.md — stable filename
• lens/PRISM_lens_library_v0_9.md — versioned copy, byte-identical

Known-limitations section and full schema inside the artifact. Feedback and field-observations welcome.

v1.10.4 — Initial public release

Initial public release of PRISM — a framework for keeping multi-prompt, multi-session LLM research and audit work coherent across prompts, sessions, and the context limits of any single chat.

What ships in v1.10.4

• PRISM.md — the full framework (Monitors, Gates, Standing Principles, two-layer convergence, version management, and the What's next? entry point)
• PRISM_backlog.md — roadmap: active proposals, deferred items, and declined ideas with rationale
• PRISM_teaser.pdf in /assets — 10-slide overview if you'd rather skim before reading the framework

Quick start

Grab PRISM.md, attach it to a Claude session, tell Claude the subject and goal of your project, send What's next?. That's the entry point — it also works to resume any existing project mid-stream.

The framework runs on any capable LLM (Claude, Gemini, Perplexity, ChatGPT) and is designed around deliberate multi-vendor triangulation rather than single-vendor lock-in.

Origin

Distilled from real competitive research and product audit projects across 2026, with iteration driven by dogfooding the framework on its own evolution.