Immutable opportunity record
The recommendation is allowed to move. Every release preserves the candidates, evidence threshold and sprint contract we were willing to stand behind at that moment.
A changed lead is not a broken record. It is the expected result of stronger evidence. The failure would be silently rewriting the earlier claim—or dressing weak signal as continuity.
Return to the current Opportunities page →Newest first. Every card opens the preserved page exactly as that release was framed.
This table grows with every release, so the final sprint review can compare the starting hypothesis with the recommendation without reconstructing either.
| Compare | v0.1.0 | v0.2.0 | v0.2.1 | v0.2.2 | v0.3.0 | v0.3.1 | v0.3.2 | v0.3.3 | v0.3.4 | v0.3.5 | v0.3.6 | v0.4.0 | v0.4.1 | v0.5.0 | v0.5.1 | v0.5.2 | v0.5.3 | v0.5.4 | v0.5.5 | v0.5.6 | v0.6.0 | v0.6.1 | v0.6.2 | v0.6.3 | v0.7.0 | v0.7.1 | v0.8.0 | v0.8.1 | v0.9.0 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Verdict | Paid-Test Conversion is the lead; Payment Recovery is the internal challenger; Residential Vantage is the external probe. | Paid-Test Conversion leads validation; Payment Recovery is conditional; Residential Vantage remains discovery only. None is launch-ready. | Paid-Test Conversion leads validation; Payment Recovery is conditional; Residential Vantage remains discovery only. None is launch-ready. | Paid-Test Conversion leads validation; Payment Recovery is conditional; Residential Vantage remains discovery only. None is launch-ready. | Payment Recovery leads the inner lane on measured evidence with its flip-down condition published; Retention economics (pricing x churn) enters at #2; Paid-Test Conversion parks with a drop-by date; Residential Vantage leads the outer lane with the crypto-native wedge second by evidence, first by test order. | Payment Recovery leads the inner lane on measured evidence with its flip-down condition published; Retention economics (pricing x churn) enters at #2; Paid-Test Conversion parks with a drop-by date; Residential Vantage leads the outer lane with the crypto-native wedge second by evidence, first by test order. | Payment Recovery leads the inner lane on measured evidence with its flip-down condition published; Retention economics (pricing x churn) enters at #2; Paid-Test Conversion parks with a drop-by date; Residential Vantage leads the outer lane with the crypto-native wedge second by evidence, first by test order. | Payment Recovery leads the inner lane on measured evidence with its flip-down condition published; Retention economics (pricing x churn) enters at #2; Paid-Test Conversion parks with a drop-by date; Residential Vantage leads the outer lane with the crypto-native wedge second by evidence, first by test order. | Payment Recovery leads the inner lane on measured evidence with its flip-down condition published; Retention economics (pricing x churn) enters at #2; Paid-Test Conversion parks with a drop-by date; Residential Vantage leads the outer lane with the crypto-native wedge second by evidence, first by test order. | Payment Recovery leads the inner lane on measured evidence with its flip-down condition published; Retention economics (pricing x churn) enters at #2; Paid-Test Conversion parks with a drop-by date; Residential Vantage leads the outer lane with the crypto-native wedge second by evidence, first by test order. | Unchanged ranking, second view of the same evidence. The board now plots its own claims: every load-bearing claim placed at (evidence level, grade). Levels L3 and L4 hold nothing — no claim is reconciled against an external ground truth and none is owner-validated. | Payment Recovery leads the inner lane on measured evidence with its flip-down condition published. Retention economics (pricing x churn) enters at #2 on measured warehouse evidence. Paid-Test Conversion parks on MN-side data with a drop-by date. No candidate is yet a proven opportunity. | Payment Recovery leads the inner lane on measured evidence with its flip-down condition published. Retention economics (pricing x churn) enters at #2 on measured warehouse evidence. Paid-Test Conversion parks on MN-side data with a drop-by date. No candidate is yet a proven opportunity. | Payment Recovery leads the inner lane on measured evidence; its open gate is now measured. Retention economics at #2. Paid-Test parked. Vantage leads the outer lane. No candidate is yet a proven opportunity. | Payment Recovery leads the inner lane on measured evidence; its ceiling is now bounded to Stripe. Retention economics at #2. Paid-Test parked. Vantage leads the outer lane. No candidate is yet a proven opportunity. | Payment Recovery leads the inner lane on measured