Ramp run — one run of Jerry OS, on the record

Ticket: JOS-29 (sub-ticket of JOS-26). Authority: D-024 (Founder approved Ramp as the demo company for the Company-Specific Proof run, 2026-09-04; cited as D-022 until the verifier caught it on 2026-09-06), D-060 (Material Claim Verification Gate), D-065, D-066. Operator: claude-worker-cd-sweep, 2026-09-06, local interactive session with direct web access (the 2026-09-04 pass hit a full network-egress block — EGRESS_BLOCKED, confirmed against a Wikipedia control — and worked from search synthesis; this pass had direct web access).

What this file is. The workbench for the flagship case study. Every step of the run is on this page, in the order it happened: the trigger, the frame, the evidence and its grades, the theses and which one died, the verification gate with the actual before/after edits, the independent check, what shipped, what we'd measure. The public page at /run/ is built from this file and shows nothing this file doesn't contain.

What this run is not. Not work done for Ramp; not a claim about Ramp's internal numbers; not a prediction dressed as a result. Public information only. Every material claim carries a label from the OS's eight-label claim-status vocabulary (CHARTER.md); the four that carry the public-claim work are Publicly verified, Reasoned hypothesis, Internal input required, Illustrative output (the other four grade the mechanism's own maturity: Workflow tested · Live in operating system · Prototype · Requires integration).

Lineage (D-066). Modus Create was an earlier production version of this operating model (first PMM function built from zero, 2024–2025). What follows is the current system, run transparently on a company Jerry is not pursuing.

0. Trigger and frame

Trigger. Ramp shipped Router on 2026-08-19/20: per Ramp, a single API endpoint that routes each AI request to the lowest-cost model meeting a developer's quality bar. It launched into a category that consolidated the same week (Stripe announced its OpenRouter acquisition on 2026-08-19; Palo Alto Networks had closed on Portkey in May). A fintech entered AI infrastructure. That is a product-marketing event: a new buyer, a contested category claim, a launch whose readiness nobody outside can see.

Frame (the question the run answers). If this launch were mine, what would I be able to say is true, what would I position on, how would I tier and measure it, and what would I refuse to claim? Four artifacts answer that, in the order the OS produces them: product truth → ICP and positioning → launch brief and workback → KPI tree. Company selection came first and is on the record too.

Why Ramp (Jerry's decision, not the system's). The Company-Selection Gate scored four candidates on a 100-point rubric; Ramp scored 95, Deel 88, Gusto 83, and a security vendor 71 (name withheld on the public page: an application of Jerry's is open there; the internal scorecard is unchanged). The Founder approved the system's recommendation, Ramp, on 2026-09-04 (D-024). The rubric was the system's; the selection was Jerry's; the rule that a company he is applying to can never be public proof is his (D-065).

1. Inputs (all in this repo)

2. Claims ledger (graded evidence)

Extracted read-only from the five documents on 2026-09-06 before any edit. "Label" is the label the document gave the claim; "Risk" is the gate's pre-verification read. Rows marked ▲ went to the verification gate in §4.

Rows 39–43 were added after the verifier found them missing (§6).

3. Thesis tournament (positioning)

Corrected twice, on the record. The ICP document ran Module 02's durability check on Ramp's headline claim, stated one positioning hypothesis, and selected one line for the technical evaluator. The first version of this section presented a reconstruction as the record (caught by the first verifier, §6). The second version, and the document itself, said the neutrality line and the spend link were "OS-authored candidate messages that Ramp is not currently running." A second, independent verification fetched Ramp's launch post and proved that false (§6, second pass): both are Ramp's own launch messaging, snapshotted under 03_Company_Proof/sources/ramp-router-launch-2026-08-19/. The system's contribution is the selection per buyer, the durability test, and the OpenRouter limit.

What a reader should take from the record: the system did not keep the headline; it chose which of Ramp's existing claims goes to which buyer and how far each can be pushed; and twice, when it overstated its own contribution, an outside check caught it and the correction shipped.

4. Verification gate (verify → revise → remove), with the edits

D-060: every material public claim is verified against a primary source, revised where the source says something narrower, or removed where it cannot be verified. Run on 2026-09-06 by fetching the primary documents directly: Ramp's release (PRNewswire, 2026-08-19), TechCrunch (2026-08-20), router.com, Stripe's newsroom (2026-08-19), Palo Alto Networks' releases, OpenRouter's Series B announcement, and the review sites. Fourteen edits landed in the source documents; the commit is a692b12 plus the follow-up in this ledger's commit. Each entry shows the text as it was and as it is now.

