PRISM v2.21.0

v2.21.0 — external-review hardening (methodology lane) + runtime front door (MINOR over v2.20.2).

Methodology-lane refinements from two external reviews, triaged against the repo and ratified, plus a runtime front door. Framework prose changes this release (unlike the v2.20.x packaging patches). Lens Library stays v0.16; lint catalog stays v5.

What changed

• Turn-close ordering (G1). The per-turn Monitors are evaluated before the continuous-state artifacts (band header, Master, What's next) are emitted; the evaluation is internal — no visible per-turn transcript.
• Self-check visible-recompute (G3). Step 6 now writes each recompute out in full (operands + operation, in the Appendix K form) before comparing, rather than asserting "recomputed/matches".
• Self-check first-token echo (G5). Step 0 emits the matched Dispatch ID as the first token on a clean gate (the one-line truncation report stays first on a failed/truncated gate), so a silently-skipped transport gate is now observable in the response itself. Refuse-all was dropped.
• Dispatch record vs execution seat paste (N3). A clean split between the orchestration dispatch record (never pasted whole to a vendor) and the self-contained execution seat paste — new template E.1a, the Envelope minus the triangulation fields with Vendor: resolved to the single executing seat. The single-arm strip rule now explicitly retains the single Vendor: / Vendor config:.
• ━-vs-XML rationale (G4). Documents why PRISM keeps ━━━ heavy-line dispatch delimiters rather than migrating to XML tags (fence-collision-safe; the vendor-parsing ledger indicts fences, not the heavy line; reliable tag-extraction is a vendor dependency the dispatch surface avoids). A new Appendix H section + a calibration item queue a real-device mobile-paste test. No format change.
• J.3 clarifier (G2). The "no trailing-newline normalization" pin is scoped as a re-stage rule only — it has no bearing on return verification (which reconciles by Dispatch-ID instance match + Prompt-digest copy-through, never a rehash of clipboard bytes).
• Runtime front door (C). A new /prism-help command and an orientation-before-engagement core behavior: an opening "how do I use this?" is answered with a 30-second orientation and an offer to begin — before the missing-input halts — instead of a framework dump.

Calibration / report-back

• Item 5 — multi-vendor Self-check. Does a clean Step 0 emit the matched Dispatch ID as its first token on Gemini / ChatGPT / Perplexity? (Claude-family verified.)
• Item 10 — ━ transport survival. Does the ━━━ heavy-line glyph survive mobile-app paste on ChatGPT / Claude / Perplexity / Gemini? (Seeds the new Appendix H rows.)

Install

• Marketplace (Claude / Cowork / Claude Code) — see the README Quick start; installed plugins update to 2.21.0.
• Upload-plugin (no marketplace) — download PRISM-plugin-2.21.0.zip below → Customize → Plugins → Personal → + → Upload plugin.
• Single file — attach PRISM.md (or the version-pinned PRISM_v2_21_0.md).

Surface-drift map: Appendix D. Version-pinned snapshot at this tag: PRISM_v2_21_0.md (byte-identical to PRISM.md).

PRISM v2.20.2

PATCH over v2.20.1 — packaging + front-door hardening. No framework mechanic changes: PRISM.md is byte-stable apart from version strings, the Lens Library stays v0.16, and the lint catalog advances to v5. This release acts on two external reviews (a file-aware ChatGPT pass and a Gemini methodology pass).

Packaging / loader fixes

• Root SKILL.md: malformed YAML description (unquoted, with embedded colons) → block scalar; stale PRISM_v2_18_0.md reference → the current snapshot.
• /prism-close command description colon → em-dash (the frontmatter now parses); all six command argument-hints quoted (they parsed as YAML lists, which render garbled in autocomplete).
• Skill core SKILL.md: bare §7.2 → the named reference §{section.lens-schema-what-orchestration-consumes} (a cross-file lint error that had been shipping).

A real lint gate (lint-v2). CI previously validated only PRISM.md named-refs, and silently skipped frontmatter when a private schema URL was unreachable — so the fixes above could ship "lint clean." Catalog v5 now runs on push, tags, and PRs, wires the previously-unrun cross-file linter over the Skill archive, and adds a package-integrity linter (PRISM-LINT-10–13): every shipped SKILL.md/command frontmatter parses, command field types are correct, plugin.json / marketplace.json are valid JSON with versions consistent against VERSION, and the built plugin ZIP is asserted clean.

