# Release 0.4.7 — Downgrade rendering-equivalent spans instead of spending the bullet budget **Coordinated packages:** `ds4-context-core`, `ds4-context-reference-adapter`, and `ds4-context-engine` 0.4.7. **Implementation commit:** `1483bce`. ## Summary Behaviour patch that removes the source of the recurrent fail-closed fallback. Until 0.4.6 the repair treated a re-rendered exact value exactly like an invented one: it deleted whole bullets, and beyond eight affected bullets it refused to repair at all and handed the session to Pi. The class reports had shown for four releases that this was the common case — rendering and composition, not invention. The repair now retracts only the backticks of a span whose information the evidence holds in another rendering, keeps the text and the fact, and re-validates. Invented values, unrelated compositions, one-character deviations, case changes and unanalysed spans keep the existing fail-closed bullet removal. Validation strictness, the eight-bullet and 25% bounds, the fail-closed fallback, SQLite schema 16, canonical JSONL history and every provider-facing default are unchanged. ## Changes - `packages/core/src/compaction/summary-contract.ts`: - `RENDERING_EQUIVALENT_RELATIONS` names the relations that mean "the evidence holds this information in a different rendering": `escaped-form-present`, `unescaped-form-present`, `whitespace-collapsed-present`, `typographic-variant-present` and `composed-adjacent-present`. - Deliberately excluded, because prose would hide a wrong claim rather than a re-rendered one: `case-variant-present` (a case change can alter an identifier), `single-deletion-present` (a one-character deviation is indistinguishable from a wrong value), `composed-two-present-parts` (the evidence never contains that association) and every `not-classified-*` relation (*not analysed* is not evidence of support). - `downgradeRenderingEquivalentSpans` removes the two backticks of each rendering-equivalent span, byte-range based and applied after the bullet removals so every span is matched at its final position. - `analyzeUnsupportedExactValueBullets` partitions the rejected spans, applies the unchanged bounds to the ones that still need removal, and reports the new counters. `ExactValuePruneResult` and `ExactValuePruneAttempt` gained `downgradedSpans`, and the attempt status gained `downgraded` and `unrepairable`. - The span classifier is no longer described as observation-only: it is `analyzeSpanClasses` internally, and its relations now also select the repair treatment. Validation still never consults them, and the public `classifyUnsupportedExactValueSpans` report is unchanged. - `src/pi-adapter/summary-generator.ts`: the repair outcome is recorded as counts only — `unsupported-exact-bullets-pruned` for removed bullets and the new `unsupported-exact-spans-unquoted` for retracted quotes. - `tests/unit/summary-contract.test.ts`: five cases — a composed span with parts adjacent in one source is downgraded and the summary still validates; a span whose whitespace-collapsed and JSON-unescaped renderings are present is downgraded too; unrelated compositions and one-character deviations still cost their bullet; twelve affected bullets downgrade where the same summary without the adjacent rendering still answers `too-many-bullets`; a mixed summary removes only the one bullet that needs it and downgrades the other nine. - `tests/integration/compaction.test.ts`: eleven affected bullets complete the compaction with `validationStatus: "warning"` and `validationIssueCodes: ["unsupported-exact-spans-unquoted"]` instead of falling back, and the same fixture with the two parts at unrelated positions still returns `repair=too-many-bullets` with Pi compacting the session. - `docs/COMPACTION.md`, `docs/ADR/068-downgrade-rendering-equivalent-spans.md` and the ADR index record the decision. ## Why The affected-bullet series from production was 30/20 → 23/19 → 19/15 → 17/11 → 16/12, every run ending in `compaction.custom_fallback` with `unsupported-exact-value; repair=too-many-bullets`. The class-only reports attributed the rejected spans to rendering or composition in every interpretable run: `escaped-form-present` 11 of 16 in the first, then composition 9 of 14 with 2 one-character deviations, 2 absences and 1 over-length span in the second. `no-near-miss` — the class that means invention — was the smallest in both. Each earlier patch removed one class (escaped equivalence in 0.4.4, prompt