evidence; its ceiling is bounded to Stripe. Retention economics at #2. Paid-Test parked. Vantage leads the outer lane. No candidate is yet a proven opportunity. | Payment Recovery leads the inner lane on measured evidence; its ceiling is bounded to Stripe. Retention economics at #2. Paid-Test parked. Vantage leads the outer lane. No candidate is yet a proven opportunity. | Payment Recovery leads the inner lane on measured evidence; its ceiling is bounded to Stripe. Retention economics at #2. Paid-Test parked. Vantage leads the outer lane. No candidate is yet a proven opportunity. | Payment Recovery leads the inner lane on measured evidence; its ceiling is bounded to Stripe. Retention economics at #2. Paid-Test parked. Vantage leads the outer lane. No candidate is yet a proven opportunity. | Payment Recovery leads the inner lane on measured evidence; its ceiling is bounded to Stripe. Retention economics at #2. Paid-Test parked. Vantage leads the outer lane. No candidate is yet a proven opportunity. | Payment Recovery leads the inner lane on measured evidence; its ceiling is bounded to Stripe. Retention economics at #2, now with a measured tier cut. Paid-Test parked. Vantage leads the outer lane. No candidate is yet a proven opportunity. | Correction release — four board strings restated after the adjudicated Tris red team of the MN DB audit (ACCEPT WITH MATERIAL CORRECTIONS, 2026-08-17). No ranking, status or decision-rule change. | Wording repair — one phrase in W7 restored to the adjudicated sentence after the opposing verification of the v0.6.1 integration found it paraphrased. No ranking, status or decision-rule change. | Payment Recovery leads the inner lane on measured evidence with its flip-down condition published. Retention economics (pricing x churn) enters at #2 on measured warehouse evidence. Paid-Test Conversion parks on MN-side data with a drop-by date. Residential Vantage leads the outer lane; the crypto-native access wedge runs the first smoke test before Sept 15. No candidate is yet a proven opportunity. | Payment Recovery remains #1; Retention Economics remains #2; GoProxies Conversion System replaces the narrower Paid-Test label at #3 without promotion; outer order holds; no torpedo. | #1 and #2 hold. #3 gets wider, not stronger. | Payment Recovery remains #1; Retention strengthens; GoProxies and Residential hold; Crypto-native is the most vulnerable #5. Delight stays inside #2 and Key Maker enters as a challenger/enabler. No torpedo. | No formal rank swap. Key Maker is visibly named as the contender to #5; Crypto-native Access holds pending the normalized portfolio comparison. | Five credible mechanisms for next decision attention; no approved bets and no material pursuit authority. |
| Four-week contract | Seriously test three to five candidates across internal and external lanes; recommend only those that clear the evidence threshold. | Screen three to five, deepen one or two, prepare one complete torpedo contract, and launch at most one only through the six-part gate. | Screen three to five, deepen one or two, prepare one complete torpedo contract, and launch at most one only through the six-part gate. | Screen three to five, deepen one or two, prepare one complete torpedo contract, and launch at most one only through the six-part gate. | Two evidenced internal candidates, one external discovery candidate, one dated external smoke test, a standing generative ring, and a live rejection log; methodology and the 12-row blindspot map published with the board. | Two evidenced internal candidates, one external discovery candidate, one dated external smoke test, a standing generative ring, and a live rejection log; methodology and the 12-row blindspot map published with the board. | Two evidenced internal candidates, one external discovery candidate, one dated external smoke test, a standing generative ring, and a live rejection log; methodology and the 12-row blindspot map published with the board. | Two evidenced internal candidates, one external discovery candidate, one dated external smoke test, a standing generative ring, and a live rejection log; methodology and the 12-row blindspot map published with the board. | Two evidenced internal candidates, one external discovery candidate, one dated external smoke test, a standing generative ring, and a live rejection log; methodology and the 12-row blindspot map published with the board. | Two evidenced internal candidates, one external discovery candidate, one dated external smoke test, a standing generative ring, and a live rejection log; methodology and the 12-row blindspot map published with the board. | Seriously test three to five candidates across internal and external lanes; recommend