Front door (README). An action-first Start here block: a "choose your path" table, quick brief vs full audit, a follow-the-card loop, and pointers for resuming and for repo-backed (cross-device) operation.

Install / update: marketplace.json is at 2.20.2, so the Claude desktop Update button offers it; on Claude Code, /plugin install prism@prism-marketplace. Non-marketplace path: download PRISM-plugin-2.20.2.zip below, then Customize → Plugins → Personal → + → Upload plugin.

Provenance: version-pinned snapshot PRISM_v2_20_2.md, byte-identical to PRISM.md at the tag. Build: apply_v2_20_2_bump.py version sweep + project_skill_archive.py re-projection. Lint clean at the scope the gate now actually covers — named-refs on PRISM.md, cross-file on the Skill archive, and package-integrity (0 error-severity findings).

PRISM v2.20.1

PATCH over v2.20.0 — plugin command-surface revision. No framework mechanic changes: PRISM.md is byte-stable apart from version strings, the Lens Library stays at v0.16, and the lint catalog stays at v4. The projected Skill core is a clean re-projection.

The Skill's slash commands now track the Setup → rounds → Closure lifecycle as a coherent set:

Command	Role
/prism-start <subject>	Begin — Setup, probes P1–P7 (the lifecycle entry point)
/prism-whats-next	The PRISM Desk — resume, re-sync from the repo, the one next action, and stage the next dispatch as a paste-ready Envelope
/prism-converge	The dispatch return leg — transport-integrity check (Dispatch ID), Layer-1 absorption + reconciliation, write the convergence delta + What's next
/prism-status	The STATE view — dependency / critical-path map + progress timeline, side by side, from verified repo state
/prism-close	The closure gate — three-layer close sweep + deliverables (report + optional interactive workbook)
/prism-meta	The methodology (contributor) lane

The three new commands (converge, status, close) complete the loop that the prior start / whats-next pair only started. The natural-language form ("converge these returns", "what's next", "close this engagement") remains equivalent, and is the portable path on non-Claude vendors.

Install / update: marketplace.json is bumped to 2.20.1, so the Claude desktop Update button offers the new commands; on Claude Code, /plugin install prism@prism-marketplace picks them up.

Provenance: the version-pinned snapshot at this tag is PRISM_v2_20_1.md (byte-identical to PRISM.md at the v2.20.1 tag). Build is the workshop decompose leg — apply_v2_20_1_bump.py (version-string bump, assert-then-replace, fail-loud) plus the projection pin. Lint clean (0 error-severity, named-refs on PRISM.md and cross-file on the projected tree).

PRISM v2.20.0

MINOR over v2.19.0. A re-embed release: it carries the new Lens Library v0.16 into the single-file form (Appendix G) and names its new lens in the Independent Validation Dispatch default kit. The only framework behavior change is additive; no construct renumbering, no operating-rule change.

Paired Lens release: https://github.com/Ronkupper/PRISM/releases/tag/prism-lens-v0.16

What's new

Reader-respect lens — LL-D-022 "Does this respect the reader?" (Lens Library v0.16, Pack 1). A reader-respect pass that flags filler which spends the reader's attention and returns nothing — empty preambles, the prompt restated back, contentless transitions, content-free hedges. It completes the document-review family alongside LL-D-019 ("Who said otherwise?"), LL-D-020 ("Help or ammunition?"), and LL-D-021 ("Does a stranger follow?").

Named in the §4.14 IVD default kit. The Independent Validation Dispatch's default lens kit (and the Appendix E.13 validation-dispatch envelope) now run LL-D-022 by default — the full reader-respect family is the IVD's standing document-quality gate. (This is the one additive framework behavior change in v2.20.0.)

recommended_sources: curation. The two source-bearing lenses — LL-D-008 ("Compared to what?") and LL-D-009 ("Does it pay back?") — gain two sources each, every one shipped with its mandatory bias-and-handling framing: and a recency: posture: Failory (a founder-interview / post-mortem corpus) and Start-Up Nation Central — Finder (a deep Israeli-tech-ecosystem structured source, framed as depth-by-ecosystem — globally material in Israel-led categories such as cybersecurity). The existing pitch-deck-libraries source now names bestpitchdeck.com among its open examples.