V-01 · REVISE · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

| 1 | Router — multi-model AI request routing | Contested (GA-styled marketing site vs. one closed-beta/waitlist report) | TechCrunch 2026-08-20; router.com; X post at launch | Safe with caveat | State as "launched Aug 2026; public sources conflict on current access tier — confirm before citing as generally available" | 2026-09-04 |

After

| 1 | Router — multi-model AI request routing | Live, self-serve (Ramp's release: "live now at router.com"; TechCrunch: "launched"; router.com: email sign-in link, no gate) — one social post at launch described a waitlist and is outweighed by three primary sources | Ramp release 2026-08-19 (PRNewswire); TechCrunch 2026-08-20; router.com fetched directly 2026-09-06 | Safe with caveat | State as "launched Aug 2026, self-serve; U.S.-only; enterprise features 'coming soon' per Ramp" | 2026-09-06 |

V-01b · REVISE · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

1. Router's shipped-status is genuinely unresolved in public evidence. One contemporaneous source

After

1. Router's shipped-status was unresolved at the 2026-09-04 pass and was resolved on 2026-09-06 by fetching the primary documents directly (row 1 above now reads "Live, self-serve"). The original entry is preserved below as the record of what the mechanism refused to guess. One contemporaneous source

V-02 · REVISE · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

| 3 | Router model roster / count | Shipped, exact roster unconfirmed | Two inconsistent secondary syntheses (see Known evidence) | Not safe (specific count/list) | Re-verify directly against live model picker before citing a number | 2026-09-04 |

After

| 3 | Router model roster / count | Shipped; provider roster attested by two primary documents that agree (TechCrunch 2026-08-20: OpenAI, Anthropic, DeepSeek, Moonshot, Minimax, Nvidia, xAI, Z.ai; Ramp's release 2026-08-19: OpenAI, Anthropic, xAI, Gemini "coming soon", open models Nvidia/Kimi/DeepSeek/GLM/Qwen via Fireworks AI, Google, AWS, Together AI, Baseten, Crusoe) | TechCrunch; PRNewswire release; both fetched directly 2026-09-06 | Safe for provider names; not safe for a model count (no primary source publishes one; "27 models" remains an uncited secondary figure) | Cite providers by name with the source; never a count | 2026-09-06 |

V-02b · REVISE · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

2. Router's exact model roster is inconsistently reported across near-simultaneous sources. One synthesis lists

After

2. Router's exact model roster looked inconsistent across secondary syntheses on 2026-09-04; on 2026-09-06 the two primary documents were fetched directly and agree on providers (row 3 above), while still publishing no model count. The "SpaceXAI" entry turned out to be the release's own wording, not a search artifact. The original entry is preserved below as the record of the refusal. One synthesis lists

V-03 · REVISE · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

| 9 | Competitive-timing signal: Stripe agreed to acquire OpenRouter, reported 2026-08-16, 4 days pre-Router-launch | Published | Bloomberg, Stripe newsroom, TechCrunch |

After

| 9 | Competitive-timing signal: Stripe's agreement to acquire OpenRouter was reported 2026-08-16 (Bloomberg, "over $7 billion") and announced by Stripe on 2026-08-19 — the same day Ramp's Router release went out; Stripe's announcement states no price, press reports range from over $7B (Bloomberg) to more than $8B (Semafor, Axios) | Published | Bloomberg 2026-08-16; Stripe newsroom 2026-08-19 (fetched directly 2026-09-06); Semafor 2026-08-21; TechCrunch |

V-03b · REVISE · 03_Company_Proof/02_icp-positioning-ramp.md

Before

The same week, Stripe announced its acquisition of OpenRouter (reported $7.5–8B), with the two announcements landing roughly 70 minutes apart according to independent tech press.

After

The same day (2026-08-19), Stripe announced its acquisition of OpenRouter — Stripe's own announcement states no price; press reports range from "over $7 billion" (Bloomberg, 2026-08-16) to more than $8 billion (Semafor, 2026-08-21; Axios) — with the two announcements landing roughly 70 minutes apart according to independent tech press.