only those that clear the evidence threshold. | Two evidenced internal candidates, one external discovery candidate, one dated external smoke test, a standing generative ring, a live rejection log, and the blindspot map as a working tool: ten quests in three lanes plus one watched and one standing row, statuses derived from an event record. | Two evidenced internal candidates, one external discovery candidate, one dated external smoke test, a standing generative ring, a live rejection log, and the blindspot map as a working tool: ten quests in three lanes plus one watched and one standing row, statuses derived from an event record. | Payment lane evidence + blindspot map events | Blindspot map event (W1 killed) + Payment Recovery ceiling clause | Blindspot map events (W2 signal + kill) + data-unlock B2 sharpened | Blindspot map events (W3 claim + signal) | Blindspot map events (W4 claim + signal) | Blindspot map events (W5 claim + signal + kill) | Blindspot map events (H1 claim; C1 claim + signal) | Blindspot map rows W6–W8 (added, claimed, signalled); W3 reworded; retention #02 sixth graded claim | W2 marketing signal (F-T2) · W7 why + signal (F-T5, F-T6) · W8 why (F-T7); ranking, candidates, rules, statuses unchanged | W7 why (F-T5 named query); nothing else | Same four-week contract as v0.6.2: evidence-first ranking with pre-published flip conditions; payment recovery leads on measured Stripe evidence; retention (pricing × churn) at #2 on measured warehouse cohorts; paid-test parked with drop-by date; blindspot map running its 30-day mandate. | Same evidence-first MN top five, plus a separate Kairos capability slate. No material pursuit, build, outreach programme, transfer or spend is authorized. | Additional warehouse and operating evidence through 20 August 2026; MN commercial ranking held, with a separate Kairos leverage slate. | Same evidence-first top five and September contract. No material pursuit, build, outreach programme, transfer or spend is authorized. | Clarity correction only: contender relationship, evidence comparison, promotion path and park conditions; no ranking or evidence change. | Three atomic evidence checks before 11 September; rerank on results or preserve the provisional order. |
| Change from previous | Initial public hypothesis board. It ranked three survivors but blurred candidate screening with serious real-world testing inside the four-week window. | Corrected the sprint contract: screening is not testing, a real torpedo is conditional, and a no-launch result can still be decision-grade. | Made the opportunity board an immutable evolving series. Added version links, an append-only changelog, a hash-chained manifest, and a side-by-side sprint timeline without changing the candidate ranking. | Added accessible hover and keyboard-focus explanations to all three evidence-map candidates, with a reserved explanation band and a deploy-time contract that refuses tooltip drift. | Re-ranked with the evidence-first Area Decision Protocol after a blind alien-eyes derivation adjudicated against the sealed prior board; Payment Recovery leads on measured evidence; Retention economics enters on measured warehouse evidence; Paid-Test parks on MN-side data with a drop-by date; a generative ring and blindspot map are added; v0.2.3 repair items absorbed. | Evidence-ladder tier chips gain hover/focus explanations mapping each tier (L0-L4) to where it appears on this board. Board content and ranking unchanged from v0.3.0. | Clarity pass, no ranking or evidence change: every candidate gains a visible plain-words explanation with a concrete example and a plain-language subtitle; the title's 'torpedo' is defined on the page; the signal map gains a how-to-read line; twenty-one working terms gain hover explanations. Standard set by Lee: clear, not clever. | Chip-gloss pass, no ranking or evidence change: all 81 repeated chrome labels become learnable on hover or keyboard focus - every candidate score chip (15) explains its evidence grade and what it refers to on that candidate, plus authority chips, blindspot criteria chips, rejection status tags, rank status lines, map-key markers and hero meta; a reject-list styling defect that shrank jargon terms inside card headings is corrected; the release verifier gains a chip-gloss guard proven red-first. | Theme seam unification, no ranking or evidence change: the board now rides the fleet light/dark module (fleet-theme.js - lm-theme cookie remembered across leematulis.com surfaces) instead of a page-private theme key with no toggle; a floating light/dark control appears on every page; the war-room-wide check-theme guard, proven red-first, makes a page that hand-rolls its own theme scaffold