Housekeeping folded in

• The Supersedes header paragraph (stale — still described v2.18.0) is rewritten for v2.20.0.
• The Scope-Integrity-Test generic-gate count and the README lens count (both stale at 23) and the glossary domain count are reconciled to the live catalog (27 total / 22 domain).
• The single-file SKILL.md version-pinned snapshot pointer (stale at PRISM_v2_18_0.md) is updated to PRISM_v2_20_0.md.

Verification

Lint clean — named-refs on PRISM.md 0 errors, cross-file on the projected Skill tree 0 errors. The version-pinned snapshot at this tag is PRISM_v2_20_0.md (byte-identical to PRISM.md). Re-embed and version bumps were applied by deterministic, fail-loud transforms; the Skill core + bundles were re-projected idempotently. See RELEASING.md.

PRISM v2.19.0 — Core-slimming refactor

MINOR, additive (no construct renumbering). Slice 5 — the terminal slice of
the dependency-ordered sequence: a projection-only, behavior-preserving
packaging refactor (Decision T). PRISM.md stays the canonical monolith; only the
projection and the load contract change.

Surface

• Core-slimming decouple (§A2 / Decision T). The always-loaded core sheds its
  phase-, seam-, and condition-scoped mechanics into four new on-demand
  bundles keyed to the role that needs them, leaving the execution-hot spine
  resident. Core ~24% leaner (5092 → 3895 lines).
  • reference/setup.md — Setup mechanics: the seven probes (P1–P7),
    three-layer readiness, the Setup-artifact procedures, strategy stability,
    onboarding (§6.1–6.6); Library-reference-at-Setup (§7.1); the Scope-Integrity
    Test + specialist-pass promotion (§7.3–7.4); the Claude-Project-as-Setup
    recommendation (§8.1). Trigger: no Master yet / operator initiates an
    engagement (P0).
  • reference/currency.md — point-refresh (§7.5), the Update session (§7.6),
    the session-open currency check (§18.3). Trigger: an Update session / the
    currency-check-at-open.
  • reference/continuity.md — migration handoff, failsafe recovery,
    defensive migration (§5.4–5.6), missing-handoff recovery (§14). Keeps a
    minimal in-core recovery floor resident (write-at-every-turn-close +
    recover-from-the-repo) so a session in trouble is never stranded. Trigger: a
    resume with a missing handoff, or M5 band 🟠/🔴.
  • reference/corpus-access.md — the corpus-access dispatch (§4.13).
    Trigger: the dispatch-builder classifies a pass as corpus-access.
• Fail-loud load triggers. Each gutted area leaves a small in-core trigger
  stub — a detector + pointer + a visible SP-4 self-check load line ("phase =
  Setup → setup.md ✓") — so operating without a required bundle is detectable,
  not a silent gap. The §3.7.6 phase→bundle manifest and the SKILL.md
  core-load gate route the loads; the v2.16 bundle-load integrity layer is the
  safety net.
• What stays in core: the execution-hot spine — Standing Principles,
  Monitors, the triple contract, the prompt-package engine, the dispatch
  lifecycle, the §6.7 closure gate, and the §7.2 lens schema (the dispatch
  interface).
• One doc-hygiene content correction folded in. The Appendix C tag-count
  summary is reconciled to its list-entry counts (structural 44→50, methodological
  12→18; totals 74 / 81), with a lockstep-invariant note so it can't silently
  re-drift; the shipped reference/provenance.md table now matches.

Behavior-preservation (the bar for a projection-only slice)

• Union of core + bundles is content-equivalent to the prior projection
  (line-accounting verified: core-uncovered + bundles + discarded == body; 0
  source lines lost, +75 generated stub lines). Every new bundle is reachable
  via its in-core trigger + the manifest + the SKILL.md load step; a miss is
  caught by the integrity layer. Projection idempotent. Lint 0 errors (named-refs
  on PRISM.md; cross-file on the projected tree).