V-04 · REVISE · 03_Company_Proof/03_launch-brief-workback-ramp.md

Before

OpenRouter (unified gateway, 300+ models across 60+ providers; ~$113M Series B)

After

OpenRouter (unified gateway; "400+ models from more than 80 providers" per Stripe's 2026-08-19 acquisition announcement; $113M Series B at a $1.3B valuation, 2026-05-26, led by CapitalG)

V-05 · VERIFY · 03_Company_Proof/03_launch-brief-workback-ramp.md

Before

Portkey (observability/DevOps-focused; acquired by Palo Alto Networks, May 2026)

After

Portkey (observability/DevOps-focused; acquisition by Palo Alto Networks announced 2026-04-30 and completed 2026-05-29, per Palo Alto Networks' own releases)

V-06 · REMOVE · 03_Company_Proof/03_launch-brief-workback-ramp.md

Before

• OpenRouter, Portkey, and Martian together account for roughly 97.8% of tracked capital raised in the AI-inference-gateway category over the trailing 24 months (~$181M total, ~$153M of it OpenRouter's).

After

• (Removed 2026-09-06 by the Material Claim Verification Gate: a "97.8% of tracked category capital" figure with no named source could not be verified and carried false precision. The verified funding fact is OpenRouter's $113M Series B above.)

V-07 · REVISE · 03_Company_Proof/04_kpi-tree-measurement-ramp.md

Before

with one unnamed company cited at 92% — self-reported, press-relayed figures, not independently audited.

After

with one named customer (Valentin De Matos of Delphi, quoted in Ramp's release coverage) cited at 92% — self-reported, press-relayed figures, not independently audited.

V-08 · VERIFY · 03_Company_Proof/04_kpi-tree-measurement-ramp.md

Before

• Data handling: Router stores inputs/outputs/metadata to improve the service by default, with an opt-out and a zero-data-retention option on U.S.-hosted models.

After

• Data handling: Router records model inputs, outputs and tool calls for one year by default (opt-out, not opt-in; PII removed before use, per Ramp as reported by TechCrunch 2026-08-20), with a zero-data-retention option on U.S.-hosted models (router.com, fetched 2026-09-06).

V-09 · REVISE · 02_Company_Selection/scorecard.md

Before

Deel confidentially filed an S-1 on 2026-02-15 at a reported $17.3B valuation, with an IPO "as soon as 2026" reported by CNBC. [Publicly verified — https://www.cnbc.com/2025/02/04/hr-unicorn-deel-prepares-for-ipo-as-soon-as-2026/, and corroborating S-1 filing date per https://www.techstackipo.com/ipo/deel]

After

CNBC reported in February 2025 that Deel was preparing for an IPO "as soon as 2026" [Publicly verified — https://www.cnbc.com/2025/02/04/hr-unicorn-deel-prepares-for-ipo-as-soon-as-2026/]; a confidential S-1 filing on 2026-02-15 at a $17.3B valuation is asserted by one IPO aggregator (https://www.techstackipo.com/ipo/deel) and, as of 2026-09-06, is not corroborated by any major outlet or company statement. [Reasoned hypothesis — downgraded 2026-09-06 by the Material Claim Verification Gate: the CNBC citation predates the asserted filing by a year and cannot corroborate it]

V-10 · REMOVE · 03_Company_Proof/02_icp-positioning-ramp.md

Before

• Independent secondary commentary (a comparative reviews digest) notes reviewers "reported issues with enterprise-level visibility and permissions," i.e. some public signal that Ramp's platform experience is stronger for smaller orgs than for enterprise buyers. [Reasoned hypothesis — single secondary source, not cross-verified against a second independent outlet]

After

• (Removed 2026-09-06 by the Material Claim Verification Gate: a single secondary digest's claim that reviewers "reported issues with enterprise-level visibility and permissions" could not be corroborated in G2 or Capterra review text on re-check; a negative claim about a company's product does not ship on one uncorroborated source.)

V-03c · REVISE · 03_Company_Proof/02_icp-positioning-ramp.md

Before

now backed by Stripe's payments infrastructure and an estimated $7.5–8B acquisition price.

After

now backed by Stripe's payments infrastructure (acquisition announced 2026-08-19; price undisclosed by Stripe, reported between over $7B and $8B).

Tally: 2 verified with added precision, 10 revised, 2 removed. Two of the revisions resolve refusals the system made on 2026-09-04 (access tier, model roster) that a primary fetch could settle; the refusal and its resolution are both preserved in the Product Truth Brief so a reader can see the mechanism stall and then clear.

What the gate caught that reviewers would not have: a Deel citation whose URL predates the filing it was said to corroborate by a year (V-09); a "97.8% of category capital" figure with no source (V-06); a negative claim about Ramp's product on one uncorroborated digest (V-10); an OpenRouter scale figure that understated the competitor's own published scale (V-04); a price for the Stripe deal that Stripe never stated (V-03).