un-deployable. | Register re-cut, no ranking or evidence change: the 'In plain words:' prose prefix is cut as bloat (Lee, 2026-08-13) - the Provenance card now opens directly on its content; labeled plain-language translation blocks are unchanged. | Adds the evidence grid to section 02: all 24 load-bearing claims plotted on the signal map's up-and-right grammar, with four region labels (well-argued guess / decision-grade / rumour / sloppy measurement), the ghosts-troops divider, and an off-field column for claims carrying no evidence at all. Derived from the claim lists by build_claim_grid.py and guarded in release.py verify, so a changed grade cannot leave a stale dot. Dot numbering now follows the signal map rather than per-ring data-rank. No evidence, ranking or verdict changed. | The blindspot map becomes a tool, not a list: rows now carry a lane and pull order, a plain subtitle, a prior, an unlock line, an owner column, a decision rule written before the data, a predicted days-to-signal with calibration against the actual, and a drop-by date; status is derived from an append-only event record through a state machine (open, scouting, signal, climbed, killed, parked) and nothing between the markers is hand-typed; the map carries its own kill test (unused by day 21, the page says so, computed live); ordering changes: metric lineage and the owner map lead the human lane as the L4 gate. Board ranking, candidates and evidence grades unchanged. | Consistency correction, no content change: the hero evidence-state date returns to 13 Aug 2026 to match the machine brief — the evidence ladder did not change on v0.4.0, only the blindspot map's structure did; the release date (16 Aug) stays in the footer. | The payment lane's open gate is measured: the pre-registered matched cancellation test ran on 16 Aug against MN's own churn dates — of July's still-unpaid Stripe renewals ($19,150 face) only 2.8% belong to customers who had already cancelled; about 97% is involuntary; both stop rules cleared; the published flip condition (rescuable value below ~$2k/mo) is not triggered. Two claims added at invoice grain: the pool is already retried about six times per failing invoice with 17% recovered (retry quality, not count, is the lever), and fees do not kill it (Stripe nets ~95%; Apple fees recorded only from Dec 2025; fee table frozen 28 Apr). Header chip, thesis, plain words, ceiling card and next-test box updated; ranking unchanged. Blindspot map: W1 Non-Stripe payment funnels claimed and signalled — Adyen renewal charges fail 87–89% at charge grain (2025-07 to 2026-03) but Adyen is being wound down; the stores carry no retry visibility; the map's own kill test now reads passed (first claim on day 7). Evidence-state date moves to 16 Aug 2026. | Blindspot map: W1 Non-Stripe payment funnels is killed — the first quest to clear. Both pre-written clauses fired: the proceed clause on Adyen (renewal charges fail 87–89% at charge grain, 2025-07 to 2026-03) and the drop clause on the stores. The drop clause carries the base: Apple and Google — about 45% of July gross and $38k of the month's $87k successful renewals — bill and retry inside the stores, so no charge, retry or recovery outcome reaches the warehouse, and the store lifecycle export has been frozen since September 2024. Adyen cannot carry the ceiling: it is being wound down (4,404 renewal charges in Dec 2025 to 170 in Jul 2026; feed stopped 16 Jul 2026). PayPal and Coingate (about $11k of July renewals) have no charge-grain table — a data ask, not a scouting task. Payment Recovery's honest-ceiling card now says the ceiling does not rise from the warehouse. Ranking, evidence grades and evidence-state date unchanged. | Blindspot map: W2 Marketing economics is signalled and then killed — the second quest to clear, both on day 7. Signal (the first L2 number, hours after claim against a 21-day prediction): attribution in the canonical revenue table collapsed after 28 April 2026 — New revenue was 39–53% channel-attributed through March, 14% in April, 0.0% from May to August; July clean revenue is 6.8% attributed overall. The daily web-session model still tags 22% of July new web revenue to a paid click (Google, Impact, Microsoft, Meta, CJ) and 78% to no paid click; GA4 purchase events read 99% direct or null. Kill: the drop clause fired — spend data is absent from all 24 datasets, verified live over the project's whole INFORMATION_SCHEMA (503 tables; the only cost columns are proxy bandwidth COGS and an Intercom company attribute), so CAC cannot be computed from the warehouse at any grain. Spend becomes an MN-side ask: data-unlock B2 now says one monthly spend line per channel is enough. The attribution collapse is a governance