Calibration / report back

• Do the in-core trigger stubs + SP-4 self-check reliably load the right bundle at
  the right moment across surfaces (deterministic skill-loader vs model-driven)?
• Does the leaner resident core measurably lower M5 context pressure on long
  Execution sessions?
• Is the in-core recovery floor enough to never strand a resuming session before
  it fetches continuity.md?

Provenance. Lens Library unchanged (v0.15). Snapshot PRISM_v2_19_0.md
(byte-identical at the v2.19.0 tag). The projection script (workshop
scripts/decompose/project_skill_archive.py) was reconciled onto origin's
v2.14.0 base + the 2.15→2.18 deltas (verified byte-for-byte against the committed
v2.18.0 tree) before adding the four phase-bundle regions. Decision T (full
conservative set, Setup-first, recovery-last with an in-core stub) and Decision S
(core-light routing) resolved.

---

Part of the PRISM v2.15.0 → v2.19.0 promotion arc. See RELEASING.md.

PRISM v2.18.0 — Lifecycle & deliverables

MINOR, additive (no construct renumbering). Slice 4 of the dependency-ordered
sequence: closes the engagement lifecycle and codifies the deliverable, building
on the 2.16 lanes/Desk backbone and the 2.17 dispatch round-trip.

Surface

• Engagement closure (new §6.7). The symmetric gate out of an engagement —
  a three-layer close sweep (object / lane-meta / operator) that bookends Setup's
  three-layer readiness, opposite in direction and lighter in machinery (a
  verification sweep, not a probe battery). Layer 1: every pass closed-or-deferred,
  falsifiers checked, the verdict shipped as the comprehensive report, polish only
  on a validated deliverable, shipped == validated. Layer 2: every lane's inbox
  drained (the orphan-sweep, now a gate) and the reconcile-at-close codification
  sweep. Layer 3: explicit operator close ratification + externally share-ready.
• The engagement deliverable, codified (new reference/report-architecture.md,
  Appendix K). The comprehensive final report named as the deliverable of
  record — verdict-first, executive-scannable, all work reflected, tight body +
  curated appendices, a linkable "Created with PRISM" credit, a plain-language
  methodology & disclosures section. The canonical body skeleton, the ~11 craft
  conventions (in-one-breath takeaways, how-to-read orientations, the ●/◐/○
  status-glyph legend with Confidence + Verification as separate axes, visual
  scenario-fencing, visible recompute, a self-decoding glossary), the interactive
  workbook cockpit pattern (editable cells → a live decision gate; SP-18
  tie-back), the presentation house-style (HTML→PDF done well; polish only
  validated; inspired-not-impersonating — a closure convention, not a new SP),
  and the reconcile-at-close checklist.
• Deliverable bump-atomicity (§11.2). Bump atomicity now governs the
  client-facing deliverable: cosmetic + text corrections bump too; the shipped
  revision must equal the validated revision (a post-validation correction
  re-stamps + re-validates-if-material); one current file per revision. The IVD
  (§4.14) validates the complete deliverable. An SP-18 deliverable-transform guard
  (§10.1.9) tokenizes standalone figures only so a clarity edit doesn't trip a
  false figure-drift halt; HTML→PDF render hazards in reference/vendor-parsing.md.
• External share (new reference/external-share.md, Appendix L).
  One-repo-per-engagement, externally share-ready by close; the de-coded share
  archive (logical tree + descriptive filenames over SP-14 + a root Index.html,
  expansive to show the rigor); operator-selectable share modes (present-only
  vs repo-committed · report-in-archive vs separate · 2- vs 4-file send); and the
  invariant image-redaction procedure (visual pass → rebuild from PII-safe
  screens → content-hash-verify → grep→0). The read-the-repo-not-the-mirror
  canonical-source rule ties SP-8 (§10.2.1).
• Operator guide (MO-6). Multi-lane session legibility — rename
  <plain name> | <Lane>-<N>, suffix superseded sessions _Archived;
  name-disambiguation relies on the explicit manual-rename step (a summarizing
  surface drops the moniker), with a flag-fatigue note.

Calibration / report back

• Does the closure gate actually catch loose ends (un-drained inboxes, unchecked
  falsifiers, shipped≠validated) before an engagement is declared done?
• Does reconcile-at-close convert the applied-but-not-codified pattern into a
  standing close-time habit, or does craft still accrete uncodified?
• Is the report-architecture bundle enough for a fresh session to build a
  verdict-first report without re-deriving the craft conventions?