Second pass — from the outside review (§6), 2026-09-06

The verifier's findings produced a second set of edits. Same rule, same record. 35 edits: 3 verify, 31 revise, 1 remove.

R-01 · REVISE · finding F1 · all five documents

Before

canon/DECISIONS.md D-022 (cited as the Founder's approval of Ramp)

After

canon/DECISIONS.md D-024 — the actual entry: "JOS-25 flagship case study: Ramp approved as demo company; Jerry-add test made a hard QA gate — Founder". D-022 in the repo log is an unrelated heartbeat/coordination decision; the confusion is the two-numbering-systems hazard D-022 itself names. 9 occurrences corrected.

R-02a · REVISE · finding F2 · 03_Company_Proof/04_kpi-tree-measurement-ramp.md

Before

Router records model inputs, outputs and tool calls for one year by default (opt-out, not opt-in; PII removed before use, per Ramp as reported by TechCrunch 2026-08-20), with a zero-data-retention option on U.S.-hosted models (router.com, fetched 2026-09-06).

After

Router records model inputs, outputs and tool calls for one year by default, with PII removed before use (per Ramp, as reported by TechCrunch 2026-08-20); router.com states users "can control some of" these settings and can select U.S.-hosted models with zero data retention — no explicit opt-out mechanism is published in either primary source (router.com fetched 2026-09-06).

R-02b · REVISE · finding F2 · 03_Company_Proof/03_launch-brief-workback-ramp.md

Before

a one-year data-retention window on inputs/outputs/tool calls, opt-out rather than opt-in by default.

After

a one-year default retention window on inputs/outputs/tool calls (TechCrunch 2026-08-20); no explicit opt-out mechanism is published — router.com says users "can control some of" these settings and offers zero-data-retention U.S.-hosted models (corrected 2026-09-06).

R-02c · REVISE · finding F2 · 03_Company_Proof/03_launch-brief-workback-ramp.md

Before

(1) the opt-out, one-year data-retention default, flagged explicitly as a trust concern;

After

(1) the one-year default data-retention window, flagged explicitly as a trust concern;

R-03a · REVISE · finding F3 · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

• No public source states Router's actual current access tier (open GA vs. still-gated waitlist) as of 2026-09-04, roughly two weeks post-launch.

After

• No public source states Router's actual current access tier (open GA vs. still-gated waitlist) as of 2026-09-04, roughly two weeks post-launch. (Resolved 2026-09-06 — see ledger rows 1 and 3: access tier and provider roster were settled by direct fetch of Ramp's release, TechCrunch and router.com; this paragraph is preserved as the 2026-09-04 record.)

R-03b · REVISE · finding F3 · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

| Router GA vs. beta status | Current release-tracker status for the Router account-provisioning gate | Router PM / eng release owner |

After

| Router GA vs. beta status (resolved 2026-09-06 from public sources — row 1; kept for the record) | Current release-tracker status for the Router account-provisioning gate | Router PM / eng release owner |

R-03c · REVISE · finding F3 · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

Rows 1, 3, 4, and 5 are the ones the mechanism explicitly refuses to certify without internal access — that is the output the assignment asked for: not a clean Ramp overview, but a marked map of exactly where public evidence stops being sufficient.

After

Rows 1, 3, 4, and 5 were the ones the mechanism explicitly refused to certify without internal access on 2026-09-04 — that is the output the assignment asked for: not a clean Ramp overview, but a marked map of exactly where public evidence stops being sufficient. On 2026-09-06 rows 1 and 3 cleared from public sources alone (direct fetch of the primary documents); rows 4 and 5 still cannot clear from outside. The sections below describe the 2026-09-04 state and are preserved as written, with dated notes where the state changed.

R-03d · REVISE · finding F3 · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

it is to hold them at "safe with caveat" or "not safe," exactly as recorded above, and stop.

After

it is to hold them at "safe with caveat" or "not safe," exactly as recorded above, and stop. (Rows 1 and 3 later cleared on public evidence, 2026-09-06; rows 4 and 5 remain held.)

R-03e · REVISE · finding F3 · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

(and for row 1, one of those two is a single social-media post, the lowest-confidence source type used anywhere in this brief).

After

(and for row 1, one of those two was a single social-media post, the lowest-confidence source type used anywhere in this brief — superseded 2026-09-06 by three primary sources).

R-03f · REVISE · finding F3 · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

four of nine rows are explicitly held short of "safe."