grading in 0.4.5) while the bound stayed exhausted by the rest, so the fallback persisted. The information in those spans was present all along; only the verbatim surface form was not, and discarding a supported fact — or losing the compaction — is the wrong response to a rendering difference. The repair now applies deterministically the ladder the 0.4.5 prompt asks the model to follow, instead of depending on the model complying ([ADR 068](ADR/068-downgrade-rendering-equivalent-spans.md)). ## Privacy and safety - Only the backticks are removed. Span text is never logged, notified or persisted in diagnostics: the recorded effects are two counters, and the failure path keeps the class-only report. - No provider call is added: the downgrade happens after generation, on text the model already produced. - The downgraded text is sanitized output that already passed privacy handling; removing backticks does not change its classification. ## Compatibility and persistence - No configuration key is added or changed. `downgradedSpans` is additive on the repair result, and `unsupported-exact-spans-unquoted` is a new issue code in the recorded node details. - The eight-bullet and 25% bounds are not raised: they now count only the bullets that still need removal. A span downgraded in place consumes no bullet. - SQLite schema 16, migrations, the Pi JSONL boundary, tool contracts and the reference adapter's public surface are untouched. ## Measured scope and limitations - Verified by fixtures, not yet by a natural production failure: the first real `compaction.custom_fallback` (or its absence) after updating the failing host remains the end-to-end confirmation. - The fallback is still possible — deliberately — when more than eight bullets contain genuinely unsupported spans or when removal would exceed 25% of the summary. Those are the cases where the evidence does not hold the information. - A downgraded value is still stated, as prose. DS4 stops asserting exactness it cannot verify but does not delete the fact, and the counters make the trade visible. - The repair now consumes the bounded class analysis (default 24000 evidence-source scans). A span left unanalysed is never downgraded, so budget exhaustion keeps the previous fail-closed behaviour. - `case-variant-present` is intentionally not downgraded, so summaries that only differ by letter case still take the bullet-removal path. If production reports show that class dominating, that is the next measured decision. ## Validation - `npm run check` on the release commit: builds, `tsc --noEmit`, and **100 test files / 649 tests passed** (7 new: 5 unit cases for the repair rule and 2 integration cases, one of which pins the unchanged fallback on unrelated composition). - `npm run pack:check` in a clean consumer: `ds4-context-core@0.4.7` (251 files), `ds4-context-reference-adapter@0.4.7` (7 files) and `ds4-context-engine@0.4.7` (110 files). - `npm run quality:compare`: candidate `task-weighted-v0.2-candidate` 0.9875 against the frozen baseline `static-ranking-v0.1` 0.808156, unchanged. - `npm run schema:context-persistence` (`passed: true`). - No provider calls are involved. `jev_verify` is not available in this environment (project verification disabled), so the checks above were run directly. ## Publication and registry verification Published manually in dependency order from `bf221b5`: `ds4-context-core@0.4.7`, then `ds4-context-reference-adapter@0.4.7`, then `ds4-context-engine@0.4.7`, all with the default `latest` tag, after `npm run check` and `npm run pack:check` on the same commit. `npm run registry:check -- 0.4.7` needed five attempts, each with a fresh npm cache, before the registry CDN served the new packuments (the known propagation delay after a publish); the fifth reported "Verified all DS4 registry packages at exact version 0.4.7". All three packages install in a clean consumer, the exact adapter/core dependencies resolve, the public core and KV exports import, the compiled reference conformance and packaged quality corpus run, the `ds4-context-storage` CLI shim responds, and the published Pi extension starts against isolated offline RPC state. `npm view dist-tags.latest` resolves to 0.4.7 for all three packages. Annotated tag `v0.4.7` and the GitHub Release were created from `bf221b5`: . No session data, credentials, provider payloads or prompt text were read or written while preparing this release, and no provider call is involved in the release procedure.