[michaldengusiak]

Merged and closed out - 2026-10-17

Merged into sow/2026-Q4 as 912ae94ef68af20b437fc92ec7e8891fd1cb640f, using a merge commit guarded to owner-approved head 4a7da6b1821d303d183fe87d275b5d0aaf05460e. No automatic merge or administrative bypass. Owner authorization came from the explicit gated merge instruction; no GitHub approval review was fabricated.

Final-head Build 39068752757 and SPDX 39068752748 passed. Post-merge Build 39070909651 and manually dispatched SPDX 39071240235 passed at the immutable merge SHA. Hosted comparator 100/100 and public-API 2/2, zero skipped. Dependency logs: SAM / SAM_Mollier sow/2026-Q4, SAM_IAPWS default branch.

Closeout: 68eded0060047af714853a02a152fb9c960bfeb1, docs-only PROJECT_PROGRESS.md on Q4 with [skip ci]. Fully merged task branch claude/tas-openstudio-comparison was deleted remotely with an expected-SHA lease and locally with branch -d after all cleanup guards passed; worktrees/frozen quarters preserved. Primary checkout is clean Q4. This removes dependency-branch shadowing for downstream UI checks.

No local test suite/build or real simulation was rerun during merge/closeout. Post-merge tests executed in CI; saved reports/hashes and existing acceptance artifacts were independently inspected. Numerical engine disagreement, HungaryHouse Tas crash, unavailable input manifests and producer deployment remain unresolved. UI#237/#238 are held for missing G1 evidence/explicit deferrals and approval of their documentation-corrected pins. Earlier implementation record follows; this merge status supersedes its not-merged wording.

---

Presentation-only two-decimal comparison Markdown

Status: implemented; not merged. Companion to SAM_UI#238. Follow-up to merged SAM_Validation#14.

• Base: sow/2026-Q4 at 8fa79f5c8559c2db11e0ce2a26193e5ccad80e79
• Tested content SHA: a39e4a9fd11019de9400f62fe5e68179539a4661
• Head: claude/tas-openstudio-comparison at 4a7da6b (record-only commit after the tested content)

Scope

Presentation only. comparison.csv and comparison-summary.json keep their schema, columns and full precision; Format.Number/Format.Percent remain the raw contract. No simulation, conversion, metric mapping, unit conversion, tolerance, gate or comparison arithmetic changed.

What changed

• New PresentationFormat: two-decimal engineering values, absolute differences and percentages; hour-of-year values and hour differences stay integers; null/NaN/infinity render n/a and are never zero; informational label Informational - excluded from numerical status; unit mismatch remains distinct; space name/GUID and input-manifest helpers.
• ComparisonResult adds contributing/informational Fail counts using the comparator's existing classification. The saved example is 10 Fail (4 contributing, 6 informational); overall/numerical Fail, coverage Warn, provenance/reconciliation Pass are unchanged.
• Markdown now explains the normalized difference formula and 5.00%/15.00% thresholds, the n/a/- markers, designDaySource=None, the input-manifest limitation and engine warnings/notes. Reconciliation uses two decimals and adds a normalized difference column; space metrics show the space name beside its GUID.
• Review finding fixed (fractional peak-hour thresholds): WriteMarkdown.AppendToleranceProfile cast HourWarnAbsolute/HourFailAbsolute to int, so a valid profile with 1.5 h / 24.5 h was reported as 1 h / 24 h. It now renders both with Format.Number, exactly as comparison.csv does. Integer formatting remains where it belongs: hour-of-year indices and hour differences. The default profile still renders 1 h / 24 h, so the goldens are unchanged.
• Markdown goldens refreshed for the presentation change. CSV and summary JSON goldens are byte-identical.

Validation

Check	Result
dotnet test SAM.Analytical.Benchmark.Compare.Tests -c Release	PASS 100/100 (includes the fractional-hour regression)
dotnet test SAM.Analytical.Benchmark.Compare.ApiTests -c Release	PASS 2/2
SAM_Validation.sln Release rebuild	PASS, 0 errors (1 pre-existing warning)
Saved box-both documents	Markdown regenerated; CSV and summary JSON SHA-256 byte-identical to the saved run
Regression cases	CSV and JSON byte-identical; Markdown golden refreshed

The new regression ReportTests.MarkdownKeepsFractionalHourThresholdsAndAgreesWithTheRawProfile asserts the Markdown contains 1.5 h / 24.5 h (and not 1 h / 24 h), that the summary JSON's toleranceProfileDetail.hourWarnAbsolute/hourFailAbsolute equal the raw profile, and that every CSV row carries 1.5,24.5.

Saved comparison invariants: 6 Match / 0 Warn / 10 Fail / 2 N/A; 4 Fail rows contribute to the numerical status and 6 are informational. Overall/Numerical Fail, Coverage Warn, Provenance/Reconciliation Pass are unchanged. Regenerated comparison.csv SHA-256 8092B384DD9D9BA9289E1C10B77DB7C93BCD95E911B4A0F32CEE93373C0ECD3D and comparison-summary.json SHA-256 36B9D353B35CDFEB1FBAC70B78317BF36C4F393159F114FCF39716D825FF4A98 are byte-identical to the saved run.

Gaps / follow-up

• The UI grid and its 1180/820 screenshots belong to SAM_UI#238, which now uses fixed readable column widths and the grid's own horizontal scrolling. This repository has no UI layout code.
• Numerical discrepancy is not resolved here: TAS heating/cooling and peak values differ from OpenStudio (box example: heating 742.78 vs 2810.02 kWh, cooling 2156.92 vs 3832.50 kWh). The 871/872 heating-hour and cooling-peak discrepancies are separate engineering investigations. No tolerance or engine change is proposed.
• Per-engine input manifests remain unavailable (SAM_OpenStudio#30, SAM_UI#230).

Hosted CI passed at final head 4a7da6b: Build (Windows) run 39068752757 succeeded, including Test benchmark comparator (dotnet test); SPDX run 39068752748 succeeded. The additional local suite and build results are listed above.

Next step

Merge this presentation PR before SAM_UI#238 so the UI and Markdown use the same helpers, then rerun #238 CI. Do not merge automatically.