After

four of nine rows were explicitly held short of "safe" at first run (two of nine after the 2026-09-06 re-verification).

R-03h · REVISE · finding F3 · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

this row cannot sit at "safe with caveat" indefinitely once that information becomes available.

After

this row cannot sit at "safe with caveat" indefinitely once that information becomes available. (Triggered and executed 2026-09-06: router.com and Ramp's release both state live, self-serve access.)

R-03i · REVISE · finding F3 · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

That single market fact is the reason ledger rows 1, 3, 5, and 9 (Router's access tier, model roster, performance claims, and competitive framing) get prioritized for the next re-verification pass

After

That single market fact is the reason ledger rows 1, 3, 5, and 9 (Router's access tier, model roster, performance claims, and competitive framing) got prioritized for the next re-verification pass (run 2026-09-06)

R-04 · REVISE · finding F4 · 03_Company_Proof/03_launch-brief-workback-ramp.md

Before

the incumbent (OpenRouter) is well-capitalized (~$153M raised)

After

the incumbent (OpenRouter) is well-capitalized ($113M Series B at a $1.3B valuation, 2026-05-26)

R-05 · REMOVE · finding F5 · 03_Company_Proof/03_launch-brief-workback-ramp.md

Before

• Sequencing-plan adherence, partial: the beta-savings claim ("customers already using Router" report ~40% cost cuts at launch) is public evidence of a closed-beta-before-GA pattern, consistent with (not proof of) internal-readiness-before-external-comms discipline.

After

• Sequencing-plan adherence: (inference removed 2026-09-06 — Ramp's "customers already using Router" phrasing was read as evidence of a closed beta; no primary source describes an external beta, and Ramp's release attributes the prior usage to three years of internal use. No public evidence on sequencing either way.)

R-06a · REVISE · finding F6 · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

and is named here once per CLAUDE.md's capability-gap-disclosure rule rather than left implicit.

After

and is named here once per CLAUDE.md's capability-gap-disclosure rule rather than left implicit. (Superseded 2026-09-06: the Material Claim Verification Gate re-ran in a local session with direct web access; every entry citing "fetched directly 2026-09-06" is a primary-document read, not search synthesis.)

R-06b · REVISE · finding F6 · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

Sources consulted (access date 2026-09-04 for all, via web search synthesis; direct WebFetch blocked per the environment note above)

After

Sources consulted (access date 2026-09-04 for all, via web search synthesis; direct WebFetch blocked per the environment note above — superseded 2026-09-06 for the rows marked "fetched directly")

R-06c · REVISE · finding F6 · 03_Company_Proof/02_icp-positioning-ramp.md

Before

Facts below sourced this way are cited as [Publicly verified — via WebSearch snippet, <domain>, accessed 2026-09-04; direct fetch unavailable this session] rather than pulled from full page text.

After

Facts below sourced this way are cited as [Publicly verified — via WebSearch snippet, <domain>, accessed 2026-09-04; direct fetch unavailable this session] rather than pulled from full page text. (Superseded 2026-09-06: the Material Claim Verification Gate re-ran in a local session with direct web access; every entry citing "fetched directly 2026-09-06" is a primary-document read, not search synthesis.)

R-06d · REVISE · finding F6 · 03_Company_Proof/03_launch-brief-workback-ramp.md

Before

so facts below are drawn from WebSearch's synthesis of those sources rather than primary-document quotation):

After

so facts below were drawn from WebSearch's synthesis of those sources rather than primary-document quotation; superseded 2026-09-06 where an entry cites a direct fetch or Stripe's / Palo Alto Networks' own releases):

R-06e · REVISE · finding F6 · 03_Company_Proof/03_launch-brief-workback-ramp.md

Before

Flagged once, here, per this repo's standing rule on stating capability gaps rather than re-discovering them silently.

After

Flagged once, here, per this repo's standing rule on stating capability gaps rather than re-discovering them silently. (Superseded 2026-09-06: the Material Claim Verification Gate re-ran in a local session with direct web access; every entry citing "fetched directly 2026-09-06" is a primary-document read, not search synthesis.)