Provenance. Lens Library unchanged (v0.15). Snapshot PRISM_v2_18_0.md
(byte-identical at the v2.18.0 tag). Reference detail in the new Appendix K /
reference/report-architecture.md and Appendix L / reference/external-share.md.
Source specs (Engagements clone): ENGAGEMENT_REPORT_v1 (ER-A/ER-K),
ENGAGEMENT_WORKBOOK_v1, ML-63/72/78/96/108, ML-81/89/91/92, R6/R7. Decisions B
(presentation = closure convention, not an SP) and E (reconcile-at-close =
closure-gate line + report-architecture checklist, not a Monitor/SP) applied;
E was the genuinely-open decision — flagged to the operator for ratification.

---

Part of the PRISM v2.15.0 → v2.19.0 promotion arc. See RELEASING.md.

PRISM v2.17.0 — Dispatch round-trip

MINOR, additive (no construct renumbering). Slice 3 of the dependency-ordered
sequence: makes the dispatch round-trip symmetric and named end-to-end, building
on the 2.16 lanes/Desk backbone.

Surface

• Dispatch lifecycle (new §4.15). The whole round-trip named as one bounded
  construct — five stages (build → dispatch → execute → return/converge →
  reconcile), the dispatch-scoped sibling of the engagement lifecycle. A frame
  that subsumes the late-bound build, the operator and return cards, and the
  failure leg by reference (no renumber). Render-from-live-model named once
  across UI view / vendor pick / dispatch paste; Dispatch-builder and
  Dispatch-consumer roles.
• Late-bound build (§6.2). Layer-1 structural completeness now requires a
  pass SPEC, not a frozen Envelope; the Envelope is rendered just-in-time at
  dispatch, anchored to the spec so it cannot silently re-scope.
• Return + failure leg (§4.4). A proactive return-handling card (default
  attach-and-the-session-handles; converge via the seam-③ clean-context
  sub-agent) and a failure leg routed by type — seat-failure (capability null
  → converge + named gap + ranked re-dispatch, never auto), digest/ID mismatch
  (re-grab before converging), substitution (absorb). Substantive divergence is
  the triangulation delta, not a failure.
• Dispatch ID + pinned digest preimage (§3.2 / §4.7). A unique
  per-dispatch-instance identifier paired with the copy-through digest, so a
  re-dispatch of the same pass (identical body ⇒ identical digest) reconciles to
  the exact instance. Reconciliation keys on the triple **filename + Dispatch ID
  • digest**; the digest preimage is pinned for re-stage reproducibility.
• Transport-integrity bracket (§3.2.5). A top-of-paste PRISM PROMPT INTEGRITY anchor + a terminal END PRISM DISPATCHED PASTE sentinel wrap the
  dispatched paste; Self-check Step 0 confirms the sentinel is present and
  its Dispatch ID matches the anchor (presence + string-match, never a hash
  recompute) before any work — so a paste copied mid-stream, before its tail
  finished writing, is caught at the execution session instead of silently
  running partial. The anchor + sentinel are excluded from the digest preimage;
  the bracket frames the triple contract, it is not a fourth block. (Operator
  refinement; adversarially design-verified — the first design keyed on the
  vendor-generated Output block and would have false-halted every paste; the
  terminal-sentinel form is the corrected mechanism.)
• IVD refinement (§4.14 + SP-15). A fanned Independent Validation Dispatch is
  a coverage fan (recall-merge), not triangulation: adjudicate each catch by
  re-derivation, never an agreement vote; prefer strong closed-document
  reasoners as seats. SP-15 gains the "verify contested first-hand" anchor.
• Passive pre-fill of self-report (§4.13). Passively pre-answer
  externally-knowable self-report items before the interview (provenance-tagged),
  triangulated against the self-report when it arrives; passive-only by default,
  active tests need explicit authorization.
• Agentic-browser driver class (§3.6) named as distinct from cowork-mcp
  (investigation-posture single-source); the single-arm paste rule (§4.12).

