[vjovanov]

Closes #253

With two github-issues blocks answering [1 2 3] and [4], ephor refresh javelin --quiet previously kept only whichever block finished last and exited 0. It now retains [1 2 3 4] in either completion order.

Both project providers and site sources retain every contribution under the existing provider identity. Each configured block contributes to that provider's slot, as clarified in §FS-001-forge-interface.9. Overlapping subjects retain their combined reasons. Shared reports also combine every report's attribution evidence before placing the subject once, preserving existing precedence and ambiguity rules. Combination and completeness are established before attribution or persistence publishes an answer (§AR-008-pipeline.2).

Any failed contribution makes the whole provider slot incomplete. Every distinct diagnosis remains visible, and the previous whole complete slot stays labelled last-good without advancing its timestamp or cursor. A successful subset cannot replace it as current; without previous data, the failure remains explicit and no rows are invented. A later complete answer replaces the old slot and clears failure state, including when empty. Discovered task stores of the same kind follow the same completeness and failure rule (§FS-001-forge-interface.6.1).

Refresh exits count actual contributions: mixed success and failure gives exit 4; every attempted contribution failing gives exit 3. Empty successful answers count, while idle projects and cached last-good data cannot turn total failure into partial failure. Five existing E2E-004 store-only failure expectations now require exit 3 because their sole attempted store failed. The manual explains repeated blocks, incomplete answers and recovery. Configuration shape, public provider identity, cache schema and binding/move policy are preserved. Query-cost work in #252 is separate.

Eight enabled regressions in E2E-065-every-provider-contribution-counts cover project and site answers, overlapping reasons and placement evidence, mixed and total failures, whole-slot last-good/recovery, and discovered stores. The overlapping shared-report regression failed before the attribution repair and passed afterward: both block orders now yield one project row with both reasons and no duplicate in the unattributed bucket. All eight passed with zero ignored tests.

Local validation at 508f2229bf finished READY: yes on the first gate run, in 8m02s:

• The original offline reproducer retained [1 2 3 4] in both forced completion orders and all six unheld turns; mixed failure returned exit 4 with a diagnosis in either order.
• pre-commit run --all-files passed Rust formatting, the warnings-as-errors locked all-target build, locked all-target Rust tests, Python tests, CI-pinned grund 0.18.0 check/format, file budgets and content guards.
• Explicit Python integration discovery passed 113 tests; check_boundary.py and check_parity.py passed.
• The full Rust suite passed again with symlinked TMPDIR, including all eight issue-253 regressions with none ignored.

These fixtures use an offline forge. Hosted CI and merge verification remain for shipping; the generated workflow summary below covers recorded invocations before shipping.

AI workflow: `rhei`, 13 agent invocations across 1 model; 8 tasks completed, 8 in progress.

1. github-issues-agent-grounds-ephor-253-implement-fba34977.ticket supervising (visit 1) — cdx, openai/gpt-6.1-sol — 1m43s — 523.7k in / 3.5k out
2. github-issues-agent-grounds-ephor-253-implement-fba34977.ticket supervising (visit 2) — cdx, openai/gpt-6.1-sol — 3m09s — 817.3k in / 7.0k out
3. github-issues-agent-grounds-ephor-253-implement-fba34977.ticket.specify specify — cdx, openai/gpt-6.1-sol — 11m32s — 4.1M in / 26.2k out
4. github-issues-agent-grounds-ephor-253-implement-fba34977.ticket supervising (visit 3) — cdx, openai/gpt-6.1-sol — 2m07s — 629.6k in / 5.0k out
5. github-issues-agent-grounds-ephor-253-implement-fba34977.ticket.implement implement — cdx, openai/gpt-6.1-sol — 12m18s — 5.4M in / 22.9k out
6. github-issues-agent-grounds-ephor-253-implement-fba34977.ticket supervising (visit 4) — cdx, openai/gpt-6.1-sol — 1m53s — 578.2k in / 4.5k out
7. github-issues-agent-grounds-ephor-253-implement-fba34977.ticket.review-1 review — cdx, openai/gpt-6.1-sol — 6m00s — 2.8M in / 14.5k out
8. github-issues-agent-grounds-ephor-253-implement-fba34977.ticket supervising (visit 5) — cdx, openai/gpt-6.1-sol — 1m53s — 493.2k in / 4.0k out
9. github-issues-agent-grounds-ephor-253-implement-fba34977.ticket.fix-1 fix — cdx, openai/gpt-6.1-sol — 14m52s — 5.9M in / 29.5k out
10. github-issues-agent-grounds-ephor-253-implement-fba34977.ticket supervising (visit 6) — cdx, openai/gpt-6.1-sol — 2m02s — 729.7k in / 4.8k out
11. github-issues-agent-grounds-ephor-253-implement-fba34977.ticket.review-2 review — cdx, openai/gpt-6.1-sol — 2m45s — 1.2M in / 6.6k out
12. github-issues-agent-grounds-ephor-253-implement-fba34977.ticket supervising (visit 7) — cdx, openai/gpt-6.1-sol — 1m54s — 840.2k in / 4.4k out
13. github-issues-agent-grounds-ephor-253-implement-fba34977.ticket supervising (visit 8) — cdx, openai/gpt-6.1-sol — 1m59s — 527.3k in / 4.0k out

Accounting	Value
cost	$6.13
total tokens	24.6M
input tokens (incl. cache)	24.4M
input cache read	23.2M
input cache write	-
output tokens (incl. cache)	137.0k
output cache read	-
output cache write	-
coverage	Complete