R-07 · REVISE · finding F7 · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

Shipped; provider roster attested by two primary documents that agree (TechCrunch 2026-08-20: OpenAI, Anthropic, DeepSeek, Moonshot, Minimax, Nvidia, xAI, Z.ai; Ramp's release 2026-08-19: OpenAI, Anthropic, xAI, Gemini "coming soon", open models Nvidia/Kimi/DeepSeek/GLM/Qwen via Fireworks AI, Google, AWS, Together AI, Baseten, Crusoe)

After

Shipped; provider rosters from two primary documents overlap and are reconcilable under model/provider-family naming (Moonshot≈Kimi, Z.ai≈GLM), not identical — TechCrunch 2026-08-20: OpenAI, Anthropic, DeepSeek, Moonshot, Minimax, Nvidia, xAI, Z.ai; Ramp's release 2026-08-19: OpenAI, Anthropic, "SpaceXAI" (the release's own wording, reproduced uncorrected), Gemini "coming soon", open models Nvidia/Kimi/DeepSeek/GLM/Qwen via Fireworks AI, Google, AWS, Together AI, Baseten, Crusoe; TechCrunch alone names Minimax, the release alone names Qwen

R-08a · REVISE · finding F8 · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

**[Publicly verified as Ramp's stated claim, sourced to Ramp's own launch materials as reported by TechCrunch/Unite.AI; no independent audit or methodology disclosure found.

After

**[Publicly verified as Ramp's stated claim — primary source is Ramp's own release (PRNewswire, 2026-08-19: "reduced their inference costs by 40% on average", "approximately 30%", "99.9%+ reliability"), relayed by Unite.AI; TechCrunch's launch article carries none of these figures (citation corrected 2026-09-06); no independent audit or methodology disclosure found.

R-08b · REVISE · finding F8 · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

| Vendor-stated only | TechCrunch/Unite.AI reporting Ramp's own materials | Not safe (as independent fact) | Attribute to Ramp explicitly; no third-party audit found | 2026-09-04 |

After

| Vendor-stated only | Ramp's release (PRNewswire 2026-08-19, fetched directly 2026-09-06); router.com; Unite.AI relay | Not safe (as independent fact) | Attribute to Ramp explicitly; no third-party audit found | 2026-09-06 |

R-08c · REVISE · finding F8 · 03_Company_Proof/02_icp-positioning-ramp.md

Before

**[Publicly verified as a claim — via WebSearch snippet, techcrunch.com/unite.ai, accessed 2026-09-04;

After

**[Publicly verified as a claim — primary source Ramp's own release (PRNewswire 2026-08-19), relayed by unite.ai; citation corrected 2026-09-06 (TechCrunch's article does not carry the figure);

R-10 · REVISE · finding F10 · 03_Company_Proof/02_icp-positioning-ramp.md

Before

the "neutral router" and "existing spend-data advantage" claims pass the same check today because no competitor evidence gathered here claims either.

After

the "neutral router" and "existing spend-data advantage" claims pass the same check today because no competitor evidence gathered here claims either (caveat added by the 2026-09-06 gate: OpenRouter also sells no models, so "neutral router" differentiates against Cursor- and Meta-style routers, not against OpenRouter; the spend-data claim remains unmatched).

R-12 · REVISE · finding F12 · 03_Company_Proof/03_launch-brief-workback-ramp.md

Before

| Competitive timing | Confirmed favorable — no evidence a competitor (OpenRouter, Portkey, Martian) shipped a comparable capability in the same window that would have forced a defensive re-tier | Publicly verified, Section 1 |

After

| Competitive timing | Favorable on absence of evidence — no evidence found that a competitor (OpenRouter, Portkey, Martian) shipped a comparable capability in the same window; absence of evidence, not evidence of absence (relabeled 2026-09-06) | Reasoned hypothesis, Section 1 |

R-15 · REVISE · finding F15 · 02_Company_Selection/scorecard.md

Before

— the S-1 is confidential and no listing date exists as of this research.

After

— the confidential S-1 filing itself is asserted by a single aggregator and is not corroborated (see criterion 1, downgraded 2026-09-06), and no listing date exists as of this research; the criterion score was not re-run after the downgrade.

R-18 · REVISE · finding F18 · 03_Company_Proof/02_icp-positioning-ramp.md

Before

with Enterprise a distinctly smaller share. [Publicly verified — via WebSearch snippet, g2.com, accessed 2026-09-04]

After

with Enterprise a distinctly smaller share. [Publicly verified — via WebSearch snippet, g2.com, accessed 2026-09-04; not independently re-fetched by the 2026-09-06 gate (G2 blocks direct fetches) — single-snippet, load-bearing for A2, flagged]

R-19a · VERIFY · finding F19 · 03_Company_Proof/04_kpi-tree-measurement-ramp.md

Before

[PRNewswire, 2026-08-20]

After

[PRNewswire, 2026-08-19]