Calibration / report back

• Does the Dispatch ID close the re-dispatch reconciliation ambiguity in live
  failure-leg runs without adding operator friction?
• Does the late-bound build (spec-not-Envelope) reproduce correctly on long
  engagements, refreshing the rendering without re-scoping?
• Does the proactive return-handling card remove the "what do I do with these
  returns?" ask?

Provenance. Lens Library unchanged (v0.15). Snapshot PRISM_v2_17_0.md
(byte-identical at the v2.17.0 tag). Reference detail in the new Appendix J /
reference/dispatch-conventions.md (promotion map + canonical paste model +
digest-preimage recipe); card schemas in reference/templates.md. Source specs
(Engagements clone): DISPATCH_LIFECYCLE_v1, DISPATCH_CONVENTIONS §A–H,
ML-43/69/73/74. Operator gate confirmed: dispatch-id + pinned preimage.

---

Part of the PRISM v2.15.0 → v2.19.0 promotion arc. See RELEASING.md.

PRISM v2.16.0 — Operating-model backbone

MINOR, additive (no construct renumbering). Slice 2 of the dependency-ordered
sequence: the lanes/Desk backbone the dispatch round-trip (2.17) and closure
(2.18) build on.

Surface

• Lanes / roles operating model (new §3.7). An engagement runs as parallel
  lanes (each = {resume pointer, append-only log, inbox, one concern}) worked
  by sessions in a flexible role × context-tier matrix. Three isolate options
  (inline / separate session / clean-context sub-agent), with the sub-agent the
  default for a clean-context role when the holding session is contaminated
  (self-trigger). The net-helpful bar: the machinery is the session's
  bookkeeping, never the operator's.
• PRISM Desk + PRISM Meta standing-lane constructs, with the operator-facing
  PRISM UI (View+Controller over the repo Model; STATE/HEALTH/ACTION
  surfaces; render-from-verified-model), <Lane>-<N> session numbering, and the
  /prism-meta lane-entry command. Desk discipline: dispatch-not-execute,
  state-correction, propagate-don't-make-the-operator-the-checker, catch-one →
  propose-a-sweep. The stop-set: pause only at real gates (ratify strategy,
  scope change, ratify close, anything sent to the client), else auto-handle and
  report.
• Cross-lane OPEN_ITEMS inbox (§3.7.3) — tier classification (append-only /
  single-writer pointer / canonical), per-writer namespaced IDs, append-only
  dispositions, owner-drain, cross-lane handoffs, and the commit-discipline rule
  (rebase-on-live-tail / commit-only-own / commit-then-confirm). What's next
  splits into the single-writer body + the inbox; resume gains a drain step.
• Bundle-load integrity (§3.7.6) — a phase→bundle manifest in core + a
  per-session fail-loud entry-check + an inbox bundle-miss flag + a Desk
  re-sync audit (the safety net for the v2.19 core-slimming refactor).
• Setup onboarding & quick mode (new §6.6). SETUP_ONBOARDING: Setup
  generates the per-engagement SI from a template (gather → generate → ratify →
  commit → emit project-create/install cards), two-project model.
  SETUP_QUICKMODE: a first-class light mode (one ephemeral session, sub-agent
  fan-out; keeps the mini decision-brief + lite Probe-1 + SP-16/17/18 gates;
  graduation seeds a repo_backed Master).
• M5 self-band — fires per standing session's own context, not only the
  engagement band (Monitor extension, not a new SP).

Calibration / report back

• Does the lane/role model stay invisible to the operator (the net-helpful bar),
  or does it leak machinery into the operator's view?
• Does the OPEN_ITEMS inbox + commit-discipline eliminate the pointer/snapshot
  clobber class in live parallel runs?
• Does the stop-set draw the auto-vs-ask line where the operator wants it?

Provenance. Lens Library unchanged (v0.15). Snapshot PRISM_v2_16_0.md
(byte-identical at the v2.16.0 tag). Reference detail in Appendix I /
reference/lanes-ui.md. Source specs (Engagements clone): LANES_ROLES_PRISM_UI,
OPEN_ITEMS_inbox, SETUP_ONBOARDING, SETUP_QUICKMODE.

---

Part of the PRISM v2.15.0 → v2.19.0 promotion arc. See RELEASING.md.