finding for the data owner, not a warehouse gap. Ranking, evidence grades and evidence-state date unchanged. | Blindspot map: W3 Pricing / packaging elasticity is claimed and signalled the same evening — the first L2 number arrived from saved cohort files hours after claim, against a 21-day prediction. Neither pre-written clause fires. Tiered plans (basic/plus/pro) exist only since January 2026, so the readable cohorts are monthly, January to May 2026, censored: five cells carry 200 or more renewals — tier-1 basic 30.3% vs plus 34.3%, tier-2 plus 21.7%, tier-3 basic 12.7% vs plus 16.1%. The largest gap between price tiers is 4.0 points, under the 5-point bar, and the dearer tier renews better; pro is thin everywhere and lowest. Geography moves first renewal by about 18 points, price tier by about 4. Yearly and two-year tiered plans have no renewal window before January 2027. The comparison is cross-sectional and self-selected: a price change, not a tier comparison, is what would measure elasticity. The row stays at signal under its own rules (park is written as never; drop needs thin cells across the board, and they are not). Ranking, evidence grades and evidence-state date unchanged. | Blindspot map: W4 Support economics is claimed and signalled the same evening — the fourth warehouse row in a row to return its first L2 number in hours against a 21-day prediction. The proceed clause is met. Over the twelve complete months August 2025 to July 2026 the VPN support inbox logged 38,397 conversations; 58% carry a human tag (one per conversation) and since January 2026 an automated title layer lifts classified coverage to about 70%. The tags separate billing cleanly: billing-related contacts (cancellation 19.9% plus payment and refund 7.7%) are 27.5% of all conversations and 47.5% of tagged ones, above the 15% floor even if every untagged conversation were non-billing, so the drop clause cannot fire and coverage does not park the row. Billing-related conversations also draw 39% of agent replies (cancellation 27%, payment 12%), technical 26%; untagged conversations are short (2.7 agent replies, 14 hours median resolution against about 230 hours for tagged). Load runs 300 to 370 conversations per 1,000 active subscriptions a month, of which billing-related 75 to 100 per 1,000 and rising since August 2025. Three inbox seats carry about 3,200 VPN plus 460 node and 70 proxy conversations a month. Cost per contact needs a salary line MN holds, not the warehouse; that goes on the data-unlock list. Ranking, evidence grades and evidence-state date unchanged. | Blindspot map: W5 Fraud / abuse is claimed, signalled and killed the same evening — the fifth warehouse row to return its first number in hours. The drop clause fires and the row closes as a win: since June 2025 the Stripe door shows 1 dispute on 34,934 successful charges, 13 dollars of 563,541, 0.002% of gross, far under the 0.5% floor. It was not always so — November 2024 to January 2025 was a real wave, 385 disputes on about 6,500 successes (US 5.1%, Japan 35.8%; 3DS-covered charges 0.36% against 6.3% without). Refunds, not chargebacks, are the leak now: 5.8% of Stripe dollars over the last twelve months. Card testing is small (bursts of five or more failed attempts per card per day are 4% of failed charges), but 21% of failed Stripe charges sit on cards that never succeeded, so a payment-failure denominator built on all failures overstates dunning by about a fifth — that note passes to Payment Recovery, whose denominator has to be renewal failures on known cards. Two conditions travel with the win: the zero is read from a flag whose sibling refund flag keeps updating after capture and which carried the 2024 wave, so the instrument is live, but one Stripe-dashboard dispute count for 2026 H1 from MN confirms it outright and is asked for; and the Stripe replica has holes — no charges at all for February to October 2024 and February to April 2025. Ranking, evidence grades and evidence-state date unchanged. Five of ten quests are now off the open pile: W1, W2, W5 killed; W3, W4 signalled. | Blindspot map: the human and corpus lanes open. H1 Metric lineage and owner semantics is claimed — the sit-down question list for the owners of the two measured candidates' fields is drafted and handed over, so the L4 gate can start moving before it becomes the critical path (drop-by 28 Aug). C1 Finance perimeter is claimed and signalled the same evening from the public company registry, the first denominator on this board that is citable as it stands: the two Lithuanian operating entities together booked 3.10M euro of sales in 2025 and lost 298k (about 25k a month); the older entity lost 672k in 2024, the year the all-hands