R-19b · VERIFY · finding F19 · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

flag as Reasoned hypothesis only if used that way | 2026-09-04 |

After

flag as Reasoned hypothesis only if used that way | 2026-09-06 |

R-19c · VERIFY · finding F19 · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

Test event: Ramp's AI Router / "Router" launch, 2026-08-20,

After

Test event: Ramp's AI Router / "Router" launch, 2026-08-19/20 (release dated 2026-08-19; press 2026-08-20),

R-21a · REVISE · finding F21 · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

press reports range from over $7B (Bloomberg) to $8B (Semafor)

After

press reports range from over $7B (Bloomberg) to more than $8B (Semafor, Axios)

R-21b · REVISE · finding F21 · 03_Company_Proof/02_icp-positioning-ramp.md

Before

to $8 billion (Semafor, 2026-08-21)

After

to more than $8 billion (Semafor, 2026-08-21; Axios)

R-21c · REVISE · finding F21 · 03_Company_Proof/02_icp-positioning-ramp.md

Before

reported between over $7B and $8B

After

reported between over $7B and more than $8B

R-22 · REVISE · finding F22 · 03_Company_Proof/01_product-truth-brief-ramp.md

Before

one social post at launch described a waitlist and is outweighed by three primary sources

After

one social post at launch (a technology-news account; not retained by the 2026-09-04 pass and not re-located on 2026-09-06) described a waitlist and is outweighed by three primary sources

R-23 · REVISE · finding F23 · 03_Company_Proof/02_icp-positioning-ramp.md

Before

could not be corroborated in G2 or Capterra review text on re-check;

After

could not be corroborated on re-check (2026-09-06, search-snippet pass over the G2 and Capterra Ramp review pages, which block direct fetch — a negative search, not a read of every review);

5. Where Jerry's judgment sits, and where the system's capability sits

This is stated so the case study cannot be read as "the AI did it" or "the person did it."

Operator note, in plain words: this run was executed by an AI operator session under Jerry's operating system and rulings. The judgment that shapes the system, and the decisions at every gate, are his. The leverage under it is AI.

6. Independent verification (read-only adversarial pass)

Who: Mike's read-only verifier (D-003), an agent with no write tools, briefed to refute rather than confirm, dispatched after §1–§5 and the first-pass edits were on disk. It read all six files and fetched seven primary sources. Its verdict, verbatim:

MIXED — the 14 edits are all really present and the ten I could reach a primary source for are substantively correct, but six blocking defects remain (a ruling ID that resolves to the wrong decision, an unsupported "opt-out" privacy claim about Ramp shipped as Publicly verified, three artifacts whose stale "we couldn't fetch anything" notes contradict their own new direct-fetch citations, a survived $153M figure orphaned from a removed claim, a "closed beta" assertion the gate itself refuted, and a Product Truth Brief whose six downstream sections still say the two resolved rows never cleared), plus a thesis tournament that reads cleaner and more decisive than the document it claims to record.

Twenty-three findings; every one answered. Blocking (6): F1 wrong ruling ID (D-022 → D-024, nine occurrences) → R-01. F2 "opt-out" unsupported by TechCrunch or router.com → R-02a–c, label downgraded. F3 Product Truth Brief downstream sections still said rows 1 and 3 never cleared → R-03a–i, dated notes. F4 orphaned "~$153M raised" → R-04. F5 "closed beta" inference the gate had refuted → R-05 removed. F6 stale egress notes in three artifacts → R-06a–e superseded notes. Should-fix (13): F7 roster "agree" overstated; SpaceXAI wording → R-07. F8 40/30/99.9 cited to TechCrunch, which doesn't carry them; primary is Ramp's release → R-08a–c. F9 tournament cleaner than the record → §3 rewritten with the document's sentences. F10 T2 caveat contradicted the document → R-10. F11 "four labels" vs. the charter's eight → header corrected. F12 "Confirmed favorable" on a negative search → R-12. F13 unattributed opening sentence → "per Ramp". F14 §8 stripped of its Illustrative label → corrected. F15 Deel sanity-check bullet and un-rerun score → R-15 + noted. F16 "cloud environment" not in the record → header corrected. F17 five material claims missing from §2 → rows 39–43. F18 gate asymmetry on the G2 57/38 snippet → R-18 flagged. F19 three date residues → R-19a–c. Notes (4): F20 "by a hundred models" false precision → corrected. F21 Semafor says more than $8B → R-21a–c. F22 the waitlist social post is unidentified → R-22 says so. F23 the V-10 negative check is unreproducible → R-23 names what was checked.