PRISM v2.15.0 — Foundations

MINOR, additive (no construct renumbering). First of a dependency-ordered
sequence consolidating engagement learnings into the framework.

Surface

• SP-21 Trust structure, self-report advisory — load-bearing mechanisms rest
  on structure, the operator-set filename, and the copy-through digest; a
  producer's self-report is advisory. Rewrites §4.7 reconciliation: filename
  authoritative, self-ID an advisory cross-check, a mismatch a logged drift
  signal — never a halt.
• SP-22 Surface translation — nothing crosses a vendor/operator/card
  boundary un-rendered for that reader (parent of §4.12 self-containment, SP-17,
  the dispatch-card rule).
• retrieved | inferred finding-basis axis (§4.11) — decides recall-merge
  vs triangulate; SP-12 extended to inferences.
• SP-16 → visual-polarity/render-integrity channel; SP-10 → operator-environment
  state + external-send events; SP-13 substrate self-ID corroboration-only note.
• Defect fixes D1/D2/D3/D5; M10-rerun vs Follow-up (R1); M2 clobbered-pointer
  tie-break.
• Setup stakeholder-positioning (every stakeholder and the operator, stated
  positively); Setup-artifact re-audit; accumulate-never-overwrite.

Calibration / report back

• Does filename-authoritative reconciliation (SP-21) absorb real vendor self-ID
  drift without false halts?
• Does the retrieved|inferred axis route convergence correctly on mixed recon
  returns?
• Does stakeholder-positioning catch mis-scoped framing at P0 rather than late?

Provenance. Lens Library unchanged (v0.15). Snapshot PRISM_v2_15_0.md
(byte-identical at the v2.15.0 tag). Surface-drift map: Appendix D.

---

Part of the PRISM v2.15.0 → v2.19.0 promotion arc. See RELEASING.md.

PRISM Lens Library v0.16

One lens added, two source lists curated. Catalog grows 26 → 27 entries (5 universal + 22 domain across 6 packs). No schema change.

Release: https://github.com/Ronkupper/PRISM/releases/tag/v2.20.0 (paired framework release — re-embeds this catalog as Appendix G and names LL-D-022 in the §4.14 IVD default kit)

LL-D-022 — "Does this respect the reader?" (new, Pack 1: Using the product). A reader-respect pass that flags filler which spends the reader's attention and returns nothing — empty preambles, the prompt restated back, contentless transitions, and qualifiers that hedge without informing. Every paragraph is read against one question: what does the reader lose if this is cut? It completes the document-review family alongside LL-D-019 ("Who said otherwise?"), LL-D-020 ("Help or ammunition?"), and LL-D-021 ("Does a stranger follow?") — and joins them in the default kit of the Independent Validation Dispatch (§4.14). Specialist: line editor / substantive editor (reader-advocate posture). Evidence class: document. Informed by: the Strunk & White / Zinsser editorial tradition, the plain-language tradition, and generated-text review practice (boilerplate / filler detection).

recommended_sources: curation (O1). The two source-bearing lenses — LL-D-008 ("Compared to what?") and LL-D-009 ("Does it pay back?") — gain two sources each, each shipped with the mandatory bias-and-handling framing: and a recency: posture:

• Failory (both lenses) — a founder-interview and post-mortem corpus. Framing flags failure-selection plus interview-subject self-presentation bias; a founder's own account is a claim, not a finding. Answers lost-to enumeration (LL-D-008) and willingness-to-pay / burn post-mortems (LL-D-009).
• Start-Up Nation Central — Finder (both lenses) — a deep, continuously-updated structured source for the Israeli tech ecosystem. Framed as depth-by-ecosystem, not merely regional: in Israel-led categories — cybersecurity above all — its companies are global category innovators and disproportionately M&A targets (predominantly of U.S. acquirers), so its comparable / competitor / acquirer set (LL-D-008) and its exit / transaction record (LL-D-009) are globally material cross-checks.
• The existing pitch-deck-libraries source now names bestpitchdeck.com among its open examples (the source name stays generic).

Unchanged. The other 25 entries are untouched; the entry schema and pack structure carry over from v0.15.

Snapshot: lens/PRISM_lens_library_v0_16.md (byte-identical to the canonical file at this tag).