fix(content): withdraw the break-even claim, correct the slippage range, sync to v0.13.3 (#137, #138, #139)

[eaitbrahim]

Closes #137. Closes #138. Closes #139.

Why

The front page's "The honest result, stated first" block carried three figures that no longer describe what keel measures — and one that keel's own cost model now falsifies.

1. The matrices were superseded (#137)

"In one measurement matrix, 0 of 90 configurations cleared; in another, 0 of 82."

Those covered three rule families on a flat-5bp engine. Five signal families ship at v0.13.3, and this site's own rendered changelog for that release already says 0 of 90 → 0 of 114 → 0 of 138. The front page was understating a measurement published two clicks away.

Now: 0 of 120 at shipped defaults, 0 of 138 across every matrix run.

2. The break-even claim is withdrawn — explicitly, on the page (#137)

"Inside the venue's fee-free monthly allowance, the reconstructed rules sit indistinguishably at break-even — a 14.9% win rate against a 14.88% break-even."

That 14.88% comes from docs/research/2026-08-20-quant-lab-note-cross-verification.md §5, computed at ψ = 5bp. The 09-01 per-product restatement measured that 0 of 24 assets reach that floor — 1.1× to 36.8×, median ≈10×. The note's own sensitivity row already fails the claim at ψ = 10bp, a fifth of the measured median.

So the summary badge's "at best they touch break-even inside the venue's fee-free allowance" was not true, and neither was "the allowance rail is the profitability boundary". It is the boundary between decisively negative and clearly negative.

It is retracted in the copy rather than quietly edited. On this site the withdrawal is the credibility, not the number.

3. Fees are no longer described as the whole cost (#137)

The block said "the taker fee is the entire result". Since #334/#335 that is not the project's own position: the cost-model error was worth 0.090 of median profit factor — larger than the best genuine improvement any rule change has produced.

4. Features page drift (#138)

• "from 5 to 50 basis points" → SLIPPAGE_CAP_PCT has been 0.01838 (183.8bp) since #523, which re-derived the cap as the corpus tail; the round 50bp was the defect that change removed. Understating the ceiling by 3.7× on the page that promises "if a claim ever drifts from the code, the link is how you catch us".
• A cost-model conflation — the sentence read as though the whole keel simulate pass is liquidity-priced. At v0.13.3 only the edge table is; the account pass and the DCA benchmark (which that same sentence names) stay flat, as simulate.py states beside its own numbers.
• The Arabic meta description still said 18 rails while the Arabic body beside it, and every EN/FR string, said twenty.

Verifiability

Every figure the new copy states is sourced from a document that is in v0.13.3, so a reader can check it at the tag the site pins. A third link — "How execution is priced" → the 09-01 restatement — puts the cost claims one click from the record that measured them. Both experiment links return 200.

The confirming measurement (0 of 240, p_be 20.52%) is CodeGateSoftware/keel#739. It is deliberately not cited here: it is not in a release yet, so the link would 404. Worth a follow-up to sharpen this copy once it ships in a tag.

Commit structure — please read

• 3aed993 — the v0.13.3 sync (Sync site copy to v0.13.3: rails 18 → 20, trade scope, the order ledger, eight console views #139): guides seven→eight views with Orders, keel scope attest, rails 18→20 in the comparison matrix, manifest additions, regenerated screenshots. This was already in the working tree and is not my work; it is committed separately so authorship stays legible.
• 9a6f309 — the corrections above. It also carries the rails 18→20 pass and the trade-scope / order-ledger / research-front-door sections for Sync site copy to v0.13.3: rails 18 → 20, trade scope, the order ledger, eight console views #139, because those edits were already in the tree in the same three files and could not be split out without reconstructing them.

Verification

• npm run check: 0 errors, 0 warnings (4 pre-existing hints in untouched news/[page].astro files)
• npm run build: 73 pages
• Rendered EN/AR/FR honest blocks inspected in dist/; all three locales carry matching four-paragraph structure
• No-break spaces normalised to U+00A0, the convention every other file here uses

One thing to check before merge

I wrote the Arabic and French copy rather than leaving translatedFromRev stale. The retraction paragraph is delicate in both — worth a native read.

🤖 Generated with Claude Code

https://claude.ai/code/session_01EXz13qp1UM3pBa6BqsRqvC

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud