Changelog
Every change to the revenue data behind the multiples — dated and sourced. Append-only. How the data is built →
Breadth, concentration, regime and valuation given visual form
Network Pulse shows breadth (bought vs sold) and flow concentration (top 5 vs rest) as proportion bars instead of bare counts, and drops the most-bought/most-sold cells the map now covers. Atlas Regime shows three discrete states rather than a continuous gauge, which is what the model actually produces, plus a history strip built only from stored snapshots. Valuation Pulse gained a log-scale multiple spectrum and lost the cheapest/highest cells it replaced. 'Largest 24h flows' compressed to Flow Extremes. The briefing heading now distinguishes a live network read from the last completed snapshot rather than implying currency it lacks.
Atlas Map — the subnet economy as one picture
Today in TAO now leads into a scatter map of every tradeable subnet: horizontal position is 24h net flow as a share of the subnet's own pool on a signed log scale, vertical position is market cap on a log scale, bubble area is pool size, and a gold ring marks subnets that disclose external revenue. Hand-rolled SVG rather than a charting library, so nothing extra ships to the homepage. Netuid 0 (Root) and subnets with no trades are excluded rather than plotted at the centre, and pools under 500 TAO are left off entirely — normalising a flow against a near-empty pool is how the site once printed '+1,688,349.3%'. One tab stop with arrow-key stepping, because an SVG element here takes focus without firing a focus event, so 113 tabbable circles would have been 113 stops that showed nothing.
Subnet scorecards, revenue coverage, and Leaderboard folded into Subnets
Subnet pages lead with a factual scorecard — valuation, revenue, revenue freshness, capital, momentum, liquidity — rather than a composite score, because weighing those against each other is the reader's judgement, not ours. Valuation Pulse gained Revenue Coverage: 5 of 128 subnets disclose external revenue, representing 12.1% of subnet market cap, tracked daily so it becomes a trend. What Changed now attributes valuation moves to CAP or CAP + REVENUE and reports capital-flow reversals and revenue-backed cohort behaviour. /leaderboard redirects to /subnets, which absorbed its momentum, capital-flow and emerging rankings as sorts and filters.
Revenue freshness is now shown wherever revenue is
Every revenue figure carries its age, measured from the end of the period it covers rather than its publication date. States are Current (<30d), Aging (30-90d) and Stale (90+d). All five figures currently on the ledger are 154-184 days old and classify as Stale — not invalid, but the multiples they sit under are comparing a live market cap against an older disclosure, and that is now stated plainly on the ledger, on subnet pages and in the Valuation Pulse.
Revenue provenance recorded separately from period
The ledger now distinguishes the period revenue was earned in from the date the figure was published, and records the source behind each figure. Publication dates were never captured for the founding entries and are shown as 'not recorded' rather than inferred from the period end. Snapshots from 2026-10-01 onward also freeze each subnet's revenue figure, confidence and period, so a historical briefing reflects what was known that day rather than a figure revised later.
Today in TAO rebuilt as a daily briefing
The daily read now leads with Atlas Regime — a deterministic RISK-ON / SELECTIVE / DEFENSIVE call scored from five measured signals, each expandable so the call can be audited. Sections were reordered to interpretation, changes, signals, metrics, then raw data. 'What to lean into / what to avoid' is gone, replaced by signal-oriented statuses (Confirming, Emerging, Reversing, Weakening, Unusual) that describe situations rather than recommend positions. A Valuation Pulse module ties the daily read back to the Subnet Multiple.
Relative-flow percentages replaced with bands
Subnet cards could show a relative flow of +1,688,349.3%. Two faults combined: the numerator used Taostats' unusable net_flow_1_day field, and the denominator floored pool size at 1 TAO, so a small flow through a near-empty pool produced an enormous ratio. Flow scale is now classified into bands (moderate / strong / exceptional, in or out) and pools below 500 TAO are excluded from the comparison entirely. The raw ratio is retained internally for scoring and tooltips.
Daily network snapshots now stored server-side
SubnetAtlas captures one network snapshot per day and keeps the raw measurements. 'What changed' now compares two of our own snapshots instead of a baseline held in the reader's browser, so every visitor sees the same changes. Archived briefings are published at /daily/<date> and are rebuilt from the numbers stored for that date rather than recalculated with current data. History starts today; no past days were back-filled.
SubnetAtlas is now one product: the Subnet Multiple
The Multiple is the homepage. The site answers one question — which Bittensor subnets are businesses and which are priced on belief — and the navigation was cut back to the standard, the directory, a portfolio read, this changelog, and the methodology. Momentum and tape surfaces (signals, leaderboard, screener, activity feed) were unlinked from the primary navigation; their routes still resolve. No revenue figures changed.
24h flow figures corrected — previous values were not TAO
Flow was rendered from Taostats' net_flow_1_day field divided by 1e9 and labelled TAO. That was wrong by roughly five orders of magnitude: it implied a single subnet moving 259M TAO in a day against a pool holding 208K TAO, and a network total of -2.2B TAO against ~10M TAO in circulation. The field's sign agrees with actual buy-minus-sell volume only about half the time, and its 7-day variant is ~7x the 1-day variant for every subnet — the signature of a running sum, not a net flow. Flow is now computed as 24h buy volume minus sell volume in TAO, bounded by both pool size and gross volume. The 1h, 7d and 30d windows have no trustworthy source and are no longer published. No multiple was affected — flow has never been an input to the Multiple.
Bitcast (SN93) disclosed TAO Institute-verified revenue
$1.9M/yr. Basis: Apr 2026 × 12 (TI verified). Creator incentives · monthly disclosure.
Source ↗Chutes (SN64) disclosed TAO Institute-verified revenue
$5.6M/yr. Basis: 90d × 4 (TI verified). Inference API · on-chain buybacks corroborate.
Source ↗Gradients (SN56) — no verified revenue
Training market · monetization pending.
Source ↗Lium (SN51) disclosed self-reported revenue
$1.6M/yr. Basis: Jan–Apr × 3. GPU rental · self-reported.
Source ↗Targon (SN4) disclosed self-reported revenue
$1.2M/yr. Basis: Apr 2026 × 12. TEE compute · Series A backed.
Source ↗Templar (SN3) — no verified revenue
Covenant AI exited Apr 2026.
Source ↗Nineteen (SN19) — no verified revenue
High-freq inference · no disclosure.
Source ↗Celium (SN12) — no verified revenue
GPU market · reports revenue, none verified.
Source ↗Vanta (SN8) disclosed self-reported revenue
$1.2M/yr. Basis: MRR Mar 2026 × 12. Self-reported MRR · unaudited.
Source ↗sundae_bar (SN121) announced a commercial deal
Commercial deal announced Mar 2026. No revenue figure disclosed.
Source ↗