were chasing +53k of growth a month towards break-even; the ring employs 25 people (July 2026) on roughly 113k a month of gross payroll, and carries negative equity of 784k on 1.8M of liabilities — funded, not self-sustaining, as the 2023 all-hands said in words. Runway is not in the free filings and the row does not climb: cash and committed funding become one narrow finance ask. Six of ten rows have now returned a first number in hours; the map's predictions remain unshortened because the remaining rows are gated on people, not queries. | Three warehouse-lane rows join the map (W6 Android product experience in the largest paying geography · W7 failed payments as a churn label and the never-recovered remainder · W8 the Primer switch and the payment label it hides under), each claimed 16 Aug and signalled 17 Aug on numbers re-tested against an opposing review before anything reached this board: churn by country tier on a month-start cohort (tier 3 turns over about 1.5× tier 1 every month, July 38% gross / 28% net of 30-day recoveries against 25% / 19%), cancel reasons standardised within platform (tier-3 cancellers name price less and product failure more on Android, iOS and Windows — a survey direction, not a churn cause), modal list price by gateway (13.49 dollars a month for the plus plan in every tier on Stripe, Coingate, PayPal and Apple; Google Play tier 3 about 7% lower), the Primer switch reconciled week by week on identical populations (web New-type dollars at 100–102% of the spine's through the 22 July cut-over; the spine still labels Primer as Stripe), and 2026 churned subscriptions counted one row each (about 48% carry a billing-retry label). W3 is reworded: because the tiers do not differ in list price, tier churn is not elasticity evidence; the lever it points at is product. Retention (#02) adds a sixth graded claim carrying the tier cut, and its next test narrows to the per-geo × price-cohort pass on 2024 vs 2026 mix. One candidate row was considered and dropped before release: iOS first-connect — among users entitled to connect at their first try, same-day success is 96.9% on Android and 96.3% on iOS; the raw platform gap was composition. Nine of thirteen quest rows have now returned a first number; predictions stay unshortened. | W2: the 22 % paid-click line is a July new-web-session share (18.3k of 82.1k rows), untested on a refuted-grain table, not a revenue share. W7: 93 % is P(≤30 d | Recurring) = 10,328 of 11,064; recovery-vs-returner is a named hypothesis (reason join 10.1 %); the 45 % dollar share is a raw sum on a raw-sum-forbidden table, labelled pending a wo.safe rerun. W8: weekly web/spine ratio 101 · 102 · 101 · 100 · 100 · 110 % — the 110 % week was dropped by the old '100–102 %'. Source: 90-system/agent-reviews/2026-08-17 — Tris review of MN DB audit (red team, adjudicated). | W7: 'Named query: reactivation rows in 2026 by ≤30 d × type' → 'Name the query: reactivation rows in 2026 by ≤30 d × type' — the adjudicated Tris correction verbatim (review F-T5). Found by Codex's findings-only verification of v0.6.1 (2026-08-17 18:19): six of seven items LANDED, this one DRIFTED by paraphrase. | Removed the unsupported raw-derived dollar-share label from current carriers; retained the measured ≈48% share of churned subscriptions carrying a billing-retry label. | Corrects the stale zero-warehouse-presence proxy claim: account, plan, product, funnel and support surfaces exist, while revenue amounts and paid-test cohort economics remain absent. Adds K1-K3 as a separate Kairos slate. Ranking order unchanged. Footer now opens the complete v0.6.3 outgoing page. | Pre-deploy interaction correction: claim explanations become a contained mobile bottom sheet and the top-bar date is synchronized and kept clear of the fleet theme control. Content and ranking are unchanged from v0.7.0. The footer preserves both v0.7.0 and the complete v0.6.3 outgoing page. | Weekly evidence refresh: Delight adds working-tier support prevalence plus confounder attenuation inside Retention; Key Maker enters as a cross-cutting MN enabler and external generative candidate; source access is corrected to 2 ready, 3 partial and 6 blocked; acquisition optionality becomes a watch lane; the formal top five do not move; 320px horizontal containment is repaired. | Adds a visible Contenders section comparing Key Maker directly with #5, explaining why it is the named challenger, why Delight remains inside #2, and what would promote or park Key Maker. Preserves v0.8.0 intact. | Blind denominator-first re-derivation expands the inventory to 26 surfaces and 25 constructs, reranks five leads, names seven contenders and preserves 21 remainder rows; v0.8.1 remains immutable. |