Making CI faster and 95% cheaper
Comments on “Making CI faster and 95% cheaper”
Claim-checked. Scored by all three judges, and its checkable claims were compared with the sources it cites. How far stories are checked
Vincent Koc describes how OpenClaw's CI runtime on Blacksmith runners fell from 158,546 hours in a September peak week to 21,395 in a November week (86.5% less), with estimated compute value down 95.5%. Blacksmith sponsored the capacity, and the workload also shrank, so the figures are not attributable to any single fix. The changes: deleting redundant tests, deferring full validation off main pushes, limiting repeat reviews, and selecting tests from the diff. Pull-request-to-merge time was not measured.
Why it ranked
No. 16 of 122 with a score of 0.66. The panel rated it highest on technical depth (0.88) and practical relevance (0.85), lowest on significance (0.57). Placed No. 16, No. 28 by score alone: the placement terms moved other stories past it.
- AI judgeJudge 2claude-haiku-5-5
- Detailed engineering write-up on CI cost reduction, test pruning and adaptive selection with measured figures, though sponsored by a runner vendor and with some unverified claims.
- AI judgeJudge 3gpt-5.6-luna
- A detailed, technically substantive CI engineering report with measured workload, cost, testing, scheduling, caching, and adaptive-selection changes. Its headline is qualified by explicit confounding caveats, though the 95% figure is not a controlled causal result.
Each judge wrote its line after answering the rubric, with the site and the author still hidden. Every answer behind this is in the table under Score.
Score
0.66total under rubric v2.3
- Significance0.57
- Reader interest0.80
- Audience breadth0.61
- Practical relevance0.85
- Substantiated0.80
- Technical depth0.88
- New information0.81
- Counts against: Promotional0.18
- Counts against: Overclaim0.42
Each judge’s answer per question (0–1), their mean and spread, and what the question adds to the score (weight × mean).
| Question | Judge 1 | Judge 2 | Judge 3 | Mean | Spread | Weight | Contribution (weight times mean) |
|---|---|---|---|---|---|---|---|
| SignificanceHow much does this change what people in tech know or can do? | 0.64 | 0.35 | 0.72 | 0.57 | 0.37 | +0.24 | +0.137 |
| Reader interestWould a well-informed senior engineer or tech founder want to read this today, and likely pass it on? | 0.78 | 0.70 | 0.93 | 0.80 | 0.23 | +0.26 | +0.209 |
| Audience breadthHow many people who build technology does it concern? | 0.62 | 0.50 | 0.70 | 0.61 | 0.20 | +0.13 | +0.079 |
| Practical relevanceHow useful is it to someone who builds software or tech products? | 0.96 | 0.70 | 0.90 | 0.85 | 0.26 | +0.22 | +0.188 |
| SubstantiatedHow well is it backed? Consider whether it is the primary source, and whether its main claims come with data, code, a demo, a reproducible method or citations. | 0.95 | 0.60 | 0.84 | 0.80 | 0.35 | +0.05 | +0.040 |
| Technical depthHow technically substantive is it? | 0.99 | 0.75 | 0.91 | 0.88 | 0.24 | +0.05 | +0.044 |
| New informationDoes it contain information absent from the stories covered in the last 7 days (list provided)? | 0.84 | 0.70 | 0.88 | 0.81 | 0.18 | +0.05 | +0.040 |
| PromotionalIs it mainly marketing with little substance (a press release, ad or sponsored post with no real information)? A launch post that carries real news is not promotional. | 0.07 | 0.35 | 0.12 | 0.18 | 0.28 | −0.19 | −0.034 |
| OverclaimDo the headline or conclusions claim more than the body supports? | 0.56 | 0.45 | 0.24 | 0.42 | 0.32 | −0.11 | −0.046 |
| Sum of contributions | +0.657 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.657 | ||||||
Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested.
The badge for this ranking, for the author's page or anyone's: one line of HTML, and it links back here.
<a href="https://rubric.hrih.si/editions/2026-11-17/16/"><img src="https://rubric.hrih.si/badge/2026-11-17/16.svg" alt="Ranked #16 on Rubric, 17 Nov 2026" width="245" height="28"></a>Claim check
Claim check: 3 matches source · 3 no source given · 2 couldn’t check. Each claim is compared with the source the article itself cites; a flag is published only when both checkers agree.
No source givenNumber
Blacksmith runner usage fell from 158,546 hours in the September peak week to 21,395 hours in the latest week, 86.5% less runtime.
The article gives no link or citation for it.
The article says “Our Blacksmith runners logged 158,546 hours in the September peak week and 21,395 hours in the latest week: 86.5% less runtime”
The source says No source to compare with.
No source givenNumber
The estimated value of the compute at Blacksmith's published prices fell 95.5% between the peak week and the latest week.
The article gives no link or citation for it.
The article says “The estimated value of that compute fell 95.5%”
The source says No source to compare with.
Couldn’t checkQuote
Peter Steinberger reported that OpenClaw deleted around 400k lines of its own tests without much change in code coverage.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “OpenClaw deleted around 400k LOC of its own tests without much change in code coverage”
The source says No passage could be compared.
Source cited by the article: x.com/steipete/status/2103147927313199260
Matches sourceBenchmark result
In one matched job of the same 240 test files, the 16-class runner cost 41.7% less ($0.672 vs $1.152) but took 16.5% longer (20m 51s vs 17m 54s) than the 32-class runner.
The cited source says the same thing.
The article says “41.7% cheaper. 16.5% slower”
The source says “Identical 240-file sample on 32/16-class: completed provider billing units imply $1.152 → $0.672 at the catalog rates; whole-job wall 1,074s → 1,251s”
Source cited by the article: github.com/openclaw/openclaw/pull/167021
Couldn’t checkNumber
Three completed PR runs executed only 15 of 4,018, 16 of 4,480 and 97 of 6,895 planned test entries after adaptive selection.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “Full inventory → executed4,018 → 1599.63% fewer entries”
The source says No passage could be compared.
Source cited by the article: github.com/openclaw/openclaw/actions/runs/42312466737
No source givenNumber
Estimated daily compute cost fell from about $7,738 on 5 November to $1,507 on 7 November, after adaptive selection rolled out around 6–7 November.
The article gives no link or citation for it.
The article says “Estimated daily compute cost then fell from about $7,738 on 5 November to $1,507 on 7 November”
The source says No source to compare with.
Matches sourceBenchmark result
Switching the typechecker from tsgo to tsrs cut a local full-typecheck from 447s to 64s, 85.7% less time.
The cited source says the same thing.
The article says “tsgo → tsrsSupported Linux / Apple Silicon hosts447s → 64s 85.7% less timeLocal”
The source says “pnpm tsgo:all` passed with no errors on both compilers: 64s with tsrs, 447s with `OPENCLAW_TYPECHECKER=tsgo”
Source cited by the article: github.com/openclaw/openclaw/pull/175711
Matches sourceNumber
One change packed 95 planned Node shards into 63, and another rebalancing change cut the slowest observed job from 403 seconds to 300.
The cited source says the same thing.
The article says “one change packed 95 planned Node shards into 63, keeping the gateway and agent suites isolated”
The source says “reducing the normal Node matrix from 95 planned shards to 63”
Source cited by the article: github.com/openclaw/openclaw/pull/95649
Also covered by
No other coverage was clustered with this story.
Elsewhere
Elsewhere: HN 3 pts, not on front page
Recorded for context only, checked 17 Nov 2026, 05:47 UTC. It played no part in the ranking.