One flattering clause it called: "Jerry decided which company to run" paired with "the system produced the four scores" omitted that the system also recommended Ramp and the Founder approved the recommendation. Corrected in §5.

Four-persona read of the public page (second reviewer, same night): a VP of Marketing, a technical recruiter, a senior PMM peer, and an AI evaluator. Consensus findings: the outside-review section was empty when they read it (this section and the page's station 08 now carry the findings); the headline read as a Ramp employee's postmortem for the first fifteen seconds; no role title above the fold; "Nine rows, four refused" followed by four rows that were mostly safe; two nav items pointing at the same place. All fixed on the page. Their honest limit, kept: the page evidences process, not execution under pressure or outcomes; the "first two weeks" test passed two sentences of three (what he'd do alone) and failed the third (what he'd do with people). That gap is real and is not this page's to close; it is what the six rooms and the Modus record are for.

Second independent verification (Fable 2, claude-worker-fable-ac2f66, 2026-09-06 21:10 CT; Opus verifiers, Gates A/B/D): Verified FAIL, 26 applicable, 17 passed, 7 failed, 2 human judgment. Verbatim P0: "neutral-routing thesis false about Ramp in four public places — run/index.html station 05, the ledger's T2 row, /run/artifacts/icp-positioning.html, rooms.js + ROOM_MECHANISMS Positioning example. Refuted by Ramp's blog post https://ramp.com/blog/router-launch (2026-08-19, Wayback 08-24): 'We don't build models, so we have no reason to favor any particular one.' Also fires the page's own station-10 update trigger: Ramp itself already does." Answered: §3 rewritten (third version); station 05, the Positioning room example, ROOM_MECHANISMS and the artifacts regenerated from the corrected document (peer commit de6ed53 reframed §B2/§B3/§B4 and snapshotted the primary sources). Other items from that pass and their answers: dead lineage link (fixed); R-23's softened V-10 wording not propagated to the page, the room example and ROOM_MECHANISMS (propagated); "actually" in room copy (removed); "(prototype)" title and stale meta on the homepage (replaced); #work anchors without a target (retargeted to #functions); static room pages carrying the homepage's headings without JS (shell headings demoted, homepage hidden until the app boots); the adoption proof citing installs as if earned during the tenure (now a plain size fact per canon/company/CAREER_HISTORY.md: the install base pre-dates the tenure; Cloud Fortified was earned during it); ROOM_MECHANISMS "staged, not wired" (updated); "fourteen edits" (49); /run/ missing from the sitemap (generator now includes it); the R-series skipping R-03g (it was a no-op guard, dropped before logging). Human-judgment item routed to the Founder on the ticket: the public page criticized a company with an open application; its row is anonymized on the page by default.

7. What shipped

• The four artifacts above, corrected by the gate, in this repo.
• This ledger, including the second-pass edits and the reviewer's findings.
• The public case-study page at /run/ on the review site, built from this ledger in the site's room grammar, with the artifacts openable.
• One example per room, wired into each operating view as the first disclosure ("One example, from one run") with a link into the station of this run where that function shows up (run/ROOM_MECHANISMS.md is the source; rooms.js carries the copy; build-routes.js regenerates the static room pages).
• On the homepage: the Jerry OS line, once, in the after-section, with the link to this run; the nav's "Approach" now resolves to /run/.

8. What we'd measure, and what would change the system's mind

If this were a live launch (Illustrative output — no Ramp data was accessed; this is the tree the OS would build at kickoff). Top metric per the KPI tree: Qualified Routed Volume, weekly, from sustained-use accounts, deliberately excluding one-time signups chasing the $26 credit. Two measurement eras, forced by the free-through-2026 fact. Sales enters only at two named signals; most signals have no sales owner by design.

What would change the system's mind (update triggers already written into the artifacts): - Router pricing or enterprise terms are published → re-run the buying-committee map (the economic buyer resolves). - A Router deal is publicly case-studied as sales-assisted → T3 moves from conditional to live. - A competitor other than OpenRouter puts neutrality in its own copy → the selected line loses what durability it has and is re-tested (Ramp's own copy already carries it; OpenRouter shares the attribute without the copy). - Ramp publishes cohort, window or denominator for the 40% → row 5 of the Product Truth Ledger moves from "attribute to Ramp" toward "safe". - An independent audit of the savings figure appears → the same row becomes citable as fact.

[op:claude-worker-cd-sweep | Engagements/JOS-26-one-line/run | 2026-09-06 | authority: JOS-29 Current Contract; D-060; D-066]